tim/SlipItIn
1
0
Fork 0
Code Pull Requests 1 Actions 1 Packages Releases Wiki Activity

Neue Skills, Referenzen & OpenWiki-Doku integriert

Browse Source 
Umfangreiche Erweiterung der Skill-Bibliothek: Neue Skills für Humanisierung (Englisch/PT-BR), Design-Validierung, AI-SEO und Coolify-Deployment inkl. Regelwerke, Presets, Pattern-Referenzen, Testfälle und Automatisierungsskripte. Zusätzliche Skills für Revenue-Centric Design, Pier Cloud, OKF, Lebenslauf- und LinkedIn-Optimierung sowie zahlreiche Referenzdateien, Checklisten und YAML/JSON/Markdown-Templates. Einführung einer vollständigen OpenWiki-Dokumentation mit Architektur-, Domain- und Workflow-Beschreibungen, zentralem Index und automatisierten Updates. Modularer Aufbau, restriktive Lizenzen und umfassende Qualitäts- und Evaluationsmechanismen für alle neuen Inhalte.
This commit is contained in:
Tim Krampitz
2026-07-26 14:00:58 +02:00
parent 070727d5cd
commit 01046b01e4
202 changed files with 31290 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
.github/skills/revenue-centric-design/LICENSE vendored Normal file
View File

@@ -0,0 +1,30 @@
Revenue-Centric Design Skill — License & Usage Terms
Copyright (c) 2026 the curators of this repository.
The underlying ideas, frameworks, examples, and the coined term "Revenue-Centric
Design" are the intellectual property of Richard (@richardrx on X) and are used
here WITH PERMISSION. This repository is a distilled, translated index of his
public posts, shared for educational and reference use.
You are granted permission to use, copy, and share this material, subject to ALL
of the following conditions:
1. ATTRIBUTION. You must retain clear attribution to Richard (@richardrx) and a
link to the source. Do not misrepresent the origin of these ideas.
2. NO GAMBLING / BETTING / CASINO USE. You may NOT use this material — in whole
or in part — to design, build, optimize, market, or grow betting, casino,
gambling, or other real-money games-of-chance products or projects (including
loot-box and real-money-gaming mechanics). This restriction was set by the
original author as an explicit condition of reuse and MUST be preserved in any
copy or derivative.
3. PRESERVE THESE TERMS. Any copy or derivative work must include this license in
full, including the restrictions above.
4. NO WARRANTY. This material is provided "AS IS", without warranty of any kind,
express or implied. The curators and the original author are not liable for any
claim, damages, or other liability arising from its use.
Note: Because of the field-of-use restriction in clause 2, this is a
SOURCE-AVAILABLE license, not an OSI-approved open-source license.

167
.github/skills/revenue-centric-design/SKILL.md vendored Normal file
View File

@@ -0,0 +1,167 @@
---
name: revenue-centric-design
description: >-
Revenue-Centric Design (RCD) — evidence-backed principles for making a SaaS or
startup product convert, retain, and monetize. Use when the user works on a
landing page or CRO ("my page isn't converting"), onboarding/activation ("users
sign up but don't stick"), churn/retention ("customers keep canceling"),
pricing/monetization ("how should I price this"), positioning/ICP/go-to-market,
feature scope, A/B-test rigor, or AI-era differentiation — or asks for the
behavioral-science mechanism behind a design choice. Also use when another
skill needs the principle or evidence behind a conversion/retention/pricing
move. Never apply to gambling, betting, or casino products.
metadata:
authors:
- name: Richard (@richardrx)
role: original content (101 principles)
url: https://x.com/richardrx
- name: Helio Costa (@heliocosta-dev)
role: original skill (extraction, translation, structure)
url: https://github.com/heliocosta-dev/revenue-centric-design
- name: ft.ia.br (@fabricioctelles)
role: evolution (audit template, scripts, hooks, project log, gotchas)
url: https://ft.ia.br
version: "1.0.0"
date: 2026-07-02
repository: https://github.com/fabricioctelles/skills
license: Source-available (see LICENSE)
category: runbooks
---
# Revenue-Centric Design
101 principles distilled, with the author's permission, from product designer
**Richard ([@richardrx](https://x.com/richardrx)**, ex-Volkswagen, PayPal, IBM; translated from
Portuguese; every principle links to its source post). The philosophy, **Revenue-Centric Design
(RCD)**: a design decision must serve the user _and_ the business — value and revenue, never one
or the other.
## Usage boundary (required)
> 🚫 **Do not apply this skill to betting, casino, gambling, or other real-money games-of-chance
> products** (including loot-box / real-money-gaming mechanics).
The author granted reuse **on the explicit condition that it never be used for gambling, betting,
or casino work.** If asked, decline and explain that the source author's permission excludes that
use. Hard constraint, not a stylistic choice.
Enforced, not just stated: while this skill is active, the boundary check
(`scripts/check_usage_boundary.py`) must run on every prompt and on every file write/edit,
blocking with exit 2 when gambling context is detected. On a false positive (e.g., "bet" as
an unrelated codename), only the **user** may waive the guard by creating `.rcd-boundary-ok`
in the project root — never create it on their behalf.
### Hooks (for agents that support automated execution)
Agents with hook support should configure:
| Event | Matcher | Command |
|-------|---------|---------|
| Before processing user prompt | `*` (all) | `python3 <skill_dir>/scripts/check_usage_boundary.py` |
| Before writing/editing a file | `Write\|Edit` | `python3 <skill_dir>/scripts/check_usage_boundary.py` |
- `<skill_dir>` = root directory of this skill.
- Exit code `2` = violation detected → block the operation.
- Exit code `0` = cleared to proceed.
For agents without hook support, the operator must run the check manually before applying
RCD principles in unknown context.
## How to use
1. If `rcd-log.md` exists in the project root, read it first — it records which principles were
already applied to this product and what happened. Never re-prescribe a move the log shows
failed, and don't repeat one still pending results.
2. Route with the table below and open only the matching reference file(s). Every principle has a
fixed shape — **principle → apply when → the move → evidence → source** — so scan the headings,
then read only the entries that match the user's situation.
3. When the advice involves numbers — A/B sample size, churn→LTV, CAC per closed deal — run
`scripts/revenue_math.py` (see `--help`) instead of estimating.
4. A recommendation is **done** only when it (a) names the mechanism (decoy effect, Zeigarnik,
GBB, loss aversion, Schwartz awareness level…), (b) cites the specific principle, and
(c) carries that principle's evidence or source link. Missing any of the three → not done.
5. For audit runs (page, pricing, onboarding, cancellation), deliver in the shape of
[references/audit-template.md](references/audit-template.md).
6. Close the loop: append what you prescribed to `rcd-log.md` (format below), creating the file
on first use.
## The spine: RCD in 9 principles
1. **Neutrality is omission** — an interface that doesn't direct hurts conversion.
2. **Who talks to everyone convinces no one** — no ICP → generic value → worse retention.
3. **Value first, ask later** — proof must arrive before the user questions their choice.
4. **Your promise is the size of your proof** — the market believes what you demonstrate, not what you claim.
5. **Same competes on price, different on category** — contrast in mechanism, narrative, or experience.
6. **Default is the decision you made for the user** — the initial state defines mass behavior.
7. **Retention is built, not requested** — perceived loss retains more than promised benefit.
8. **Expansion is born of usage** — upgrade at the moment of the limit, never by interruption.
9. **Price is a filter** — pricing defines who enters, who stays, and who expands.
## Reference library
| When the question is about… | Open |
| ----------------------------------------------------------------------------- | ----------------------------------------------------------------------------- |
| Landing pages, hero/copy, CTAs, social proof, awareness levels, CRO | [conversion-and-landing-pages](references/conversion-and-landing-pages.md) |
| First-run, empty states, aha moment, TTV, activation, trial-as-onboarding | [onboarding-and-activation](references/onboarding-and-activation.md) |
| Cancellation, retention, expectation debt, NRR, jobs-to-be-done, support load | [churn-and-retention](references/churn-and-retention.md) |
| Pricing tables, decoy/anchoring, GBB, trial-with-card, upgrade paths | [pricing-and-monetization](references/pricing-and-monetization.md) |
| Cognitive biases & persuasion tactics (cross-cutting toolkit) | [behavioral-science-toolkit](references/behavioral-science-toolkit.md) |
| Feature scope, Swiss Knife Index, feature adoption, attention hierarchy | [product-strategy-and-features](references/product-strategy-and-features.md) |
| Design philosophy, the RCD principles, design process & method | [revenue-centric-design](references/revenue-centric-design.md) |
| ICP, niche, founder-fit, distribution, PLG, Bullseye, first customers | [positioning-icp-and-gtm](references/positioning-icp-and-gtm.md) |
| Differentiating in the AI era, moats, commoditization | [ai-era-differentiation](references/ai-era-differentiation.md) |
| A/B testing rigor, vanity metrics, churn→LTV math, signal quality | [metrics-and-experimentation](references/metrics-and-experimentation.md) |
Some principles carry a **Visual.** line — a text description of the diagram or screenshot from
the original post. The original image is always one click away via the principle's **Source** link.
## Gotchas
- **Scarcity must be real.** Booking's "1 room left" works because it's true. Fabricated
scarcity destroys trust when detected (and is illegal in several markets). Never invent
counters, timers, or stock levels.
- **Loss aversion vs dark pattern** — the line: the claim is true and the exit stays easy.
Framing a real loss is persuasion; manufacturing fear or trapping cancellation is not.
- **"Kill outbound links" is a conversion-page rule.** Blog posts, docs, and SEO pages need
outbound links; don't export LP rules to content.
- **4.2–4.5 stars means _let real criticism show_** — never fabricate negative (or positive)
reviews to hit the number.
- **Don't answer this skill with 30 A/B tests.** The metrics principles themselves warn against
underpowered tests: compute the sample-size floor first (`scripts/revenue_math.py sample-size`),
test big levers, and below the floor decide by qualitative research.
- **Evidence is benchmark, not guarantee.** Figures come from the author's cases (mostly
Brazilian SaaS, values in BRL). The mechanism transfers; the exact percentage may not.
## Related skills
RCD supplies the **principle and its evidence**; execution skills own the workflow. The skills
below are from [Corey Haines' marketingskills](https://github.com/coreyhaines31/marketingskills)
— if they aren't installed, apply the RCD principles directly instead of deferring.
Full page-audit workflow → `cro` · cancellation-flow build → `churn-prevention` · test design &
stats → `ab-testing` · writing the copy → `copywriting` · pricing-page build → `pricing` ·
post-signup flow build → `onboarding`. When one of those runs, cite RCD principles inside it
rather than duplicating its process here.
## Project log (`rcd-log.md`)
Per-project memory, kept in the project root — read at the start of every engagement (step 1),
appended at the end (step 6). One entry per engagement:
## 2026-07-02 — pricing page redesign
- via: rcd (direct) # or the skill that led the run: cro, pricing, churn-prevention…
- principle: Decoy effect (pricing-and-monetization)
- move: added GBB middle tier at 80% of the top price
- result: pending # update when data arrives: "+12% upgrades", "no effect"
The `via:` field doubles as trigger telemetry: if entries accumulate where RCD led a run an
execution skill should own (a full page audit, a cancellation build), that is the signal to
narrow this skill's description to the principle/evidence angle.
## License
Source-available, **not** open-source — see [LICENSE](LICENSE) (must accompany any copy or
derivative, in full): attribution to @richardrx required; gambling/betting/casino use
prohibited. This skill is a derivative of
[heliocosta-dev/revenue-centric-design](https://github.com/heliocosta-dev/revenue-centric-design),
restructured and extended here (gotchas, audit template, revenue-math script, project log).

57
.github/skills/revenue-centric-design/references/ai-era-differentiation.md vendored Normal file
View File

@@ -0,0 +1,57 @@
# AI-Era Differentiation & Moats
> Curated, distilled wisdom from @richardrx ("Richard — Design for startups"), translated from Portuguese. Each entry is a reusable principle linked to its source post.
## Faster building doesn't fix churn — activation does
**Principle.** Build speed was never the bottleneck. Shipping the same confusing interface faster is just a more efficient route to churn. The real gap is the space between a user entering the product and understanding what to do.
**Apply when.** Vibe coding is sold to a founder as "the product got cheaper to build," and the team equates speed with progress.
**The move.** Obsess over activation, not velocity. Attack the three things vibe coding never touches: onboarding, attention hierarchy, and value delivery in the first sessions (TTV). More products now compete for the same user attention, so close the entry-to-understanding gap.
**Voice.** "Build speed without an obsession for activation is just a more efficient way to reach churn."
**Source.** [@richardrx · 2026-04-02](https://x.com/richardrx/status/2039685818273378644)
## Same engine, different UX: don't compete on the commodity
**Principle.** AI turned your codebase into a near-commodity — under the hood ~90% of new tools call the same APIs. Engineering solves the base function; design and packaging are what differentiate and resist copying.
**Apply when.** Your "engine" is effectively identical to a competitor's and a generic interface is pulling you into a price war.
**The move.** Win on UX architecture, not the engine. Superior UX (1) removes initial friction → lifts conversion; (2) fits the user's workflow → cuts churn, raises LTV; (3) eases continuous/collaborative use → enables upsell. This builds a differentiator that lowers copy risk and keeps the customer paying.
**Evidence.** VW Up, Seat Mii, and Skoda Citigo share the exact same platform — chassis, drivetrain, and the identical EA211 engine — yet are designed and packaged for different ICPs (young, pragmatic-utility, reliability).
**Voice.** "The engine may be identical, but it's the architecture of the user experience that builds your moat."
**Source.** [@richardrx · 2026-03-03](https://x.com/richardrx/status/2028837448717926518)
## A validated idea is a short-term game — plan the moat
**Principle.** If your only advantage is the codebase, you've merely built a validated MVP for better-funded competitors to execute. Structural barriers to entry (the moat) are planned, never accidental.
**Apply when.** Your tech is easy to replicate and the product identity is generic — clones can ship within days.
**The move.** Plan three deliberate moats: (1) Brand Power — a proprietary visual identity with above-average UX signals less risk and sells perceived safety (conversion); (2) Switching cost via UX — intuitive flows users have internalized make moving to a 20%-cheaper clone costly in productivity (retention); (3) Expansion architecture — internal network effects (invite-to-collaborate) are harder to copy and pull in new users (LTV). Users who perceive a value ecosystem prefer paying more over adapting to a worse, cheaper product.
**Evidence.** The "Roast My Startup" tool was cloned within a week — copies flooded the timeline — proving a codebase-only edge is no defense.
**Voice.** "Clones can't copy trust."
**Source.** [@richardrx · 2026-03-03](https://x.com/richardrx/status/2028777831233114152)
## Design for shrinking attention spans
**Principle.** Short-form video acts on the brain like a variable-reward slot machine, switching off the attention filter and eroding self-control — leaving users with high anxiety and low focus. If your user's attention keeps shrinking, design must be militarily focused.
**Apply when.** Building any product whose users are conditioned by infinite short-video feeds (attention economy).
**The move.** Engineer for the attention limit: (1) drastically reduce cognitive load; (2) direct absolutely toward the target task (conversion); (3) build interfaces that respect the human attention ceiling.
**Evidence.** An EEG study (Fabiano et al., Frontiers in Human Neuroscience) links short-form-video addiction to reduced frontal-lobe activity and weakened ability to focus.
**Voice.** "TikTok is the hot dog of social media — hyper-palatable, but nutrient-poor: you consume endlessly and it never nourishes you."
**Source.** [@richardrx · 2026-02-26](https://x.com/richardrx/status/2026978144343687177)
## Sell the value, not the feature list — and show the product
**Principle.** AI auto-generates landing pages, but ~90% share the same defects. A generic, inconsistent hero is what loses visitors, not the absence of fancy design.
**Apply when.** Auditing an AI-generated LP that leans on features and trendy gradients instead of the user's pain.
**The move.** Fix the three recurring failures: (1) generic, inconsistent aesthetics (every component a different color, purple/green gradients on dark/white); (2) over-indexing on features instead of the value/pain they address; (3) barely showing the actual product. Build the hero to pass the 5-second test — what does this do, why care, what next.
**Visual.** Supafast's "SaaS Hero Section Formula" — 5 elements with before/after copy: Headline (≤8 words, attack the #1 pain), Subheadline (show the transformation), Primary CTA (specific to outcome), Secondary CTA (low-commitment), Trust Bar (5 logos or one specific number)
**Source.** [@richardrx · 2026-02-21](https://x.com/richardrx/status/2025246347587162602)
## Don't let AI-to-Figma-to-code factory technical debt
**Principle.** A Claude → Figma → Code flow looks like speed but creates two documents that drift out of sync, plus inconsistent components — a maintenance Frankenstein, not velocity.
**Apply when.** A tool promises round-tripping AI output through the design canvas into code as a shortcut.
**The move.** Refuse the false shortcut. Without a design system and context, generated components are superficially similar but fundamentally inconsistent (random button padding, off-brand colors, inconsistent UX patterns). Manual tweaks don't flow back to code, so the source of truth reverts to the canvas and the two files desync. For devs it's useless (V0/Lovable already emit code without the full-seat toll); for designers it's a distraction that skips information architecture to spit out a screen fast.
**Voice.** "A technical-debt factory."
**Source.** [@richardrx · 2026-02-18](https://x.com/richardrx/status/2024076972565963186)
## Beat the four AI failure modes that make a vibe-coded SaaS feel like a fraud
**Principle.** When the barrier to entry tends to zero, competition tends to infinity. AI lowered that barrier and amplified Dunning-Kruger — you feel omniscient but lack the base to judge if its output is a solution or wasted time. You shipped code, not a seductive product, and you're diving into a red ocean. Code is no longer the asset — just one ingredient.
**Apply when.** A "complete SaaS in a weekend" is validated but feels hollow, generic, and clonable by Wednesday.
**The move.** Fix the four pillars where AI fails: (1) **Generic-product trap** — AI is trained on the internet's average, and average builds nothing extraordinary; escape commoditization by building the only possible tool for an ignored niche (not "CRM for doctors" but "CRM for facial-harmonization clinics" with a `last_toxin_date` field and a 110-day retouch-alert cron). (2) **Value delivery / TTV** — don't ship login → empty dashboard (the 99% default); build an on-ramp to value, an onboarding assistant, not a desert. (3) **Trust / visual confidence** — in a sea of V0/Tailwind templates, aesthetics, personality, and consistency are the last remaining trust proxies; intentional, human-aligned pixels signal authority, build trust, and lower CAC. (4) **Human touch** — cheaper code should buy more time for the memorable details (kind error messages, a 404 that returns the user, business logic that anticipates mistakes, a 200ms confirming micro-interaction) — humans are predictably irrational, full of bias.
**Evidence.** Johnson & Goldstein (2003), *Science* — a mere "opt-out" default produced +90% organ-donation consent, proving small design choices move behavior.
**Visual.** The Dunning-Kruger curve — confidence spikes at "Ignorant" (low knowledge), craters at "Cultured," and climbs toward "Expert," with a labeled "confidence gap"
**Voice.** "You can own all the cement in the world, but without the blueprint and structural engineering you're just a pile of gray concrete."
**Source.** [@richardrx · 2026-01-19](https://x.com/richardrx/status/2013264068518289753)

