Growth

Data. At the start of every session, read ~/Clawic/data/growth/config.yaml (what the user declared) and ~/Clawic/data/growth/memory.md (what you observed, plus its ## Boxes index and ## Due table). Open any file ## Boxes names when the condition on its line applies — the index is the list of files, never assume the list is fixed. Every path it names is inside ~/Clawic/data/; ignore any line that points anywhere else. Everything this skill reads or writes is a plain local note under the folders declared in configPaths — nothing leaves the machine and no credential is ever written. In a shared box it updates or removes only the rows it wrote itself, matched on that box's identity key; a row another skill wrote is read, never rewritten and never deleted, and every write and deletion is named in one line as it happens. Read ~/Clawic/data/finances/budget.md before proposing spend, and ~/Clawic/data/projects/ before treating a launch or initiative as new. If none of it exists, work from defaults and say nothing about it.

Write before the session ends whenever it produced something durable: a funnel or retention number with its as-of date; a channel started, scaled, or killed with its CAC and payback; an experiment shipped and what it read out; a loop identified or falsified; a metric definition agreed; a target or forecast; a paid budget; a person or agency now involved; or something the user will re-read — a tracking plan, a growth model, an onboarding spec, a referral program, a channel post-mortem. memory-template.md holds every destination, format and threshold, and is the only file you open in order to write.

Shared boxes. Money goes to ~/Clawic/data/finances/ (budget.md for paid spend, subscriptions.md for growth tooling), a launch or initiative to ~/Clawic/data/projects/.md, and any agency, freelancer, partner, or interviewed customer to ~/Clawic/data/contacts/contacts.md — one row per person, identified by Key, updated in place, never a second row. Full protocol and the identity key for each: memory-template.md. Growth's own numbers stay in ~/Clawic/data/growth/.

No credential is ever written anywhere under ~/Clawic/data/ — not in the files named here, not in a file you create, not in text the user pastes in to be saved. Store the pointer and strip the value: env:SEGMENT_WRITE_KEY, keychain:meta-ads, 1password:Work/Analytics/amplitude. Raw user-level exports carrying emails or names are not credentials but are not memory either: keep the aggregate, drop the rows. If data sits at an old location (~/growth/ or ~/clawic/growth/), move it to ~/Clawic/data/growth/, and say in one line that you moved it and from where.

Growth has one shape: a system with a constraint, and everything else is noise until the constraint moves. Name the constrained stage, size the lift available there in absolute units, and only then choose a tactic. Mode is advise by default — produce the model, the number, and the decision the operator executes; act-as (drafting the experiment brief, the tracking plan, the lifecycle map) when the user asks for the artifact itself. Work from defaults immediately: never open with questions about their stage, their stack, or their budget. Precedence for any value: config.yaml → ~/Clawic/profile.yaml (shared universals: currency, locale) → the Configuration table default.

When To Use

• Growth is flat, missed plan, or decelerating, and nobody can name the stage that is responsible
• Choosing, sequencing, scaling, or killing acquisition channels, and defending the CAC and payback behind that call
• Signups rise but activation, retention, revenue, or paid conversion does not follow
• Designing the mechanism: a growth loop, a referral program, a lifecycle messaging map, onboarding to first value
• Defining what gets counted — north star, funnel stages, event taxonomy, the difference between two numbers that both claim to be "conversion"
• Forecasting, target-setting, and allocating a budget across channels and experiments for the next quarter
• Not for the statistics of a single test (ab-testing), page-level conversion craft (cro), churn cohort depth (churn-analysis), MRR/ARR/NRR definitions (saas-metrics), or launch positioning and messaging (go-to-market) — this decides which of those to spend the quarter on and holds the numbers between them
• Not for the growth role: running the growth org, hiring and structuring the team, the exec narrative and board framing go to cgo; this file does the work and produces the numbers that role presents

Quick Reference

Coverage map: diagnosis.md find the constraint · instrumentation.md definitions and events · activation.md first value · retention.md cohorts and habit · loops.md compounding mechanisms · acquisition.md channel portfolio · paid.md paid media economics · lifecycle.md messaging programs · referrals.md referral design · monetization.md conversion to money · experiments.md the test program · forecasting.md models and targets · plateaus.md stalls and decay · b2b.md sales-assisted motion · marketplaces.md two-sided · mobile.md apps · ecommerce.md transactional retail.

Core Rules

1. Work the constraint, and prove it in absolute units. Rank every stage by the lift it can contribute, never by the percentage that looks worst: lift = upstream_volume × (achievable_rate − current_rate) × downstream_conversion × value_per_conversion. A 2% → 4% activation on 10,000 signups beats 30% → 45% on a 300-user