34
.github/skills/revenue-centric-design/references/audit-template.md vendored Normal file
View File

@@ -0,0 +1,34 @@
# Audit Output Template
Deliver every audit run (landing page, pricing page, onboarding flow, cancellation flow,
features page) in this shape. A row may only enter the table once it passes the completion
criterion — named mechanism + cited principle + attached evidence.
## Header
> **Target:** <URL or flow name> · **Date:** <date>
> **ICP:** <buying criteria — trigger, pain, prior attempt, proof needed; not demographics>
> **Awareness level:** <Schwartz stage of the arriving traffic>
> **Verdict:** <one sentence — the single biggest revenue leak found>
If ICP or awareness level can't be stated, that **is** finding #1 — the debug order starts
there (ICP → awareness → proof → visual), never at the visual layer.
## Findings
Ordered by expected revenue impact, not by position on the page.
| # | Finding | Mechanism | Principle (reference file) | The move | Evidence |
|---|---------|-----------|----------------------------|----------|----------|
| 1 | Hero opens with the product category, not the visitor's pain | 5-second test | "Pass the 5-second test — lead with the problem" (conversion-and-landing-pages) | Rewrite the first line to the pain: "Your team loses 6 hours a week hunting for information" | Descriptive vs transformation-led LPs: 0.5% vs 3% across 30 Brazilian SaaS ([source](https://x.com/richardrx/status/2045154631974539650)) |
When a finding involves numbers (A/B sample size, churn→LTV, CAC per closed deal), paste the
actual `scripts/revenue_math.py` output into the Evidence cell — never an estimate.
## Close
- **Do first:** the top 1–3 moves, each with one line on why it outranks the rest.
- **Don't:** any tempting change the SKILL.md gotchas rule out (fabricated scarcity, LP rules
applied to content pages, review manipulation, underpowered tests…).
- Append the prescribed moves to the project's `rcd-log.md` (format in SKILL.md) so the next
engagement starts from what was already tried.

60
.github/skills/revenue-centric-design/references/behavioral-science-toolkit.md vendored Normal file
View File

@@ -0,0 +1,60 @@
# Behavioral Science & Persuasion
> Curated, distilled wisdom from @richardrx ("Richard — Design for startups"), translated from Portuguese. Each entry is a reusable principle linked to its source post.
## You can't un-hear your own product
**Principle.** Once you know how your product works, that knowledge permanently rewrites your perception — what feels obvious to you is just "tap-tap-tap" to a first-time user. You'll mistake confused users for dumb users.
**Apply when.** You think onboarding is unnecessary because "the product is simple," or a question your support answers weekly seems already-answered on screen.
**The move.** This is the curse of knowledge — you can't switch the music off, so collect feedback from people who've never seen the product, without steering or naming things, and watch behavior. Run it continuously: use support/CX as an insight collector (tabulate each issue by %, impact, insight), Clarity/PostHog for heatmaps and session replays, sampled user interviews, and competitor benchmarking.
**Evidence.** Tapping-vs-listening study: tappers hear the full song in their head; listeners only get the taps. Listeners guessed 3 of 120 songs correctly (2.5%) — far below tappers' expectations.
**Voice.** "You're humming the whole song in your head; your user only hears 'tap, tap, tap.'"
**Source.** [@richardrx · 2026-06-04](https://x.com/richardrx/status/2062509937037590997)
## Set the default — it's the most underrated lever in conversion
**Principle.** The pre-selected option captures the overwhelming majority of choices, because deciding is expensive and the lazy brain takes the easiest path. Smart defaults beat copy persuasion.
**Apply when.** Any choice the user must make — pricing tier, billing cadence, seat count, notifications, checkout — especially before you spend hours rewriting CTAs.
**The move.** Exploit status-quo bias plus cognitive-load reduction. Pre-select the mid-tier you want to sell (the default takes 60–80% of choices); default billing to annual to lift contracted MRR without changing price; start the seat selector at your ICP's typical count (anchoring); run a reverse trial where premium is the default and free is the opt-out, so the user must actively give up what they already have. Three rules: defaults must be ethically defensible (checkbox tricks become churn and complaints), smart defaults beat copy, and a default acknowledges the user won't burn energy deciding what's trivial to you.
**Evidence.** Organ-donor study (Science, 2003): opt-out countries register ~6× more donors than opt-in. Germany (opt-in) ~12% vs Austria (opt-out) ~100% — culture/religion don't explain it; it's a pre-checked box. Richard raised average ticket 60% and saw up to 4× LTV applying this to plan acquisition.
**Visual.** Bar chart of effective organ-donor consent by country: opt-in nations low (Denmark 4.25%, Germany 12%, UK 17.17%, Netherlands 27.5%) vs opt-out nations ~100% (Austria, France, Hungary, Portugal).
**Voice.** "You can spend the rest of your life optimizing copy, or you can change 5 defaults over the weekend."
**Source.** [@richardrx · 2026-05-22](https://x.com/richardrx/status/2057872036718899256)
## Reinforce the decision the user just made
**Principle.** After committing to a choice, people actively seek information that supports it — choice-supportive bias. You can feed that need to make the decision feel right.
**Apply when.** Right after signup, purchase, or any meaningful commitment, when buyer's remorse or doubt could creep in.
**The move.** Use choice-supportive bias deliberately: send a strong welcome email with clear next steps so the new user feels embraced and validated in having chosen you. (Analogy: someone joins an EV-lovers group right after buying the car.)
**Source.** [@richardrx · 2026-04-29](https://x.com/richardrx/status/2049392897598849333)
## Architect for what users fear losing, not just what they gain
**Principle.** Builders obsess over features (gains), but conversion and retention are cemented by what the user fears losing — the pain of abandoning a built-up ecosystem outweighs the pain of paying a subscription.
**Apply when.** Designing trials, retention/renewal flows, and offboarding for any product where users accumulate data, history, or workflows.
**The move.** Exploit loss aversion and sunk cost. Convert with zero risk — let users import real competitor data in shadow mode so they can test without fear. Retain by designing the product to make users build workflows and accumulate history from day 1, so at renewal they weigh the headache of rebuilding from scratch, not the monthly fee. At offboarding, don't add friction — make cancellation easy but the loss tangible (e.g., "You'll instantly lose 41 active automations and 6 months of data").
**Evidence.** Sunk cost is what keeps many users on certain LLMs — fear of losing your memory/history, even when it could be exported with a copy/paste.
**Visual.** Hotel-listing UI using scarcity ("Only 2 rooms left") — illustrative example of a loss-framed cue.
**Source.** [@richardrx · 2026-03-02](https://x.com/richardrx/status/2028471666297217460)
## Use precise numbers, not round ones, to signal truth
**Principle.** Exact figures read as more credible than rounded ones; round numbers signal marketing while specific numbers signal reality.
**Apply when.** Writing any claim, stat, or social-proof number — landing pages, ads, results, testimonials.
**The move.** Lean on the precise-number effect: say "526 houses," not "over 500." Nothing about the claim changes except the precision, yet trust rises.
**Evidence.** Schindler & Yalch (2006), 199 participants, fictional deodorant: claims of "47%" or "53%" longer-lasting were judged ~10% more accurate than the rounded "50%" claim — only the precision changed.
**Visual.** Real billboard: "LAST YEAR WE SOLD 526 HOMES. YOUR COUSIN SOLD 2. LIST RESPONSIBLY." — the precise number doing the persuasion.
**Voice.** "'526 houses' inspires confidence; 'over 500 houses' signals marketing."
**Source.** [@richardrx · 2026-02-18](https://x.com/richardrx/status/2024141244717281514)
## Guide the eye — don't give every option equal weight
**Principle.** The brain uses contrast to make fast decisions (Von Restorff effect). When competing options carry identical visual weight, you create mental friction, decision time rises, and conversion falls. Guiding the user isn't manipulation — it's respect for their time.
**Apply when.** You have 3 plans, two equally-weighted buttons, or any "democratic" interface where everything looks the same (a common founder error, and a default of AI-generated UIs).
**The move.** Exploit the Von Restorff effect: make the value-generating option visually dominant and de-emphasize the rest (e.g., a ghost-styled "Cancel" beside a bold, colored primary). If you know your ICP's pains and desires, you have a duty to highlight the highest-value solution. Slow decisions accumulate into a "hard-to-use" perception that becomes churn.
**Visual.** Good/bad confirm dialog: bad = both buttons same green weight; good = a ghost-text "Cancel" beside a solid red "Delete now," so the primary action stands out.
**Voice.** "If everything grabs attention, NOTHING grabs attention."
**Source.** [@richardrx · 2026-01-22](https://x.com/richardrx/status/2014317885494059106)
## Cut cognitive load — every choice you remove can lift conversion
**Principle.** Each extra field, choice, or block of complex text spends the user's mental energy and triggers analysis paralysis. Your product can be complex; your interface doesn't have to be.
**Apply when.** Checkout, signup, and subscription screens — anywhere the user must decide or input under doubt.
**The move.** Strip the interface to the essential decision; when options can't be cut, break the flow into smaller steps (e.g., a 4-step checkout). Remember the failure is invisible: users don't complain or open tickets — they close the tab as "silent churn" and your CAC is wasted. Then dogfood your own onboarding as if you were a stranger.
**Evidence.** Removing 1 checkout field raised conversion 10%. Richard has seen reworked subscription screens lift LTV 200% just by simplifying the decision.
**Visual.** Mobile checkout labeled "Analysis paralysis": payment options split into smaller stages — advice to break choices into ~4 steps.
**Voice.** "They don't file a support ticket. They just close the tab — and your CAC goes in the trash."
**Source.** [@richardrx · 2026-01-20](https://x.com/richardrx/status/2013597792447394034)

66
.github/skills/revenue-centric-design/references/churn-and-retention.md vendored Normal file
View File

@@ -0,0 +1,66 @@
# Churn & Retention
> Curated, distilled wisdom from @richardrx ("Richard — Design for startups"), translated from Portuguese. Each entry is a reusable principle linked to its source post.
## Churn and payback are one problem, measured in two places
**Principle.** Churn and payback look like two problems (often with different owners) but are the same one — both decided in the user's first session, in the gap between entering and feeling the product works.
**Apply when.** You're fixing churn at the cancel screen and chasing cheaper CAC in the ad manager at the same time.
**The move.** Both are the wrong place — the decision was made earlier. Activation fixes both: shrink TTV and you retain more AND get each customer across the payback line before they vanish. A customer who dies in month 2 with a 4-month payback never closes the account. Caveat: not all churn is activation (price, a bad channel exist) — but before chasing cheaper CAC, count how many customers die before repaying what they cost.
**Voice.** "A customer who dies before repaying his CAC is a bill you paid and never collected."
**Source.** [@richardrx · 2026-06-30](https://x.com/richardrx/status/2071931705573748896)
## Switching cost is what turns months of LTV into years
**Principle.** The same product, designed differently, yields months vs. years of LTV. Switching cost = the effort a user *perceives* in leaving; low switching cost means thin history and an easy exit.
**Apply when.** Designing for retention/lock-in, or explaining why a useful product still churns.
**The move.** Engineer switching cost deliberately — a "compound interest" that grows the product's value over time. Five levers: **muscle memory** (Superhuman/Photoshop shortcuts), **mental model** (Mac↔Windows, Gmail labels), **accumulated personalization** (Spotify playlists, home-screen layout), the **vault effect** (iCloud/Drive/years of WhatsApp), and **autopilot/habit** (variable-reward Skinner-box loops). Sunk cost holds them the way it holds an investor in a falling stock; habit can take months to install but is the difference between LTV of months and years.
**Source.** [@richardrx · 2026-06-23](https://x.com/richardrx/status/2069382946214080985)
## Tell the existing base about upgrades before they want to leave
**Principle.** Reactive improvement communication is a sneaky churn vector: if you only market new versions to cold traffic to avoid cannibalizing the old product, your base assumes the old version is the ceiling and leaves when a competitor looks better.
**Apply when.** You shipped a better version/feature but only announced it externally; support pitches the migration only at the cancel moment.
**The move.** Proactively offer upgrades and migrations to active users — not release emails nobody reads or an Instagram post. Ask: "When we shipped the last relevant feature or version, how many active customers were told?" Offering migration at cancellation converts a clean expansion into emergency retention. Track Net Revenue Retention (NRR); in B2B SaaS, NRR above 110% separates sustainable growth from a leaky funnel.
**Voice.** "The migration only showed up as a reaction to my complaint, after I started making noise."
**Source.** [@richardrx · 2026-05-19](https://x.com/richardrx/status/2056715097796411514)
## Design the cancel screen — it's your last conversation, not a form
**Principle.** The cancellation screen is the most ignored yet one of the most important pages in the product; treating it as a bureaucratic form wastes your final chance to retain.
**Apply when.** Cancel flow is just "Are you sure?" with two buttons, or Stripe's default template, while signup was crafted with care.
**The move.** Three plays: (1) Show concrete loss — "You'll lose access to relationship data on your 476 configured clients and 8 months of history"; concrete loss outweighs abstract benefit (loss aversion). (2) Offer an alternative before goodbye — "Pause 30 days instead?" or one more free month (ChatGPT nails this). (3) Collect the reason usefully via an open question — "What was missing for you to stay?" — not a dropdown that rarely lists the real reason.
**Voice.** "If you invested to bring your user here, invest the last 30 seconds trying to keep them."
**Source.** [@richardrx · 2026-05-13](https://x.com/richardrx/status/2054562119962501186)
## Churn starts on the landing page, not at cancel
**Principle.** If the LP promises one thing and the product delivers another, you create an expectation debt that charges interest every day the user thinks "this isn't what I expected" — and the disappointment is pre-programmed even if the product is excellent.
**Apply when.** 30-day churn is high but NPS is fine — the problem is likely what you promised before they entered, not the product.
**The move.** Audit three common mismatches: (1) result promise vs. tool delivery ("Increase sales 30%" → a metrics dashboard); (2) simplicity promise vs. complex product ("Set up in 5 minutes" → 47 fields, 3 integrations, 20-min tutorial); (3) promise aimed at the wrong ICP (LP speaks to a 2-person startup; product was built for a 15-person team at scale). Recalibrate the promise to match real delivery and ICP.
**Voice.** "The user bought a result and got a colorful spreadsheet."
**Source.** [@richardrx · 2026-04-29](https://x.com/richardrx/status/2049568514311172355)
## Anchor one-time-job products to a recurring life event
**Principle.** A product hired to solve a one-off problem generates structural churn: a user who loves it and still cancels isn't unhappy — they finished the job they came to do, and there was no continuous value to bring them back.
**Apply when.** You have 4.2 stars and positive NPS yet rising churn; the product solves a point problem (resume builder, contract/legal-doc generator, data migration tool, pitch-deck builder, due-diligence platform).
**The move.** Use jobs-to-be-done thinking. Instead of faking engagement, anchor the product to an event that already recurs in the user's life and returns yearly without a push. Ask: "If the job my product does is finished, what life event justifies the user coming back?" If there's no answer, it's a business-model problem, not a product problem.
**Evidence.** TurboTax (US income-tax software) tied itself to tax season, turning the product into a ritual because the event makes the use inevitable.
**Source.** [@richardrx · 2026-04-24](https://x.com/richardrx/status/2047620778338795918)
## Treat support volume as a design problem, not a staffing one
**Principle.** What looks like a support problem is usually a design problem — you cut ticket volume during onboarding itself, with an interface that answers questions before they're asked. Low-ticket digital products generate up to 3x more support than conventional tickets.
**Apply when.** Support is your biggest bottleneck, especially with low-ticket/impulse-buy products; you're tempted to just automate tickets.
**The move.** Three drivers of low-ticket support load: different buyer profile (less patience, less digital familiarity, more expectation of human help); impulse purchase (low friction → buys without understanding → seeks support); inverted opportunity cost (asking is easier than searching). For SaaS, redesign the journey: contextual in-product FAQ, self-answering UI. Automating support treats the symptom; redesigning the journey fixes the cause and can cut churn too.
**Evidence.** An old McAfee case cut support volume by 90% by implementing an FAQ — plain text, no chatbot.
**Source.** [@richardrx · 2026-04-23](https://x.com/richardrx/status/2047289409238712726)
## Strip the jargon before you blame onboarding
**Principle.** Churn that looks like a product problem is often a language problem: a technical founder writes product and sales copy in jargon, the ICP buys on a leap of faith, never perceives value, accumulates small disappointments, and cancels — looking like a missing feature on the dashboard.
**Apply when.** Churn is high and you've already revised onboarding and product — revise language and structure next.
**The move.** Watch two biases: the curse of knowledge (you know too much and forget the other person doesn't) and the easy-speech bias (simple language reads as more trustworthy and raises awareness). Rewrite dense, jargon-heavy copy into plain language even for complex topics.
**Visual.** Side-by-side: a dense, legalese contract clause (red X) vs. a plain-language rewrite "In this contract you authorize the bank…" (green check), with an "Easy-speech bias" callout.
**Voice.** "On the dashboard it looks like a missing feature; it was a mismatch between your discourse and their understanding."
**Source.** [@richardrx · 2026-04-08](https://x.com/richardrx/status/2041829519611371727)
## Engineer addiction like a game so CS isn't a churn tax
**Principle.** Customer Success is the tax you pay for a non-addictive product — if you need an army of CSMs to stop cancellations, the product failed. Your real competitor isn't another startup; it's boredom, and boredom has infinite CAC. The CNPJ buying your SaaS is the same brain that plays Candy Crush; reward neuroscience is identical. Win retention across three game-design phases.
**Apply when.** Diagnose by behavior: dropout at minutes 2–8 of onboarding = Phase 1; ~1.3 logins/week when it should be 4x = Phase 2; one departing employee kills the whole account = Phase 3. LTV:CAC below 3:1 means you're funding a product that can't stand organically.
**The move.** **Phase 1 — Time-to-value vs. cognitive load:** ditch the setup wizard (asking work before delivering value reads as hostile territory in 10 seconds); use progressive disclosure, let users create and see results before asking for email/card. Mechanism: Zeigarnik effect (incomplete-loop tension) + endowment effect (people value 3x more what they helped build). Empty states must sell the dream — never show "0 data" or blank templates; populate a demo simulating day-30 usage. **Phase 2 — Habit loop / retention as biology:** passive software that only reacts is a failure to build dependency; ship proactive variable rewards. The mesolimbic reward system releases dopamine on anticipation, not the reward itself; predictable rewards (monthly report) build tolerance, variable ones ("we detected a positive anomaly yesterday") keep the loop alive. Convert vanity metrics into loss-aversion triggers: "Your team broke a record and you haven't seen it" + temporal data scarcity ("sync in 24h or lose the weekly benchmark"). **Phase 3 — Defensive moat:** single-player products die when the champion leaves — you built dependence on a person, not the org. Build multiplayer mode + data debt via social switching cost + network effects. Make User A's work block/depend on User B; reports needing multi-stakeholder approval; dashboards aggregating 3 departments. When quitting requires an alignment meeting across Sales, Ops and Finance, you reach negative churn by bureaucratic inertia; accumulated datasets add organizational endowment effect.
**Voice.** "Stop blaming the customer. Your product is boring. And in the attention game, boring is bankruptcy."
**Source.** [@richardrx · 2026-01-30](https://x.com/richardrx/status/2017274698699067466)

132
.github/skills/revenue-centric-design/references/conversion-and-landing-pages.md vendored Normal file
View File

@@ -0,0 +1,132 @@
# Conversion & Landing Pages
> Curated, distilled wisdom from @richardrx ("Richard — Design for startups"), translated from Portuguese. Each entry is a reusable principle linked to its source post.
## "Paint the button" is judging the last line of a long cascade
**Principle.** The visual layer of a landing page is the tip of a cascade that begins at the ICP — so "make it more colorful" critiques the *end* of a process as if it were the start.
**Apply when.** Gathering or acting on LP feedback, or reaching for button color first.
**The move.** Work the cascade in order: ICP → the buyer's **awareness level** (Eugene Schwartz's 5 stages) → what you say and how → visual positioning (tone → form, color, type, space). Low awareness: open on the problem, name the pain in the first fold, then the mechanism and your fix. Higher awareness: go straight to your advantages vs. alternatives. Jump to button color and you've silently (and probably wrongly) answered who you sell to, their awareness, which pain, and what tone.
**Voice.** "They're looking at the end of a process and thinking it's the beginning."
**Source.** [@richardrx · 2026-06-23](https://x.com/richardrx/status/2069469303464730988)
## Scaling cold traffic is an honesty test for your page
**Principle.** Conversion measured on warm audiences (existing followers) is inflated — they forgive the page's flaws. Only cold traffic reveals how much of your rate is the page versus borrowed trust.
**Apply when.** Ad spend goes up but revenue barely moves, and the conversion rate "drops" even though you fixed nothing.
**The move.** Treat any baseline built on warm public as fiction, not a performance number. When you scale budget into cold traffic and the rate collapses, that gap is the page's real ceiling — fix the obvious page defects first, then judge media.
**Evidence.** A founder tripled ad spend; revenue rose only 11%. Warm-traffic conversion of ~3% (driven by Instagram followers) cratered to 0.5% on cold traffic; fixing the obvious lifted it to 2% — 4x the real number.
**Voice.** "Scaling budget into cold traffic works like an honesty test: it shows how much of your conversion is your page and how much is borrowed trust."
**Source.** [@richardrx · 2026-06-08](https://x.com/richardrx/status/2063972594223661127)
## Write the CTA microcopy, not the button color
**Principle.** The CTA label moves conversion far more than button color or shape — it's the last thing the user reads before deciding, so it must reduce the mental effort of simulating what happens next.
**Apply when.** A technical founder is A/B testing button colors for weeks over a 0.3% (non-significant) delta while the label still says "Sign up" with no click trigger.
**The move.** Make the CTA answer the three questions the brain asks before clicking: (1) what happens when I click, (2) how long it takes, (3) what it costs/commits. "Sign up" answers none; "Start free in 30s" answers two. Add a click trigger (the small line under the CTA) to answer the rest and break an objection: "Start Free Trial" + "14 days, no card". Rooted in outcome bias and Construal Level Theory (making the outcome concrete).
**Voice.** "The button is the last thing the user reads before deciding — treat it as such."
**Source.** [@richardrx · 2026-05-25](https://x.com/richardrx/status/2058875777739866490)
## Cut CAC by filtering on the landing page, not the media
**Principle.** High CAC is rarely a media problem — it's usually a filtering problem. The landing page can qualify leads before the form, killing deals that were never going to close.
**Apply when.** Clicks become leads, leads become SDR calls, calls don't close, and founders react by swapping creative or channel (or blaming the SDR).
**The move.** Pull three filtering levers so the lead self-selects without feeling filtered: (1) copy specificity — "Cash flow for service providers billing R$500k–R$3M" beats "Organize your finances"; (2) visible pricing — screens out no-budget leads so sales only meets real objections; (3) a qualification question as the first interaction (or copy/examples that play that role). CAC is cost per closed deal, not cost per lead.
**Evidence.** If you close 1 in 50 leads, dropping to 30 leads with 1 close nearly doubles media efficiency — same spend, lower CAC.
**Voice.** "The lead with no money won't magically generate money between the landing page and the sales call."
**Source.** [@richardrx · 2026-05-18](https://x.com/richardrx/status/2056385204785213446)
## Aim for 4.2–4.5 stars, not a perfect 5
**Principle.** A flawless rating reads as fake; consumers distrust unanimity. A profile that includes constructive criticism feels more authentic and converts better than pure praise.
**Apply when.** Building or curating reviews/ratings and social proof on a landing page or product.
**The move.** Let imperfection show. Stack the highest-converting combination: a detailed case study with quantifiable ROI, ideally a video testimonial from a brand recognizable to your ICP, plus customer logos. Avoid the common failure — social proof done badly: identical cards, first-name-only, all 5 stars, generic avatars instead of a real photo. Make each review verifiable (e.g., link to the actual LinkedIn post).
**Evidence.** Northwestern: purchase probability peaks at 4.2–4.5 stars, not 5. Testimonials lift LP conversion up to 34% (VWO); 5+ reviews → 270% more likely to be bought (Yotpo); 93% read reviews before buying (BrightLocal).
**Visual.** A social-proof wall done right: real faces, names + roles, verifiable reviews, and recognizable client logos (Volkswagen, Coca-Cola, Samsung, Itaú, Volvo, DocuSign, Hotmart) under a "+10,000 professionals" headline
**Voice.** "Credibility drives conversion."
**Source.** [@richardrx · 2026-05-14](https://x.com/richardrx/status/2054910531132183030)
## Compress the decision window — shorten the path to action below the path to doubt
**Principle.** The purchase decision is mostly made before checkout. Conversion is architecture: the user converts when the path to the action is shorter than the path to doubt, because every extra second of deliberation raises the odds they close the tab.
**Apply when.** You're optimizing the checkout, CTA, or headline while ignoring the deliberation time upstream.
**The move.** Compress decision time two ways: pressurize the environment with real scarcity ("Only 1 room left at this price", "Booked 3 times in the last hour") — driven by loss aversion (Kahneman: losing hurts up to ~2x more than the equivalent gain), so "1 left" registers as "I'll lose this"; or remove steps entirely (Amazon's patented 1-Click ordering).
**Evidence.** Booking runs 1,000+ simultaneous experiments (per Lukas Vermeer, 8 yrs leading experimentation); showing sold-out hotels alongside available ones *raised* bookings by sharpening perceived scarcity.
**Voice.** "The user converts when the path to the action is shorter than the path to doubt."
**Source.** [@richardrx · 2026-05-11](https://x.com/richardrx/status/2053832356293738914)
## Don't clone the page — the converting layer is invisible
**Principle.** A converting sales page is the visible shell of an invisible system. Copying the layout copies what's cheapest to produce; the expensive part (research) stays invisible and gets left behind.
**Apply when.** A page "looks like it works" and you're tempted to clone it for your own offer.
**The move.** Before copying, ask: what on this page is a function of the *product*, and what is a function of the *customer research* done before writing each line and placing each element? The hidden layer is customer vocabulary pulled from interviews, objections ordered to the ICP's specific fear, social proof hand-picked to resonate, and a core promise tuned to the stated desire. This error has a name: cargo cult — replicating the visible ritual hoping to summon the result, without grasping the causal mechanism.
**Evidence.** Two ~95%-identical pages, same niche/offer/order: 4% vs 0.6%. The 4% page belonged to Richard's client; the 0.6% was a near-pixel clone by a builder — who unknowingly DM'd the original team to complain it wasn't converting.
**Voice.** "There's a name for copying the form expecting the function. It's called cargo cult."
**Source.** [@richardrx · 2026-05-09](https://x.com/richardrx/status/2053086365781205142)
## Pass the 5-second test — lead with the problem, not the feature
**Principle.** A visitor scans your LP for ~5 seconds. If they can't answer what it does, who it's for, and why they should care, they leave — they won't stay to figure it out. Conversion begins on the first line.
**Apply when.** Your hero opens by describing what the product *is/does* (a feature/category) instead of the visitor's pain.
**The move.** Open with the visitor's problem, not yourself. "AI-powered project management platform" → meh; "Your team loses 6 hours a week hunting for information" → keep going. One talks about itself, the other talks about me. If the first line doesn't connect with the pain, the rest of the page goes invisible.
**Voice.** "If you who built it can't answer in 5 seconds, your visitor can't either."
**Source.** [@richardrx · 2026-05-01](https://x.com/richardrx/status/2050280273682510230)
## Fix contrast, layout, and trust before rewriting copy
**Principle.** The brain processes three pre-verbal variables — contrast, on-screen placement, and trust — before it reads the words, and they usually move the needle more than any headline.
**Apply when.** You're rewriting home copy for weeks chasing conversion without touching the visual/attentional layer.
**The move.** (1) Contrast is relative to surroundings, not an isolated color — a big, differently-shaped button in whitespace is *seen before it's read* (accessibility ratio 4.5:1 is a floor, not the goal). (2) Placement: Fitts's law (closer, bigger targets get clicked more) plus the F-pattern (Nielsen Norman eyetracking) — a CTA bottom-right with no visual anchor sits outside the attentional map. (3) Trust is built in four reinforcing layers: a real (non-stock) human face, a known brand or specific verifiable proof, an offer that fits the stated problem, and a solution mechanism logically coherent with the promise. When all four align, the user converts without knowing why; when one fails, they invent a rational reason to leave.
**Voice.** "Optimizing copy without working contrast, placement, and trust is masking the symptom."
**Source.** [@richardrx · 2026-04-28](https://x.com/richardrx/status/2049215050997784774)
## Specificity is the difference between decoration and persuasion
**Principle.** Generic LP copy gets discarded by the brain; specific messages are processed faster and generate more trust. A converting LP shows the *transformation* the product causes, not just what it does.
**Apply when.** Your hero reads "The complete platform for [generic category]", a 5-feature subtitle, "Start free", and no real social proof — the default Brazilian SaaS template.
**The move.** Fix the four standard failures: (1) "Complete platform" means nothing → "Cut new-dev onboarding time 40%"; (2) features on top → lead with the pain, feature as the solution (nobody wakes up wanting a "custom report feature"); (3) "Start free" is the weakest CTA (no value, no risk reduction, no urgency) → "See your first report in 2 minutes"; (4) "Used by Company X" proves nothing → "We cut Company X's churn from 12% to 6%". One page describes, the other sells.
**Evidence.** Across 30 Brazilian SaaS LPs: descriptive vs transformation-led ran 0.5% vs 3% conversion — same traffic, 6x more leads — driven by argument sequence and promise specificity, not visual design.
**Source.** [@richardrx · 2026-04-17](https://x.com/richardrx/status/2045154631974539650)
## Users scan, they don't read — give the primary action obvious contrast
**Principle.** Users scan interfaces, weighing cognitive effort against payoff, and ignore most of it. They enter with one question — "What can I do here?" — so the primary action must win on contrast within seconds.
**Apply when.** The main CTA competes with 6 other elements, or the most important action is buried inside a dropdown that requires a click to reveal.
**The move.** Make the primary action visually dominant so it's found in a glance. If the key button competes with too many elements it loses prominence; if it's hidden behind interaction it's effectively invisible. Leverage image superiority to guide attention down the page.
**Visual.** A login screen annotated with two biases — "limited choice bias" (a single dominant Google sign-in button) and "image superiority" (a vivid hero illustration pulling the eye)
**Source.** [@richardrx · 2026-04-08](https://x.com/richardrx/status/2042013751742705816)
## Judge a landing page by conversion, not by beauty
**Principle.** A landing page isn't a beauty contest — it has one measurable job: qualify the user and lower CAC by turning traffic into revenue. Evaluating a static image with no context is nearly useless.
**Apply when.** The timeline turns into "AI vs human" / "who designed it better" debates that ignore conversion, while obvious conversion flaws go unaddressed.
**The move.** Treat conversion as a continuous loop, never an isolated event: form a thesis → ship the interface and collect data → find where the user hesitated, form a new hypothesis, and optimize the next cycle. The debate defaults to aesthetics because it's easier to opine on looks than to measure results — which is exactly what keeps most LPs underperforming.
**Visual.** A side-by-side "Google Stitch vs Human" LP comparison being judged on looks alone — context for the argument, not a model to copy
**Voice.** "Product design doesn't compete with art. It sells."
**Source.** [@richardrx · 2026-03-19](https://x.com/richardrx/status/2034638793219694734)
## Converting pages are often ugly — design for performance, not applause
**Principle.** There's an invisible war between branding (looking good in the screenshot) and performance (conversion, retention, expansion). The page that puts money in the till is often visually aggressive, text-heavy, and far from the Apple aesthetic founders dream of.
**Apply when.** You're choosing between a page that wins design compliments and one built to convert.
**The move.** Engineer the converting pattern even if the artist in you cringes: (1) one obvious attention point — eyes go where they must; (2) a high-contrast CTA with redundancy; (3) a clear promise delivering easily-perceived value; (4) guarantee, social proof, and a free element to reduce risk aversion; (5) organic images by the second fold to back the offer and build trust.
**Evidence.** The analyzed page (quoted): 100 visits → 19 signups, a 19% conversion rate.
**Visual.** An "ugly-but-converts" waterproofing LP: urgency bar, high-contrast orange CTAs, a short above-the-fold form, stat row (12,500+ / 4.9★ / lifetime warranty), and trust badges
**Voice.** "Some LPs go after compliments, others go after conversion."
**Source.** [@richardrx · 2026-02-17](https://x.com/richardrx/status/2023777916673220817)
## Design the features page for skimming, not reading
**Principle.** Almost nobody reads your features page — they skim three bullets and hunt for a demo video. Design for that behavior (with the caveat that the more conscious, technical slice of your ICP *will* read the detail).
**Apply when.** You're packing a features page with dense prose expecting visitors to read it top to bottom.
**The move.** Front-load three crisp bullets and make a demo video easy to find; let the deep copy serve the minority of technical, high-awareness buyers who actually read it.
**Source.** [@richardrx · 2026-02-09](https://x.com/richardrx/status/2020815036168437814)
## No clicks out, plus brutal CTA contrast
**Principle.** For conversion you want little distraction and lots of redundancy. Any clickable element that leads off-page is lost conversion — no matter how prestigious the source.
**Apply when.** You're tempted to link out to a glowing NY Times piece or a top influencer's video, or your CTA blends into the interface.
**The move.** Kill outbound clicks entirely. Make the CTA generate strong contrast against the rest of the interface in position, size, and color — if it isn't easy to notice and click, it won't be clicked.
**Visual.** A wireframe showing the level of contrast a CTA button needs against the surrounding interface
**Source.** [@richardrx · 2026-02-05](https://x.com/richardrx/status/2019514730696565238)
## Debug ICP and awareness level before touching design
**Principle.** Conversion is context engineering — who arrives, with what pain, at what stage, needing what proof. Two invisible variables must be debugged before writing a line of code: the Who (ICP) and the When (awareness level). Visual design comes last.
**Apply when.** Your LP underperforms and you reach for button-color tweaks; or you define ICP by demographics.
**The move.** (1) Reject the demographic fallacy: geography/age/income is a bad proxy — King Charles and Ozzy Osbourne are demographic twins with opposite needs. Real ICP is *buying criteria* (trigger, pain, prior attempt, proof needed); persona is just biography, the spec for your user. (2) Map the visitor's Eugene Schwartz awareness level. Rule: headline speaks to the current stage; proof pushes them one stage forward. Selling "Solution" to the "Unaware" is proposing on a first date. Debug order: ICP wrong → awareness wrong → proof insufficient for the stage → only then touch visual design.
Awareness → angle / hero / proof:
- **Unaware** — symptom & identity / "Still doing X this way?" / simple diagnostic, checklist, benchmark
- **Problem-aware** — cost & urgency / "If you have X, you're losing Y" / numbers, before/after, calculation
- **Solution-aware** — trade-offs & selection / "3 ways to solve X — why the 3rd scales" / honest comparison, matrix
- **Product-aware** — differentiation & proof stack / "Why us, why now" / cases, demo, objections
- **Most aware** — final risk & friction / "Swap X for Y in Z days, no risk" / guarantees, onboarding, effort reduction
**Visual.** Schwartz's 5 awareness stages as a rising ramp (Unaware → Problem → Solution → Product → Most Aware); plus the demographic-twins diagram (Charles vs Ozzy, identical on paper)
**Voice.** "There's no 'conversion rate' in a vacuum. There's contextualized conversion."
**Source.** [@richardrx · 2026-01-14](https://x.com/richardrx/status/2011427153133351046)

31
.github/skills/revenue-centric-design/references/metrics-and-experimentation.md vendored Normal file
View File

@@ -0,0 +1,31 @@
# Metrics, Experimentation & Business Math
> Curated, distilled wisdom from @richardrx ("Richard — Design for startups"), translated from Portuguese. Each entry is a reusable principle linked to its source post.
## Don't mistake signups for traction
**Principle.** Signups are the cheapest action a user takes, so they measure curiosity, not value — especially in freemium. Real growth is whether people come back and do the action that delivers value.
**Apply when.** A rising signup curve on the dashboard feels like proof of traction, particularly under a freemium model.
**The move.** Treat signups as top-of-funnel only — never stop reading there. Track who returns on day 2 and day 7 (D1/D7 retention) and how many complete the value-delivering action (activation). Paid media can inflate signups while real retention stays flat.
**Voice.** "The signup curve climbs with paid traffic, but usage and activation only climb with a good product."
**Source.** [@richardrx · 2026-06-11](https://x.com/richardrx/status/2065082771987394651)
## Don't bet your product on an underpowered A/B test
**Principle.** Most A/B tests in small SaaS lack the volume to prove anything, yet founders swap the whole product on the result. Testing without enough sample to conclude is the trap.
**Apply when.** You're in traction or survival stage, ran a test for a week, saw "variant B won by 12%," and want to ship it everywhere.
**The move.** Before running, compute the minimum sample size (free calculators exist); if you can't hit that floor in reasonable time, don't start. Test big things (headline, offer, pricing structure, onboarding) since large effects need less sample. Never stop a test because the number looked pretty mid-way. With no volume, decide by qualitative research — five good interviews beat an underpowered A/B test. Beware the law of small numbers and confirmation bias.
**Evidence.** ProfitWell is categorical: don't A/B test price — you'll never have the volume or context for it to mean anything.
**Source.** [@richardrx · 2026-06-01](https://x.com/richardrx/status/2061463480868229189)
## Celebrate signal quality, not list size
**Principle.** A waitlist exists to validate that a pain is one people pay to solve — not to sell. Absolute size is a vanity metric; conversion-weighted quality is the real signal.
**Apply when.** You launch a waitlist and feel tempted to celebrate raw headcount.
**The move.** Convert size to expected customers before reacting: a good waitlist converts 15–20% to paying, above 30% is excellent. 53 people at 20% = 10 customers; 1,000 people at 1% = 10 customers — same result, different perception. Until there's a transaction, there's no validated hypothesis.
**Voice.** "Founders celebrate the size of the list when they should celebrate the quality of the signal."
**Source.** [@richardrx · 2026-04-17](https://x.com/richardrx/status/2045094511106220220)
## Translate churn points into LTV, not percentages
**Principle.** Most people watch churn %, but few compute what each point costs in accumulated LTV over 12 months. Cutting churn is a cash lever that needs no price hike or new acquisition.
**Apply when.** You're staring at a churn percentage and treating it as a vanity number rather than money.
**The move.** Do the churn→LTV math: at 25% monthly churn on 1,000 users you must add 250 new users/month just to break even — kill paid traffic and the product dies in ~4 months. Then improve retention without Figma: define the Aha Moment, measure time-to-value (TTV) from signup to it, ask "how do I deliver this faster?", break it into micro-wins if you can't, then test, measure, repeat.
**Evidence.** Finance SaaS, ARPU R$120: cutting churn 5 points (25%→20%) is +R$72,000/year in cash, with no price change and no extra acquisition.
**Source.** [@richardrx · 2026-04-06](https://x.com/richardrx/status/2041184077106004289)

148
.github/skills/revenue-centric-design/references/onboarding-and-activation.md vendored Normal file
View File

@@ -0,0 +1,148 @@
# Onboarding & Activation
> Curated, distilled wisdom from @richardrx ("Richard — Design for startups"), translated from Portuguese. Each entry is a reusable principle linked to its source post.
## An onboarding video welcomes — it doesn't teach
**Principle.** A good onboarding video isn't a manual (nobody reads their car's or iPhone's). It welcomes, builds connection, shows the product at a glance, and points to where value comes fastest.
**Apply when.** Designing first-run onboarding or a welcome video.
**The move.** Aim it at cutting TTV, support tickets, and the lost feeling — not at educating. Length follows your ICP's urgency (someone rushing vs. someone happy to build Lego). No actor or fancy set — Richard recorded his in Screen Studio and it beat many big products'.
**Visual.** A "Your account was created!" welcome modal with an embedded intro-video thumbnail and a single "Next" CTA
**Voice.** "Your product doesn't need a manual either — have you read your car's?"
**Source.** [@richardrx · 2026-06-19](https://x.com/richardrx/status/2067987722954735812)
## Rising MRR with rising churn means you lost them on day one
**Principle.** When MRR and churn climb together, the user didn't leave in month 2 — they were lost the first day, dropped into a dead empty-state dashboard with nothing guiding them to value.
**Apply when.** Churn is creeping up and you're tempted to blame the product or add features.
**The move.** It isn't a feature gap — measure **TTV** (time-to-value) and get obsessed with shrinking it. CAC, LTV and activation are *product* metrics, not marketing; with weak retention, acquiring more just fills a leaky bucket faster.
**Voice.** "It took two months to cancel, but you lost him the first day after signup."
**Source.** [@richardrx · 2026-06-18](https://x.com/richardrx/status/2067591574138052804)
## Measure activation, not signups
**Principle.** Technical founders track the wrong onboarding metrics; signups, session time, and tour completion all flatter you without proving the user reached value.
**Apply when.** You're judging onboarding by signups, time-in-product, or "completed the tour."
**The move.** Swap each vanity metric for its real counterpart: signups → activation rate, session time → time-to-first-useful-action (and its repetition), onboarding completion → D7 retention. Find your aha moment empirically: look at what every paying customer did in week one that churned users didn't (often a collaborative act — invite, share, comment). Anchor on TTV/time-to-value.
**Evidence.** Userpilot benchmark (547 companies): avg TTV 1d 12h 23m; top performers under 5 min. SaaS activation rate avg 30–37%, top quartile 40%+, under 20% = structural problem. D7 retention avg 10–15%, over 30% is strong.
**Voice.** "If you can't say how long your user takes from signup to aha moment, you're not measuring what matters."
**Source.** [@richardrx · 2026-05-27](https://x.com/richardrx/status/2059616501544468624)
## Put friction in the right place, not zero friction everywhere
**Principle.** Friction in the wrong place kills the product; friction in the right place qualifies and retains. "Less friction" is not a universal law.
**Apply when.** You're reflexively cutting clicks and fields, or your human sales team is doing qualification the product should do.
**The move.** Remove friction at trial signup (it kills acquisition), but add calibrated friction in three spots: (1) trial with card upfront filters commercial intent; (2) mandatory onboarding before the dashboard turns users into power-users faster; (3) a 6–8 field enterprise demo form (role, team size, current tool, budget, timeline) lowers lead volume but raises close rate. The mechanism is effort justification (Aronson & Mills, 1959) — same root as the endowment and IKEA effects.
**Evidence.** ChartMogul 2026: opt-in trial (no card) converts 8.9%; opt-out (with card) converts 31.4%. Superhuman requires a 30-min human call before access.
**Voice.** "How much qualification effort is your human seller doing that the product should do before they even step in?"
**Source.** [@richardrx · 2026-05-21](https://x.com/richardrx/status/2057436163841941980)
## Never ship a blank dashboard
**Principle.** The empty dashboard arrives at the user's peak of curiosity and answers it with a void — this is where most SaaS loses the trial. Every second spent deciding what to do is a second closer to quitting.
**Apply when.** A new user lands post-signup on a screen with no data and no direction.
**The move.** Four fixes: (1) empty state with a next-action hint — the CTA points straight to value; (2) seed sample data so they see the destination before starting; (3) one single clear action ("Import your first spreadsheet"), not eight, not a 12-step tour; (4) visible progress from the first click — start the bar at 20%, not 0%, so completion feels already underway.
**Voice.** "A blank dashboard looks neutral, even tidy — but it just makes the user stop and think about what to do."
**Source.** [@richardrx · 2026-05-12](https://x.com/richardrx/status/2054283657934758021)
## Design onboarding as a behavioral trigger, not a feature tour
**Principle.** Silent non-activation — users who sign up, vanish in five minutes, and never formally churn — is an activation problem, not a product one. Onboarding should fire a behavior, not narrate features.
**Apply when.** New users evaporate without complaint and you never learn their name.
**The move.** Find the single behavior that statistically separates retained from lost users, make it your activation north star, and measure every onboarding decision against it. Shift focus from explaining features to forcing that behavior fast. Exploit the Zeigarnik effect: open small loops (complete profile, invite 3 colleagues, send first message) so the user carries an unfinished task.
**Evidence.** Slack: teams exchanging 2,000 messages had 93% probability of staying. Facebook's equivalent: 7 friends in 10 days — the company's single focus, repeated at every all-hands.
**Voice.** "What's the number that separates who stays from who evaporates — and how does the user hit it in under 24h?"
**Source.** [@richardrx · 2026-05-11](https://x.com/richardrx/status/2053878928494690414)
## Map the journey from session replays, not from your diagram
**Principle.** The founder's 12-step journey is built top-down (what the product wants); the user runs 4 steps bottom-up (the specific problem they opened the tab to solve). The gap between them is where avoidable early-stage churn lives — and it's invisible because the founder only ever lived the creator's journey.
**Apply when.** You "know" the happy path but can't state what % of users actually execute it.
**The move.** Three steps, no Figma: (1) write your version of the user journey in numbered steps, on paper; (2) open five real session recordings from the first 7 days and note what each user actually does, in order, with timing — including what they try and abandon; (3) lay both lists side by side. The report is in the differences; each divergence is a hypothesis to confirm or kill.
**Voice.** "The user journey is what shows up in the replay; what's in Figma and Excalidraw is a hypothesis."
**Source.** [@richardrx · 2026-05-08](https://x.com/richardrx/status/2052711138572263474)
## Add declarative friction, cut administrative friction
**Principle.** "Good onboarding is short onboarding" is incomplete. There are two frictions: administrative (collects data the system uses later, buys the user nothing) and declarative (forces the user to state what they came to do — costs a beat, buys commitment, customization, and journey direction).
**Apply when.** Auditing onboarding steps; deciding what to cut versus expand.
**The move.** Test each step: is it collecting data or making the user declare intent? Collecting only → candidate to cut. Declaring intent → candidate to expand. A declaration ("what do you sell, what's your long-term goal?") creates a micro-commitment to the outcome before the user touches the product, and lets the journey branch (recommendations, tutorials, next actions) off that answer.
**Evidence.** Brazilian payments platform cut time-from-signup-to-first-sale from 24.2 to 2.5 days by adding a ~30-second intent step (positioned between signup and product), not by removing steps.
**Voice.** "Onboarding is also the first chance the user has to declare to themselves what they came to do."
**Source.** [@richardrx · 2026-05-07](https://x.com/richardrx/status/2052350703541039324)
## Trial conversion is a journey problem, not a pricing problem
**Principle.** A trial is a test of value; if the user never proves value to themselves, no price or trial length saves it. Conversion fails for three diagnosable reasons.
**Apply when.** Users sign up, do "a bunch of nothing," and never return.
**The move.** Diagnose which failure mode applies: (1) blank dashboard → make the first step obvious and immediate, drive to value or a micro-win, drop "explore our product"; (2) lost before value → install Clarity, watch where they stall, optimize that click; (3) lost to life → send a progress email ("you created 3 reports, your team accessed 12 times, you're in the top 20%"), not a generic "trial ending."
**Voice.** "The longest trial in the world doesn't save bad onboarding."
**Source.** [@richardrx · 2026-05-06](https://x.com/richardrx/status/2052096365128273956)
## Make onboarding active, not passive
**Principle.** Passive onboarding (tooltips, guided tour, docs — learn if you want) assumes the user will explore. They won't: they have 47 tabs open, WhatsApp pinging, and will do the bare minimum before deciding whether to return. Active onboarding designs a sequence where each action delivers value and that value triggers the next.
**Apply when.** Your onboarding opens with "Welcome, here's the documentation."
**The move.** Assume you know the shortest path to value better than the user does. Design that path, strip the friction, and make sure they arrive. Don't ask "what's the minimum the user must do?" — ask "what's the smallest action that delivers the most value in the least time?"
**Evidence.** Slack drops you into a channel and makes you send a message first — you use the product before any tutorial.
**Voice.** "Understanding by doing beats understanding by reading."
**Source.** [@richardrx · 2026-04-22](https://x.com/richardrx/status/2046959126245249288)
## Treat sub-30-day churn as an onboarding fix, not a feature gap
**Principle.** Most early-stage SaaS churn happens in the first 30 days — which means it's onboarding, not product. Adding features only makes it worse by adding complexity.
**Apply when.** Users leave before they ever liked what you built, and you're tempted to ship more features.
**The move.** Diagnose with three questions: (1) how long to the first real result? If "it depends" or over a day, you're bleeding users; (2) does the user know where they are? Use a progress bar/checklist — Progress effect: someone seeing 30% done is likelier to finish than someone at 0%, so starting at 0% is a design error; (3) what happens when they drop mid-flow — email, push, nothing? Use the Zeigarnik effect to remind them they started.
**Voice.** "Sub-30-day churn rarely dies to features; it dies to the right sequence of micro-interactions that deliver value before asking for effort."
**Source.** [@richardrx · 2026-04-20](https://x.com/richardrx/status/2046212675017887881)
## Deliver the promised result before teaching mechanics
**Principle.** Nobody wants to learn to use your product; they want the result you promised in the landing-page hero. Teaching mechanics first ("create a project → add a member → configure integrations") is boring; delivering value first converts.
**Apply when.** Your onboarding is a checklist of setup mechanics rather than a path to the outcome.
**The move.** Lead with the outcome ("In 2 minutes you'll see your first report → let's start with the data you already have → done, that's the insight competitors pay consultants for"). Reframe progress with the Progress effect by crediting effort already spent ("You've done the hard part, just 3 steps left"). Keep loops small (Zeigarnik effect — people close loops only if they look closable). Pre-select the right plan from data you collected instead of asking, then offer the upsell.
**Voice.** "The gap between 5% and 15% trial conversion is in these details — not features, not price — in the sequence of micro-decisions you designed without realizing you were designing."
**Source.** [@richardrx · 2026-04-16](https://x.com/richardrx/status/2044785090832543998)
## Qualify by behavior, not by a long signup form
**Principle.** Friction at the wrong moment kills conversion, and qualification by behavior is more precise than qualification by form. The "more qualified leads" argument for long forms usually loses.
**Apply when.** You're weighing an 8-field signup form against email + password.
**The move.** Default to the minimal form and let qualification happen later, inside the product, from real behavior. Deciding where to add versus remove friction is what separates a product that grows from one that spins its wheels — but it's contextual ("it depends").
**Evidence.** Same product, two founders: 8-field form (name, email, company, role, phone, segment, team size, how-did-you-hear) → 12% signup rate; email + password only → 34%.
**Visual.** Annotated onboarding step (DevNoodles): an ICP/intent question flagged "Zeigarnik effect" (the step progress dots) and a B2C/B2B card selector flagged "Progress effect" — showing where each bias is engineered into the flow.
**Source.** [@richardrx · 2026-04-10](https://x.com/richardrx/status/2042558822825239030)
## Celebrate the activation moment, don't just confirm it
**Principle.** At the emotional peak of activation, a number is data but a rising graph is progress — and progress triggers dopamine and an emotional memory tied to the product. Most SaaS confirms where it should celebrate.
**Apply when.** A user completes a hard-won first action (first transaction, first integration) and you respond with a static success state.
**The move.** Engineer the peak moment precisely at the point of highest emotional vulnerability in activation — right after the user clears the effort. Show motion and accomplishment proportional to the effort invested. This is the peak-end rule: users judge an experience by its emotional peak and its ending, rarely by the average.
**Evidence.** Stripe shows a rising graph (not a number) the moment the first transaction processes — the founder who integrated it at 3am remembers exactly where they were.
**Voice.** "A number would have done the job. The graph created a customer."
**Source.** [@richardrx · 2026-04-04](https://x.com/richardrx/status/2040415841628651689)
## Cut time-to-value by reorder and removal, not feature changes
**Principle.** High TTV is almost never product complexity — it's the form and order in which things happen. Each day between signup and first result is another day of abandonment risk; abandonment = churn.
**Apply when.** Your activation flow is slow and you assume the product itself is the bottleneck.
**The move.** Three iteration cycles, no product changes: (1) reduce cognitive load by grouping and standardizing what the user must fill in; (2) work with legal/compliance to strip everything required by habit but not by actual necessity; (3) invert the sequence so the user feels value before facing the heaviest step.
**Evidence.** Major Brazilian payments platform: 24 days → 2.5 days to first sale (89.7% reduction), no product changes, no cutting of mandatory compliance steps.
**Visual.** Month-by-month TTV table: Jan 24.2d → Feb 19.6d → Mar 16.8d → Apr 9.3d → May 2.5d, alongside accounts created / approved / new sellers per month.
**Source.** [@richardrx · 2026-03-27](https://x.com/richardrx/status/2037583944283996418)
## Give the trial an active goal, not passive access
**Principle.** A passive trial ("use it if you want, cancel guilt-free") builds no commitment; a trial with an active goal builds commitment before billing. Each completed day raises the psychological cost of canceling — the user starts defending a decision they already made, before paying.
**Apply when.** Your trial is open-ended access with no challenge or target.
**The move.** Set a recurring daily goal during the trial that the user opts into and completes. The named mechanism is progressive commitment. The product can be good or bad — done right, the onboarding itself is excellent.
**Evidence.** Wispr Flow challenges trial users to dictate 100+ words a day for 7 days — looks generous, is behavioral science.
**Source.** [@richardrx · 2026-03-26](https://x.com/richardrx/status/2037316988981174464)
## Calibrate the first step to the user's real willingness
**Principle.** Users don't rationally evaluate the first task — they evaluate perceived effort. Ask too much up front and the cognitive cost of even imagining the task is paralyzing, and they quit before starting. This is the activation-barrier effect.
**Apply when.** Your first onboarding step bundles setup, data import, team invites, and project creation — or demands a campaign-sized action.
**The move.** Two biases fix it: (1) started-progress effect — show what the user has already done before what's left; a loyalty card with the first stamp pre-filled beats a blank 9-stamp card, because the starting point changes perceived distance to the finish; (2) small-steps effect — "Create one story today" has radically lower perceived cost than "post 5×/week," even when the underlying task is identical. Sequence effort so the user hits first value before noticing how much they invested; build momentum.
**Evidence.** Instagram A/B test made "Create 5 new public reels" the first task — for someone who barely posts weekly, that's paralyzing.
**Visual.** Bad first-step example: a weekly-progress checklist at "0% completed" whose top item is "Create 5 new public reels (0/5)."
**Voice.** "Asking for more isn't necessarily the problem; asking for all of it at once, with no progress anchor and no commitment ladder, is."
**Source.** [@richardrx · 2026-03-24](https://x.com/richardrx/status/2036461688505909250)
## Treat onboarding as the bridge between CAC and LTV
**Principle.** Founders obsess over CAC and landing-page conversion but are blind to activation cost. Onboarding isn't an interface tutorial — it's the bridge from CAC to LTV and the point of maximum leverage to expand revenue. Cancellation happens on day one; it's merely formalized when Stripe's billing reminder lands.
**Apply when.** Users enter the trial without intent to a result and ghost before ever paying.
**The move.** Two fixes: (1) turn support into UX — every onboarding support ticket is a design failure; map recurring setup questions and convert the answers into features or in-flow tooltips; (2) compress TTV obsessively — make the user experience the product's core promise in the least time possible. If they must configure 5 screens before any result, you've already lost.
**Voice.** "Letting a user into the trial without intent of a result isn't self-service. Design's job doesn't end at signup — that's where it starts paying you back."
**Source.** [@richardrx · 2026-03-13](https://x.com/richardrx/status/2032485654811083005)
## Pick onboarding patterns by awareness × flow complexity
**Principle.** There's no best onboarding in the abstract — only the one that removes friction and delivers value early for your users. Treat the nine patterns as behavior-shaping mechanisms, not UI components, and select by two axes.
**Apply when.** Choosing or combining onboarding patterns for a new flow (SaaS, PLG, B2B early stage).
**The move.** The nine patterns (pattern → ideal use / tradeoff): **1. Welcome modal** → high-awareness ICP or low-complexity products; easy to build, easy to ignore. **2. Wizard / product tour** → B2B and complex/high-cost-of-error flows (fintech, compliance); long tours cause boredom and need constant upkeep. **3. Contextual tooltips** → advanced/secondary features; slashes support tickets but users may miss them if contrast is poor. **4. Empty state** (his favorite) → dashboards, lists, data-dependent areas; if executed well it's mandatory, directs the first action and accelerates TTV. **5. Personalization** → products serving many ICPs with different journeys; great CRM data, but too long kills signup conversion. **6. Checklists** → critical flows with mandatory prerequisites (webhook, KYC); exploits the Zeigarnik effect, but a long list breeds aversion — every item must move toward value, no bureaucratic tasks. **7. Goal-setting** → habit products (finance, productivity, health); uses commitment bias, but a broken goal can break the emotional contract. **8. Sample data** → sell the dream of a full, organized product before the user inputs anything (distinct from skeleton screens). **9. Use cases / demos** → products burning expensive resources (AI tokens) with infinite outputs; show max potential without forcing creativity from scratch.
Selection rule — two axes: **awareness level** (low = needs guidance; high = needs speed) × **flow complexity** (high = needs structure). Four cases: low-awareness + high-complexity → tours + checklists + personalization; low + low → modal + empty state; high + high → checklist + tooltip + empty state; high + low → modal + empty state (then get out of the user's way). Build vs. buy is financial/operational, not aesthetic — buy (Wistia, PostHog, Sprig) when speed is critical, dev is overloaded, or you're running A/B tests; build when onboarding is strategic, you need perfect aesthetic integration, or you must avoid third-party dependence.
**Voice.** "Onboarding's job isn't to teach the user — nobody likes an instruction manual — it's to remove cognitive effort and deliver utility as fast as possible. TTV correlates directly with churn."
**Source.** [@richardrx · 2026-02-04](https://x.com/richardrx/status/2019019293761941566)

64
.github/skills/revenue-centric-design/references/positioning-icp-and-gtm.md vendored Normal file
View File

@@ -0,0 +1,64 @@
# Positioning, ICP & Go-to-Market
> Curated, distilled wisdom from @richardrx ("Richard — Design for startups"), translated from Portuguese. Each entry is a reusable principle linked to its source post.
## Size the market with TAM → SAM → SOM (and know which one matters)
**Principle.** Your real market is far smaller than the population — TAM is a theoretical ceiling (Brazil's 213M becomes ~101M credit-card holders for a paid app). Investors often ignore a TAM under ~R$1B, but the number that matters is the **SOM** — what you can actually capture.
**Apply when.** Sizing a market, writing a deck, or judging whether a niche is big enough.
**The move.** TAM = total addressable ceiling; **SAM** = the realistic slice your model reaches (~40% in his example); **SOM** = the 1–5% you truly win in ~36 months — and SOM isn't a guessed %, it comes from real CAC, activation, support capacity and LTV. You can't change your TAM; you change how much of your SAM you convert and retain.
**Evidence.** RepareCar: ~76k mechanic shops (honest TAM) → SAM ~47k → ~3% ≈ 1,414 shops ≈ R$1.6M ARR; current pace (~10 shops/day) ≈ 7% of SAM in 12 months.
**Visual.** TAM/SAM/SOM concentric-circle diagram with definitions
**Source.** [@richardrx · 2026-06-25](https://x.com/richardrx/status/2070140923380420796)
## Frame the referral prize as a gift to the friend, not a commission to the referrer
**Principle.** Member-get-member (MGM) referral programs win on framing and timing, not just on a two-sided reward. Money makes the exchange feel transactional; an in-product benefit feels like a genuine gift.
**Apply when.** Designing or fixing a referral program and defaulting to "refer a friend, get $20."
**The move.** Apply the framing effect: surface the prize on the receiver's side ("João gave you 500MB"). Ask for the referral at the peak of value (right after a concrete win, or when the user hits a limit). Avoid cash; give a reward that deepens use of your own product. Embed it as continuous in-product operation, not a one-off campaign. Caveat: referral amplifies a product people already love; it can't fix one nobody recommends for free.
**Evidence.** Dropbox grew 3900% in 15 months (100k → 4M users), peaking near 3M invites in a single month; ~1/3 of users already arrived via word-of-mouth before the program.
**Voice.** "A referral amplifies a product people already love — it doesn't fix a product nobody recommends for free."
**Source.** [@richardrx · 2026-06-02](https://x.com/richardrx/status/2061766945582559509)
## Pick a deliberately under-served niche as your ICP
**Principle.** A clear ICP (ideal customer profile) is not "everyone who could use my product." It's a deliberately chosen, under-served niche — and a sharp niche beats no niche, because you can't out-fight the entrenched generalist giant.
**Apply when.** Early traction; tempted to "embrace the world" out of fear of a small TAM.
**The move.** Validate four ICP filters: (1) feels the pain with real weight — pain is proportional to what's lost when unsolved (a lost lead costs a face-aesthetics clinic R$3,000 vs. R$60 for a barber); (2) big enough TAM to sustain operations; (3) money to pay your required ticket so unit economics close; (4) founder-fit, giving native language, a fast validation network, and instinct that money can't buy. With a clear ICP, failure has a diagnosis ("I got the messaging wrong"); without one, you can't tell if product, copy, channel, price, or audience failed — and every test burns runway.
**Voice.** "A generalist ERP is hard to sell; an ERP for cabinetmaking is a different conversation."
**Source.** [@richardrx · 2026-05-19](https://x.com/richardrx/status/2056789797646029232)
## Don't claim PLG without the four structural conditions
**Principle.** Product-led growth (PLG) is a consequence of structural conditions, not a product decision you declare. Most B2B SaaS that pitches PLG is really sales-led wearing a PLG label.
**Apply when.** Writing a pitch deck or strategy and calling the motion "self-service" / PLG.
**The move.** Require all four conditions: (1) TTV < 10 minutes — if it needs a consultant demo, API support, or paid implementation, it's not PLG (tell: full trial, zero activation); (2) ticket below ~R$1,000 — higher means a buying committee; (3) native virality or collaboration (Notion, Figma, Slack pull users in; a CRM/AI tool needs SDRs, demos, follow-up); (4) a huge addressable market with a real bottom-up TAM. If you fail these, run sales-led honestly.
**Evidence.** Brazil has ~20,000 companies with 100+ employees, and only ~a dozen B2B SaaS where PLG makes real economic sense.
**Voice.** "Founders love PLG because it seems to delete the part they don't master — selling."
**Source.** [@richardrx · 2026-05-04](https://x.com/richardrx/status/2051262752547536941)
## Charge your first ten users from day one
**Principle.** The first ten users define the product's entire curve, and payment is the cheapest test of real pain — curiosity is free, an open wallet demands a concrete problem.
**Apply when.** Validating a new product and tempted to give early access away to "build a base."
**The move.** Source the first ten from closed communities, personal reach, or pure guerrilla. Charge even while in prototype. When someone says they can't pay, ask directly: "What does the system need to do for you to pay right now?" Treat the payment friction as part of the test. Collect dense feedback; only start visual design after ~20 paying users. For B2C apps the method shifts (e.g., pre-sale of a solution-in-progress) but the principle holds.
**Evidence.** RepareCar's first 25 auto shops tested the product in prototype; the team visited each and charged at the end, designing visuals only after 20 paying shops.
**Voice.** "Curiosity is free; an open wallet demands a concrete problem."
**Source.** [@richardrx · 2026-04-26](https://x.com/richardrx/status/2048359526487716333)
## Reverse-engineer the funnel math before celebrating an MRR target
**Principle.** Building the product is the easy part; distribution is the game. A revenue target is really a traffic-and-retention problem, and churn quietly resets the whole funnel every month.
**Apply when.** Someone asks "is it hard to hit X MRR?" or you're sizing acquisition for a target customer count.
**The move.** Work backwards: to net 2,500 customers at 5% LP conversion you need 50,000 visitors; from ads at 3% creative CTR, ~1.6M impressions (5% and 3% are top-decile — most land at 1–2% LP and under 1% CTR, so you test dozens). Then add the leaky bucket: at 20% monthly churn, average customer life is 5 months, so you replace 500 customers every month forever just to stand still (≈10,000 visitors / 333,000 impressions). The problem lives at the intersection of dev, design, and marketing — none alone owns it.
**Evidence.** 20% monthly churn → 5-month average lifetime; at low ticket many operate at 40–50% churn, so "the bucket never fills."
**Voice.** "Building the product is the easy part; distribution is the game."
**Source.** [@richardrx · 2026-04-20](https://x.com/richardrx/status/2046222319912132977)
## Concentrate channels with the Bullseye framework, not scattershot testing
**Principle.** Testing ten channels at once means you never know what drove results and you blame the channel when the business stalls. Distribution needs prioritized focus — and a perfect channel still fails if the receiving structure leaks.
**Apply when.** You're spreading content and traffic across many channels with no clear read on what works.
**The move.** Use the Bullseye framework (from the book *Traction*): three rings of priority. Inner ring = at most three highest-potential channels with total focus; middle ring = up to six channels you probe with small experiments; outer ring = everything plausible long-term, no active focus now. Choose between channels with an ICE Score (Impact, Confidence, Ease, each 0–10, divide by 3, prioritize). Crucial gap the book skips: scaling distribution onto a broken reception structure (LP, onboarding, first product steps) yields no growth — distribution and retention are simultaneous, not sequential.
**Visual.** Bullseye as nested circles — What's Possible → What's Probable → What's Working — beside a "Marketing Framework for Startups" triangle (Prioritization, Testing, Quick Iteration).
**Source.** [@richardrx · 2026-03-23](https://x.com/richardrx/status/2036035304868434115)
## Diagnose the bottleneck: no entries is distribution, leaving without paying is design
**Principle.** Design can't save a "ghost product." Design optimizes and raises the LTV of something that already has traffic; it can't manufacture demand.
**Apply when.** A builder ships an app, gets near-zero users, and hopes a redesign will rescue it.
**The move.** Split the diagnosis cleanly: if nobody enters your product, it's a distribution problem; if they enter, don't pay, and leave, it's design. Read *Traction* even if you can afford an agency or a marketing team — the lever isn't just cost-per-channel but each channel's awareness level, which drives different conversion and retention behavior depending on where and how the user arrived.
**Voice.** "If nobody enters your product, it's distribution. If they enter, don't pay, and leave, that's design."
**Source.** [@richardrx · 2026-02-28](https://x.com/richardrx/status/2027721170569564521)

87
.github/skills/revenue-centric-design/references/pricing-and-monetization.md vendored Normal file
View File

@@ -0,0 +1,87 @@
# Pricing & Monetization Psychology
> Curated, distilled wisdom from @richardrx ("Richard — Design for startups"), translated from Portuguese. Each entry is a reusable principle linked to its source post.
## The freemium trap, in numbers: higher conversion, far lower cash
**Principle.** A free plan lifts signup conversion but can crush the economics — the higher top-of-funnel number hides a worse business.
**Apply when.** You're tempted by freemium's better conversion rate.
**The move.** Run the funnel (same R$80k/mo traffic, plans from R$199). **With freemium:** ~8% → 800 signups → 80% activate → 5% pay = 32 payers (R$6,368 MRR), saturating ~160 payers under 20% churn, while 768 free users burn AI tokens (~$0.08 each) — real CAC ≈ R$2,540/payer, payback ~13 months. **Without:** ~3% → 300 payers = R$59,700 MRR (~10×), CAC ~R$275, payback ~6 weeks, LTV:CAC 5:1 that reinvests its own profit. Freemium only pays off if free brings *organic/viral* users you didn't pay for.
**Voice.** "One scenario reinvests its own profit; the other funds losses until the money runs out."
**Source.** [@richardrx · 2026-07-01](https://x.com/richardrx/status/2072312844784152628)
## Whether freemium works is decided by the cost to serve a free user
**Principle.** Freemium isn't good or bad in the abstract — the *cost of free* decides, and it hinges on (1) how much it costs to serve non-payers and (2) how long/expensive activation is.
**Apply when.** Considering a free plan, especially as a bootstrapped (non-bigtech) founder.
**The move.** If serving a free user costs almost nothing and TTV is short, free becomes an acquisition channel (Slack — first message in minutes; it's the short TTV, not the cash, that sustains it). If the product runs on AI (dollar-priced tokens) or activation is long, a free account is an expensive bet that a small fraction funds — which needs deep pockets (the exception, not the average founder). On a friendly average, only ~3–4% of freemium converts. Otherwise: charge — well, and early.
**Voice.** "For an AI product, your free user was never free."
**Source.** [@richardrx · 2026-06-30](https://x.com/richardrx/status/2071962778072469560)
## Price is the cheapest money — stop anchoring it to the cheapest competitor
**Principle.** Pricing is a SaaS's biggest lever, yet ~90% of products are underpriced — the founder, who knows every limitation, anchors on the cheapest competitor instead of on value delivered. The buyer only sees the problem solved.
**Apply when.** Setting or revisiting price; fearing a "no."
**The move.** Raise toward value. A 30% price increase doesn't yield 30% MRR (some churn), but what remains is nearly pure cash — no acquisition in between — while growing a channel 30% costs money, time, and has a ceiling. Low price costs you later: less budget to reach your ICP, a CAC-obsession trap (the real metric is the CAC↔LTV *gap*, which price widens on both sides), and higher churn (cheap attracts uncommitted buyers). Design link: the number must be sustained by perceived value — your page and first use justify or destroy it.
**Voice.** "Charging more without seeming to be worth more is just raising the price of rejection."
**Source.** [@richardrx · 2026-06-29](https://x.com/richardrx/status/2071634185228329219)
## Make the middle plan the one you actually want to sell
**Principle.** Each plan has a behavioral job, not just a price; the plan you most want to sell should sit in the middle, flanked by a decoy below and an anchor above.
**Apply when.** Building or auditing a SaaS pricing page, especially if you copied competitors without assigning each tier a role.
**The move.** Use the decoy effect: place your target (e.g. Pro) in the middle; make the tier below it clearly inferior on one important attribute (user cap, no critical integration, no priority support) so Pro looks obvious. Keep exactly three plans — four+ triggers the paradox of choice and users stall. Add a top tier (Enterprise) purely to anchor price perception. Ask: "What is my decoy today?" If you can't name one, it likely doesn't exist.
**Evidence.** Ariely's MIT test of The Economist's tiers: with the print-only decoy, 16%/84% chose online/combo; removing it flipped choices to 68%/32%, cutting combo revenue by more than half. Estimated +30–43% subscription revenue.
**Visual.** Economist subscription page; the decoy's removal shifts combo-plan share from 84% down to 32%
**Voice.** "Option B was never built to be sold — it was built to make C look obvious. It's the bait."
**Source.** [@richardrx · 2026-05-28](https://x.com/richardrx/status/2059951433827426437)
## Ask for the card in trial — but optimize for the right ICP, not raw conversion
**Principle.** Requiring a credit card multiplies trial-to-paid conversion but shrinks signups; the goal is the model that attracts and retains the right ICP, not the one with the highest headline conversion.
**Apply when.** Choosing trial-with-card vs trial-without-card (or freemium), or designing recurring billing for a Brazilian market.
**The move.** Weigh the funnel both ways. Trial-with-card converts harder but starves you of volume; trial-without-card floods the funnel with low-intent users. Run the full math, not just the conversion rate. In Brazil, also account for PIX recorrente, whose dynamics differ from monthly card billing.
**Evidence.** ChartMogul 2026 (US, 200 products): trial-with-card converts ~31.4% vs 8.9% without — 3x+. Worked funnel: 1,000 visitors → 30 trials → 9.4 paying (with card) vs 85 trials → 7.5 paying (without). Author observes PIX-recorrente cohorts churn more than card cohorts.
**Voice.** "Don't ask which model converts more — ask which model attracts and retains the right ICP."
**Source.** [@richardrx · 2026-05-15](https://x.com/richardrx/status/2055247161349054950)
## Frame the upgrade as a loss at the moment of value, not a feature you're selling
**Principle.** Low upgrade rates are usually a framing-and-timing problem, not a price problem; remind users what they've already invested and what they stand to lose.
**Apply when.** A happy, active free user never upgrades, or your upgrade rate sits below 5%.
**The move.** Three framings beat generic limit/discount/feature-gate prompts. (1) Sunk cost: surface the assets they've built — "You created 47 custom reports. On the free plan you lose access to 40." (2) Loss aversion: framing loss outconverts framing gain — "You'll lose access to 8 months of history" beats "Get unlimited history." (3) Limited-access gate timed to an imminent, known result — "Your report is ready. To export as PDF, activate Pro." The timing/context of the gate matters more than the gate itself.
**Voice.** "If your upgrade rate is below 5%, the problem probably isn't price — it's how and when you're asking."
**Source.** [@richardrx · 2026-04-21](https://x.com/richardrx/status/2046544442216054981)
## Engineer the comparison frame with a decoy and a high anchor — and drop Free from the top
**Principle.** Conversion shifts when you change the frame of comparison, not the product; equal-looking options cause delay, and showing Free first anchors everyone to zero so everything else feels expensive.
**Apply when.** You run the default Free / Pro / Enterprise (sob consulta) ladder and Pro isn't converting.
**The move.** Insert a decoy: a Starter just below Pro with irritating limitations (e.g. R$79 vs Pro R$99) so users compare Starter↔Pro and Pro wins for R$20 more. Remove Free from the visible top so the first number isn't zero — anchoring means the first price seen sets the reference; lead with a higher/previous/Enterprise price so Pro at R$99 reads as cheap.
**Evidence.** The Economist sold 3x more print+digital after adding a same-price print-only decoy nobody bought. Author cites documented tests lifting conversion 10–20% via reframing alone.
**Voice.** "You're competing against your own free plan. And losing."
**Source.** [@richardrx · 2026-04-14](https://x.com/richardrx/status/2044014136770580743)
## Tie the trial's end to value consumed, not the calendar
**Principle.** Blocking access on a fixed day count (7/14/28) is a lazy rule; the billing trigger should fire on value consumption, after the user's first real win.
**Apply when.** You copied a competitor's 14-day trial and paid conversion is failing, or you're setting trial length from scratch.
**The move.** Never paywall before a clear micro-win or solving the core problem — doing so kills conversion and breeds bad word of mouth. Set length using four variables: (1) Product complexity — enterprise needs time for compliance/security review, not just the user. (2) Time to Value — Spotify delivers in seconds, a CRM needs days of data. (3) Usage frequency — rarely-used products may need long trials, or none at all (a once-a-year tax tool shouldn't have a trial). (4) Card entry — no card means a shorter trial to create urgency; with card, watch silent next-month churn. Note: sunk cost only bites if the user built a real asset — a bad onboarding produces frustration, not switching cost.
**Voice.** "Locking access purely on the calendar is a lazy rule that can cost you dearly — you're burning CAC without knowing where value lands."
**Source.** [@richardrx · 2026-03-16](https://x.com/richardrx/status/2033502548301091057)
## Order pricing rows by the serial-position effect: killer feature first, differentiator last
**Principle.** Users don't read pricing lists linearly; attention and memory cluster on the first and last items, so feature order is itself a conversion lever.
**Apply when.** Laying out the feature rows inside a pricing card or comparison table.
**The move.** Exploit the serial-position effect (primacy + recency). Top: value anchor — never "24/7 support"; lead with the core/killer feature that solves the ICP's main pain and justifies ~80% of the ticket and the ROI. Middle: utilitarian features (exports, integrations, storage limits) the user won't memorize but will scan to compare against the next plan. Bottom (nearest the CTA): the differentiator, bonus, or loss-aversion hook — a lifetime guarantee or dedicated support. The middle of the list is "a cognitive black hole."
**Visual.** Pricing card emphasizing the bold first row (core feature) and bold last row (super bonus), with greyed utilitarian middle rows
**Voice.** "Pricing success depends not just on what you deliver, but on the order the brain is led to process the value."
**Source.** [@richardrx · 2026-03-05](https://x.com/richardrx/status/2029623167900061970)
## Build a single value axis, then tune the decoy's distance to your target plan
**Principle.** A plan ladder must read as one clear progression of value; mixing quantitative and qualitative axes muddles it, and where you place the decoy's price decides which plan looks like the deal.
**Apply when.** Naming and pricing tiers, or the "value staircase" between your plans isn't obvious to users.
**The move.** Pick one progression — quantitative (rising credits/users) or qualitative (24/7 human support, special features) — rather than blending both. Borrow Starbucks-style naming (Tall/Grande/Venti) so every tier sounds good and lifts the brand. Then position the decoy: place it near the most expensive plan and the expensive plan looks cheap; place it near the cheapest and the decoy itself becomes the most attractive option.
**Visual.** Decorative 3D price-tag illustration — no data.
**Source.** [@richardrx · 2026-02-02](https://x.com/richardrx/status/2018357024543715480)
## Engineer the pricing page with Good-Better-Best and control the comparison
**Principle.** Lost LTV is rarely about price — it's analysis paralysis from a missing choice architecture. The brain is lazy and judges by relative comparison (priming + anchoring), so if you don't design the anchor, users compare you to "nothing" or to the cheapest competitor.
**Apply when.** Designing or fixing a pricing page; conversions die at the final step despite strong CAC spend.
**The move.** Use a Good-Better-Best (GBB) structure: **Good** = a stripped entry plan that anchors a low price but is limited enough to make users feel pain and look up (never make it free — then everything above looks expensive). **Better** = your standard plan, the target for ~80% of buyers; price it closer to Good than to Best so users think "paying only ~20% more I get double?" **Best** = the value anchor that exists mainly to make Better look cheap (bicycle analogy: without the carbon-fiber Best, the carbon-wheel Better looks expensive). Golden rule: keep comparisons on one axis — don't pit "10,000 tokens" against "Priority Support"; prefer linear, ideally asymmetric, growth. Cap at 2–5 plans (6 = anxiety, paradox of choice). Then control which attributes you compare — your own "Brazil vs Paraguay" table — choosing indicators that favor your value thesis. Highlight Better with color/size/badges. "Stop making the user do the math — do the math for them."
**Evidence.** Cites Briesch et al. (1997) and Mazumdar et al. (2005) on reference-price models, and Chernev (2015) on choice overload.
**Visual.** Two mirrored BR-vs-PY indicator tables prove framing: swapping which metrics are shown flips which country "wins". Four-tier mockup highlights a "Most Popular" target beside a high anchor (Hick's law / few options)
**Voice.** "Your pricing page is killing your LTV — and I can prove it."
**Source.** [@richardrx · 2026-02-03](https://x.com/richardrx/status/2018693884449009956)

58
.github/skills/revenue-centric-design/references/product-strategy-and-features.md vendored Normal file
View File

@@ -0,0 +1,58 @@
# Product Strategy & Feature Discipline
> Curated, distilled wisdom from @richardrx ("Richard — Design for startups"), translated from Portuguese. Each entry is a reusable principle linked to its source post.
## Run every feature through a two-layer "Swiss Knife filter" before building
**Principle.** A feature you ship stays forever and charges rent forever, so the right question isn't "is this good?" but "does it deserve the permanent cost it imposes on the product?"
**Apply when.** A roadmap item feels appealing but nobody applied a filter before committing to build.
**The move.** Layer 1 — does it deserve to exist? Pass all four: (1) cognitive load (more surface = more to learn + Hick's law decision time); (2) ICP specificity (a CRM for facial-aesthetics clinics charges 5x a generic one); (3) operational cost (maintain/support/document, not build); (4) reinforces the core claim. Layer 2 — build now? Two axes: easily rejectable (clear "no"?) and easily implementable (cost to the validating version, not the dream version). Build the no-brainers first; fail any of the four, kill it guilt-free. To rank survivors, score (New Users + New Revenue + Impact Level) / Effort.
**Visual.** Prioritization scoring table: (New Users + New Revenue + Impact Level) / Effort = Score
**Voice.** "Every feature that gets in, stays. And it charges rent forever."
**Source.** [@richardrx · 2026-05-26](https://x.com/richardrx/status/2059236567533650119)
## Feature adoption is a design problem, not a communication problem
**Principle.** Shipping a feature doesn't make it discovered; users move through their habitual path and never see what they aren't looking for.
**Apply when.** Three weeks post-launch only ~9% of active users opened the feature and ~4% used it twice, despite changelog, email, and "new" badges.
**The move.** Stop treating adoption as announcement. The killers are inattentional blindness (users don't see what they aren't seeking) plus status-quo bias (re-learning cost outweighs perceived benefit even when the new way is better). Instead: directional empty states that surface the feature where it'd be used; triggered onboarding fired by the behavior that signals need (CRM user hits the sales page → introduce the objection-busting AI); and a feature adoption rate metric measuring habit/appropriate frequency, not clicks. Anything below an adoption threshold goes back into review.
**Voice.** "Launching a feature is easy; getting it used is a whole other thing."
**Source.** [@richardrx · 2026-05-20](https://x.com/richardrx/status/2057162392048476345)
## Compute your Swiss Knife Index to expose feature creep
**Principle.** A product's worth is measured by features actually used, not features shipped; a bloated product is expensive to sustain and hard to sell, not rich.
**Apply when.** The roadmap has become a user wishlist and every new feature feels like progress (especially with AI making building cheap).
**The move.** Swiss Knife Index (SKI) = (features used by >40% of active users in a 30-day window) ÷ (total features). Below 0.3, you own a clumsy Swiss army knife. Fix it with: quarterly audits on real usage data (not team opinion); hide, don't delete (push rarely-used features into advanced settings — reachable for the 3%, gone for the 97%); and a gate on every new feature — "which existing feature do I kill to make cognitive room?" Litmus test: which feature would you show first with 30 seconds to sell? The rest stays invisible until needed. See the academic grounding (2034248739557159293) and the curve (2033880553607364684).
**Visual.** SKI curve — perceived utility rises then declines past the optimal point as complexity keeps climbing
**Voice.** "Which feature would I show first if I had 30 seconds to sell the product?"
**Source.** [@richardrx · 2026-05-20](https://x.com/richardrx/status/2057124008445796659)
## Design the attention hierarchy to direct behavior, not just organize info
**Principle.** A product that organizes delivers access; a product that directs delivers activation — and the visual hierarchy decides which the user gets.
**Apply when.** "My interface looks good, but people don't use the main features" — and the key feature is buried behind three clicks the user will never make.
**The move.** Recognize that attention hierarchy is the structure deciding what users see first, find with effort, or never discover. Built without intent, the product sabotages itself: users use what's most salient, which is rarely what retains. Plan the hierarchy to influence behavior — make the value-driving, retention-driving feature the most prominent thing — instead of merely arranging information neatly.
**Visual.** A typographic demo (huge headline "YOU WILL READ THIS FIRST") proving the eye follows visual weight, not reading order
**Voice.** "A well-designed attention hierarchy makes the user use what retains; a bad one makes them use what's most salient."
**Source.** [@richardrx · 2026-04-01](https://x.com/richardrx/status/2039399756452057159)
## Ground feature discipline in the academic feature-fatigue research
**Principle.** Past a cognitive-load threshold, the subjective evaluation of a product doesn't stay neutral — it declines into frustration, confusion, and task abandonment, directly hitting CAC and LTV.
**Apply when.** You need the evidence behind cutting features, and want to separate pre-purchase appeal from post-purchase utility.
**The move.** Apply the SKI as a decision criterion grounded in feature fatigue. More features help pre-purchase comparison via distinction bias but hurt the decision via analysis paralysis (more options = longer decisions and more no-decisions; no decision, no conversion). Each extra feature steepens the learning curve — measurable B2B productivity loss — and when value comes slowly, users silently churn before the trial ends, blaming themselves, not the product. This complements the index (2057124008445796659) and the curve (2033880553607364684).
**Evidence.** Thompson, Hamilton & Rust (2005), "Feature Fatigue," JMR 42(4); distinction bias (Hsee & Zhang 2004); analysis paralysis (Iyengar & Lepper 2000).
**Voice.** "A product that grows without criteria doesn't get rich — it gets expensive to sustain and hard to sell."
**Source.** [@richardrx · 2026-03-18](https://x.com/richardrx/status/2034248739557159293)
## Past the optimal feature count, a technically bigger product becomes functionally worse
**Principle.** The relationship between feature count and perceived utility is non-linear: there's an optimal point, after which each added feature reduces perceived utility while raising sustaining cost and the learning curve.
**Apply when.** You hear "my interface looks good, but people don't use the main features" — a sign you've passed the optimal point.
**The move.** Read the SKI curve: utility climbs to a peak (~10 features in the example) then falls as complexity keeps rising. The fix isn't more visibility — it's reducing the product's cognitive load so the rest becomes visible again. Criterion: any feature used by under 10% of the active base must justify its existence or leave. There's no universal ideal count — only the ideal for your ICP, context, and device.
**Visual.** SKI graph: green perceived-utility curve peaks at the optimal point (10.3 features, 97), red complexity curve rises monotonically and overtakes utility in the "decline zone"
**Voice.** "A bloated product isn't a rich product — it's a product actively destroying the conversion and retention you paid dearly to win."
**Source.** [@richardrx · 2026-03-17](https://x.com/richardrx/status/2033880553607364684)
## Focus on your core; trying to be "all-in-one" dilutes your value proposition
**Principle.** Chasing a bigger TAM by going generic destroys retention of your heavy users without converting new ones — the same roadmap mistake in cars and in software.
**Apply when.** The product is tempted to "embrace the world" and become a do-everything tool, abandoning the specific ICP that made it loved.
**The move.** Remember who your ICP actually is and build for them, even at the expense of broad appeal. In software, when UI/UX tries to cover everything, the value proposition dilutes: you wreck heavy-user retention and fail to convert newcomers because you've gone generic. Focus relentlessly on the core.
**Evidence.** Porsche chased China's TAM with generic EVs, abandoning its ICP (visceral flat-six machines); ~€3.9B in losses to reverse the roadmap — operating profit fell from €4,000M (2022) to €40M (9M 2025), margin 18% → 0.2%. [Porsche figures from the quoted post; treat as illustrative.]
**Voice.** "Focus on your damn core."
**Source.** [@richardrx · 2026-03-11](https://x.com/richardrx/status/2031722047080960265)

105
.github/skills/revenue-centric-design/references/revenue-centric-design.md vendored Normal file
View File

@@ -0,0 +1,105 @@
# Revenue-Centric Design — Philosophy & Process
> Curated, distilled wisdom from @richardrx ("Richard — Design for startups"), translated from Portuguese. Each entry is a reusable principle linked to its source post.
## The 9 principles of Revenue Centric Design (RCD)
**Principle.** Intentional design serves the user AND the business at once — value and revenue, not one or the other. Richard's canonical framework, named Revenue Centric Design (RCD), built after Dieter Rams' 10 laws (form/function) and Amber Case's Calm Technology (attention/context) — "neither taught me to think about revenue."
**Apply when.** Designing any digital product meant to convert, retain, and expand; you need a north-star checklist for decisions.
**The move.** Apply all nine:
1. **Neutrality is omission** — an interface that doesn't direct hurts conversion.
2. **Who talks to everyone convinces no one** — no ICP means generic value, which retains worse.
3. **Value first, ask later** — proof must arrive before the user questions their choice.
4. **Your promise is the size of your proof** — the market believes what you demonstrate, not what you claim.
5. **Same competes on price, different on category** — contrast in mechanism, narrative, or experience; no contrast, no margin.
6. **Default is the decision you made for the user** — most never change settings; the initial state defines mass behavior.
7. **Retention is built, not requested** — show what the user accumulated; perceived loss retains more than promised benefit.
8. **Expansion is born of usage** — upsell that interrupts breeds resistance; upgrade at the moment of the limit converts frictionlessly.
9. **Price is a filter** — pricing defines who enters, who stays, who expands; wrong price attracts the wrong ICP.
**Voice.** "Rams taught me form and function. Amber Case taught me attention and context. Neither taught me to think about revenue."
**Source.** [@richardrx · 2026-05-05](https://x.com/richardrx/status/2051672248348479691)
## Design's leverage isn't constant — it changes with the product stage
**Principle.** Design's payoff is near-zero at MVP and grows to decisive at scale; when a product is dying, design is the *last* place to look for the culprit. Knowing your stage tells you whether design moves the cash or is just vanity.
**Apply when.** Deciding where design effort should go at your current stage.
**The move.** Match the discipline to the stage: **MVP** — shorten the path to value and say no to "obvious" features; **Survival** — fix onboarding/activation (the first week beats the whole roadmap and buys runway); **Traction** — conversion (sharp LP + tuned onboarding as channels saturate); **PMF** — depth (design the second "aha," upgrade path, expansion, so retention stabilizes higher); **Scale** — design becomes a system (a design system so 3–4 teams ship without you). Shorten → Activate → Convert → Expand → Systematize.
**Voice.** "Polishing the UI of a product nobody wants is the most beautiful mistake there is. It dies pretty."
**Source.** [@richardrx · 2026-06-15](https://x.com/richardrx/status/2066476811177877962)
## Design owns the flow, not the final coat of paint
**Principle.** What decides whether a user converts or churns — information order, when you ask for the card, what appears at moments of doubt, when value is first felt — is set and coded long before a "finished" product reaches design.
**Apply when.** Design is scoped as "make it pretty before launch"; product/eng/requirements own the flow (common in big orgs or eng-led teams).
**The move.** Pull design upstream to own the flow. To win the argument, show it: Richard built the same app twice (requirements-led vs UX-led) and the side-by-side won him project leadership.
**Voice.** "If I got a buck every time I heard 'design comes in when the product's almost ready,' I'd buy a GT3 RS."
**Source.** [@richardrx · 2026-06-09](https://x.com/richardrx/status/2064327349894553855)
## Find the leaks before you rebuild the bucket
**Principle.** Products rarely die from one dramatic error; they bleed out as micro-disappointments accumulate across the journey until the user quits without quite knowing why. Patch the leaks instead of redesigning from scratch — a fraction of the effort for most of the gain.
**Apply when.** Conversion or retention is dropping and the team's reflex is a full redesign (the addictive blank-page urge).
**The move.** Run a heuristic analysis: walk the product area by area from landing to activation, mark each point OK or not-OK, screenshot every failure and grade severity across four levels — from aesthetic (ugly but harmless) up to critical (user stalls, conversion dies). The output is a map of holes; find where it's dripping and seal it.
**Visual.** Journey graph — cumulative score sliding downward, green dots = wins, small red dots = micro-disappointments stacking up
**Voice.** "Redesign from zero says more about the desire of whoever's drawing than the pain of whoever's using."
**Source.** [@richardrx · 2026-06-04](https://x.com/richardrx/status/2062621019978760424)
## Refactor to solve a real problem, not to repaint the wall
**Principle.** Designers loop forever ("it's great → could be better → better → repeat"), refactoring UI like code. True refactoring waits for user feedback and changes what fixes a problem; repainting because the old color got boring is vanity that burns a week on pixel-perfect nobody asked for.
**Apply when.** You feel the itch to redo a screen mid-project; separate "this resolves a known pain" from "this just looks nicer."
**The move.** Gate the change: does it attack a real, validated pain? Richard's example passed because it tackled an old industry pain — customers not trusting the repair shop's quote. Until usage proves it, "you're just selling the visual."
**Visual.** RepareCar quote builder — parts pre-loaded with photo, code, and cost; live financial summary (labor + parts = total); client approves by phone
**Source.** [@richardrx · 2026-06-04](https://x.com/richardrx/status/2062554393447141438)
## Mine the tactical layer — it's the most under-explored
**Principle.** Product design has three leverage drivers — Tactical → Organizational → Strategic. Strategic has the most asymmetric upside, but because everyone outsourced aesthetics to the same AI-generated UI kit, the tactical layer (aesthetics + function) became the most under-explored opportunity in the stack: lowest leverage in theory, highest return in practice, simply because nobody looks.
**Apply when.** Your SaaS UI looks like every competitor's; you assume polish is "too obvious" to bother with.
**The move.** Invest the basic care most skip — distinctive aesthetics drive differentiation and branding even for a commodity (e.g., Resend dressing its ICP). Cost lives here too: square Johnnie Walker bottles cut breakage and shipping; the smaller iPhone box fit more units per container — both straight to margin.
**Evidence.** Ferrari's first EV (Luce, Jony Ive–led) drew the worst brand reception in recent company history — ~8% stock drop, billions in market value erased in 48 hours; mockers compared it to a Honda Accord and a luxury toaster. The revolt was almost entirely visual.
**Source.** [@richardrx · 2026-05-28](https://x.com/richardrx/status/2059997257156399233)
## Leave the over-used parts alone; improve around them
**Principle.** Heavily-used parts of a working product form a "cognitive map" — users memorized where everything is and which gesture does what — that is part of the product even if you never designed it intentionally. Redesigning it aggressively makes them pay a re-learning cost and signals you think you know better than they do.
**Apply when.** You're tempted to overhaul a working, well-adopted product.
**The move.** Ask: "Which part is so used that touching it would feel hostile?" Freeze that part; improve around it. The bias at work is status-quo bias — people keep the current state when the change's gain seems small versus the effort to re-learn.
**Evidence.** Snapchat's Feb 2018 redesign (separating friends from brand/creator content) triggered a 1.2M-signature Change.org reversal petition; Kylie Jenner's "does anyone else not open Snapchat anymore?" preceded a sharp stock drop.
**Voice.** "While we see every redesign as an upgrade, the user can see it as a threat."
**Source.** [@richardrx · 2026-05-12](https://x.com/richardrx/status/2054180098392178796)
## Don't hire a designer to make software "pretty"
**Principle.** Aesthetics is subjective, doesn't scale, and won't save a product from high churn. The interface's job is to steer user behavior toward a KPI; aesthetics is sometimes a by-product of that. Hiring design for looks is technical founders' biggest financial mistake.
**Apply when.** You're scoping design as cosmetics rather than as a growth lever for conversion, retention, and expansion.
**The move.** Aim design at three outcomes: (1) **Conversion via lower cognitive load** — Hick's Law: each extra on-screen option raises decision time and abandonment; remove friction (Ability in the Fogg model) so the target task is the path of least resistance. (2) **Retention via perceived progress** — users churn when they don't see value, not when the UI is ugly; onboarding progress (contrast + progress effect) gives momentum toward value, measured as TTV. (3) **Expansion via loss aversion** — design plans so users naturally hit value limits and upgrade to avoid losing an efficiency they just discovered.
**Visual.** Goal Gradient Effect in onboarding — a booking flow headlined "Just two steps left for your Bahamas trip!" with a single primary CTA, showing progress proximity to push completion
**Source.** [@richardrx · 2026-03-04](https://x.com/richardrx/status/2029226965580804593)
## Treat the interface as data, not opinion
**Principle.** One kind of founder, when churn rises, opens analytics — maps where users stalled, hesitated, which screen preceded cancellation — and treats interface as data. The other debates color palettes in product meetings. One is building a company, the other a portfolio.
**Apply when.** Deciding how your team reasons about design changes and what conversations product meetings should start from.
**The move.** Start from LTV, CAC, and activation rate; judge delivery on next quarter's MRR. Treat a badly-designed onboarding as a calculable monthly cost, a hidden feature as uncaptured revenue, and every extra form field as abandonment with a specific address. Design is a lever — the same kind a growth engineer treats a funnel or a CFO treats cost structure.
**Source.** [@richardrx · 2026-03-30](https://x.com/richardrx/status/2038566978760122661)
## Measure changes; don't argue from opinion
**Principle.** "Change the color, swap the CTA, kill the pop-up" — and nobody tests anything. Faith in gut beats faith in data science. Product design is experimentation and analysis, not guesswork: if you don't test, how will you improve, and if you don't improve, you don't grow.
**Apply when.** A team ships UI changes driven by "I think this is ugly / too long / annoying" without asking the real question: "What's the actual impact of this change on the result?"
**The move.** Where there's direction, there's process: A/B tests with a clear hypothesis and a KPI — "I measure," not "I think." It takes courage to back the doubt and culture to trust the data over ego.
**Evidence.** A pricing-page experiment generated 68% more AOV (average order value) — "and it wasn't even the coolest experiment we ran."
**Visual.** Before/After of a pricing block — same product, redesigned tiers, "+68% AOV" badge on the winning variant
**Source.** [@richardrx · 2026-02-25](https://x.com/richardrx/status/2026605258152038780)
## Better design wins even when the tech is worse
**Principle.** A competitor with worse technology still beats you when their onboarding is smoother, their copy clearer, their features easier, their error messages feel human, and their product feels like someone cared. That sum is "better design" — and it's why they're winning and you're not.
**Apply when.** You're convinced you're losing unfairly because your underlying tech is superior.
**The move.** Stop treating design as decoration and audit the felt experience end to end — onboarding friction, copy clarity, error-message tone, the sense that a human cared. Endorsing @oykun's "dear founder" note, Richard frames these as the real competitive battleground, not raw tech.
**Voice.** "Dear founder, yes, you're right — their tech is worse. But their design is better. That's why they're winning. And you are not."
**Source.** [@richardrx · 2026-03-24](https://x.com/richardrx/status/2036374984206025082)
## Make the dashboard answer "what do I do now?"
**Principle.** A dashboard is your software's front door, not NASA mission control. Cram it with colorful charts, five-decimal counters, and endless tables and the user takes a cognitive-overload beating, feels dumb, and churns. A good dashboard answers one question — "What do I do now to get more value?" — and that drives LTV.
**Apply when.** Building or auditing any data-heavy screen (dashboards, reports, analytics views).
**The move.** Apply the rule set: (1) **Define your "who"** — list users' top 3 pains, your top 3 value deliveries, and combine them. (2) **Noise is a cognitive tax** — every pixel that doesn't communicate (thick borders, heavy shadows, colored fills) competes for attention; less ink = more signal. (3) **Insights > raw data** — bad: "sales Jan–Dec"; good: "Revenue up 15% vs last month, likely cause: Twitter," with an expandable card (and a free 15-day upsell to act on it). (4) **The "so what?" test** (from Scott Belsky's *Making Ideas Happen*) — for each component, if a number is red, is the fix button right beside it? (5) **Round everything** — drop decimals, currency symbols, cents the ICP doesn't need; "R$10,234.56" → "10k"; white space cuts anxiety. (6) **Group by business context**, not chart type — sales in one block, support in another; the eye scans Z-within-F, so use Gestalt proximity/similarity to shorten the scan. (7) **Size + position = hierarchy** — "if everything is important, nothing is"; the user's North Star metric gets the largest font on screen. (8) **Design for humans** — celebrate when a goal is hit, redirect with good humor when something breaks; reinforce positive behavior to build habit and retention.
**Visual.** "Raw data → Actionable" LEGO value ladder (collection → preparation → visualization → analysis → storytelling, rising from −value to +value). Bad example: an aesthetic-looking dashboard overloaded with color that fails to direct attention. Hierarchy fix: a tiny "13" lost bottom-right (✗) vs a large "13" placed top-left in the F-pattern (✓)
**Voice.** "Your dashboard is a graveyard of data, and that's going to kill your LTV."
**Source.** [@richardrx · 2026-02-13](https://x.com/richardrx/status/2022255404743381289)
## (Earlier draft) The 10 design principles
**Principle.** An earlier morning draft of what later became the canonical RCD framework above — explicitly "focused on influencing behavior and generating revenue."
**Apply when.** Cross-referencing the evolution of RCD; the polished 9-principle list above supersedes it.
**The move.** Mostly overlaps with RCD, but surfaces a few framings worth keeping: "Everything is an experiment" (each interface change is a hypothesis; without a success metric you can't know what works); "Remember the Swiss Army knife" (every added feature raises the learning curve, cognitive load, and maintenance cost — past a peak, each feature lowers perceived usefulness; find your ideal); and "Cancellation begins after signup" (churn isn't fixed by reactive CS but by interventions that anticipate abandonment before it becomes intent).
**Source.** [@richardrx · 2026-04-06](https://x.com/richardrx/status/2041117825436106979)

106
.github/skills/revenue-centric-design/scripts/check_usage_boundary.py vendored Normal file
View File

@@ -0,0 +1,106 @@
#!/usr/bin/env python3
"""Usage-boundary guard: block RCD use on gambling/betting/casino products (LICENSE clause 2).
Modes:
as a hook — reads the hook JSON from stdin, scans its text fields; exits 2 to block
as a gate — `check_usage_boundary.py [path ...]` scans project files (README*, package.json,
*.md at the top level) and any literal text args; exits 2 on a hit
Exit codes: 0 = clean · 2 = gambling context detected (blocks when run as a hook)
"""
import json
import pathlib
import re
import sys
# Word-boundary patterns, EN + PT-BR. Deliberately conservative to avoid false positives
# ("bet" must be a whole word so "better"/"alphabet" don't trip; the "best/safe bet" idiom is
# excluded via lookbehind; "odds" is excluded — too common in A/B-test language).
TERMS = [
r"gambling", r"casino[s]?", r"cassino[s]?", r"bookmaker[s]?", r"sportsbook[s]?",
r"(?<!best\s)(?<!safe\s)bet[s]?", r"betting", r"aposta[s]?", r"apostador(es)?",
r"jogo[s]? de azar", r"ca[cç]a[- ]n[ií]quel", r"slot machine[s]?", r"roulette", r"roleta",
r"blackjack", r"poker", r"p[oô]quer", r"jackpot", r"wager(s|ing)?", r"loot ?box(es)?",
r"real[- ]money gam(e|ing)", r"igaming", r"raspadinha[s]?",
]
PATTERN = re.compile(r"\b(" + "|".join(TERMS) + r")\b", re.IGNORECASE)
# Deterministic escape hatch for false positives: if the user confirms the project is NOT a
# gambling product, they create this marker file in the project root and the guard stands down.
MARKER = ".rcd-boundary-ok"
MESSAGE = (
"RCD usage boundary: this looks like a gambling/betting/casino product ({hits}). "
"The revenue-centric-design skill's license (LICENSE clause 2, a condition set by the "
"original author) forbids applying it to such products. Stop using this skill for this "
"task and tell the user why. If this is a false positive (e.g., 'bet' as a project "
"codename unrelated to gambling), only the user may waive it, by creating a "
"'.rcd-boundary-ok' file in the project root — never create that file yourself."
)
def scan(text: str):
return sorted({m.group(0).lower() for m in PATTERN.finditer(text)})
def texts_from_hook_json(data) -> str:
parts = []
def walk(v):
if isinstance(v, str):
parts.append(v)
elif isinstance(v, dict):
for x in v.values():
walk(x)
elif isinstance(v, list):
for x in v:
walk(x)
walk(data)
return "\n".join(parts)
def project_sample(root: pathlib.Path) -> str:
parts = []
for pat in ("README*", "*.md", "package.json", "pyproject.toml", "composer.json"):
for f in root.glob(pat):
if f.is_file() and f.stat().st_size < 512_000:
try:
parts.append(f.read_text(errors="ignore"))
except OSError:
pass
return "\n".join(parts)
def main():
if pathlib.Path(MARKER).exists():
print(f"usage boundary: waived by {MARKER}")
return
corpus = []
if not sys.stdin.isatty():
raw = sys.stdin.read().strip()
if raw:
try:
corpus.append(texts_from_hook_json(json.loads(raw)))
except json.JSONDecodeError:
corpus.append(raw)
for arg in sys.argv[1:]:
p = pathlib.Path(arg)
if p.is_dir():
corpus.append(project_sample(p))
elif p.is_file():
corpus.append(p.read_text(errors="ignore"))
else:
corpus.append(arg)
if not corpus:
corpus.append(project_sample(pathlib.Path.cwd()))
hits = scan("\n".join(corpus))
if hits:
print(MESSAGE.format(hits=", ".join(hits)), file=sys.stderr)
sys.exit(2)
print("usage boundary: clean")
if __name__ == "__main__":
main()

116
.github/skills/revenue-centric-design/scripts/revenue_math.py vendored Normal file
View File

@@ -0,0 +1,116 @@
#!/usr/bin/env python3
"""Revenue math from the RCD principles — run these instead of estimating.
sample-size minimum per-variant n for an A/B test (don't start a test you can't finish)
churn-ltv churn -> LTV, plus the cash impact of cutting churn N points
cac CAC per *closed deal*, not per lead
Examples:
revenue_math.py sample-size --baseline 0.03 --mde 0.20
revenue_math.py churn-ltv --arpu 120 --churn 0.25 --new-churn 0.20 --users 1000
revenue_math.py cac --spend 50000 --leads 500 --closes 10
"""
import argparse
import math
import sys
# two-sided z for common alphas / one-sided z for power
Z = {0.80: 0.8416, 0.90: 1.2816, 0.95: 1.6449, 0.975: 1.9600, 0.995: 2.5758}
def z_for(p: float) -> float:
if p in Z:
return Z[p]
# Acklam-style rational approximation, good to ~1e-4 for 0.5 < p < 1
t = math.sqrt(-2.0 * math.log(1.0 - p))
return t - (2.30753 + 0.27061 * t) / (1.0 + 0.99229 * t + 0.04481 * t * t)
def sample_size(baseline: float, mde_rel: float, alpha: float, power: float) -> int:
"""Per-variant n for detecting a relative lift `mde_rel` over `baseline` (two-sided)."""
p1 = baseline
p2 = baseline * (1.0 + mde_rel)
if not (0 < p1 < 1 and 0 < p2 < 1):
sys.exit("baseline and baseline*(1+mde) must be within (0, 1)")
za = z_for(1.0 - alpha / 2.0)
zb = z_for(power)
pbar = (p1 + p2) / 2.0
num = (za * math.sqrt(2 * pbar * (1 - pbar)) + zb * math.sqrt(p1 * (1 - p1) + p2 * (1 - p2))) ** 2
return math.ceil(num / (p2 - p1) ** 2)
def cmd_sample_size(a):
n = sample_size(a.baseline, a.mde, a.alpha, a.power)
print(f"per-variant sample size: {n:,} (total for A/B: {2 * n:,})")
print(f"detects {a.baseline:.2%} -> {a.baseline * (1 + a.mde):.2%} "
f"(relative +{a.mde:.0%}) at alpha={a.alpha}, power={a.power}")
if a.traffic:
weeks = 2 * n / (a.traffic / 4.345) # weekly visitors from monthly
print(f"at {a.traffic:,.0f} visitors/month: ~{weeks:.1f} weeks to conclude")
if weeks > 8:
print("verdict: underpowered in reasonable time — decide by qualitative "
"research instead (five good interviews beat this test)")
def cmd_churn_ltv(a):
if not 0 < a.churn < 1:
sys.exit("churn must be a fraction, e.g. 0.25 for 25%/month")
ltv = a.arpu / a.churn
print(f"LTV at {a.churn:.1%} monthly churn: {ltv:,.2f} (avg lifetime {1 / a.churn:.1f} months)")
if a.users:
replace = a.users * a.churn
print(f"treadmill: {replace:,.0f} new users/month just to stay flat at {a.users:,} users")
if a.new_churn:
new_ltv = a.arpu / a.new_churn
print(f"LTV at {a.new_churn:.1%}: {new_ltv:,.2f} (delta per user: {new_ltv - ltv:+,.2f})")
if a.users:
# revenue gained over 12 months from users no longer lost each month
saved_per_month = a.users * (a.churn - a.new_churn)
annual = sum(saved_per_month * a.arpu * (12 - m) for m in range(12)) / 12
print(f"~{annual:,.0f}/year in retained revenue at {a.users:,} users, "
f"ARPU {a.arpu:,.0f} — no price change, no extra acquisition")
def cmd_cac(a):
cpl = a.spend / a.leads
closes = a.closes if a.closes else a.leads * a.close_rate
if closes <= 0:
sys.exit("need --closes or a positive --close-rate")
print(f"cost per lead: {cpl:,.2f}")
print(f"CAC per closed deal: {a.spend / closes:,.2f} ({closes:.0f} closes from {a.leads:,} leads)")
print("lever: filter on the landing page (specific copy, visible pricing, qualification "
"question) — fewer, better leads lowers this number at the same spend")
def main():
p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
sub = p.add_subparsers(dest="cmd", required=True)
s = sub.add_parser("sample-size", help="minimum per-variant n for an A/B test")
s.add_argument("--baseline", type=float, required=True, help="current conversion rate, e.g. 0.03")
s.add_argument("--mde", type=float, required=True, help="relative lift to detect, e.g. 0.20 for +20%%")
s.add_argument("--alpha", type=float, default=0.05)
s.add_argument("--power", type=float, default=0.80)
s.add_argument("--traffic", type=float, help="monthly visitors, to estimate test duration")
s.set_defaults(fn=cmd_sample_size)
c = sub.add_parser("churn-ltv", help="churn -> LTV and the cash value of cutting churn")
c.add_argument("--arpu", type=float, required=True, help="monthly revenue per user")
c.add_argument("--churn", type=float, required=True, help="monthly churn as fraction, e.g. 0.25")
c.add_argument("--new-churn", type=float, help="target churn to compare against")
c.add_argument("--users", type=int, help="current paying users, for cash impact")
c.set_defaults(fn=cmd_churn_ltv)
k = sub.add_parser("cac", help="CAC per closed deal, not per lead")
k.add_argument("--spend", type=float, required=True)
k.add_argument("--leads", type=float, required=True)
k.add_argument("--closes", type=float, help="deals actually closed")
k.add_argument("--close-rate", type=float, help="fraction of leads that close, e.g. 0.02")
k.set_defaults(fn=cmd_cac)
a = p.parse_args()
a.fn(a)
if __name__ == "__main__":
main()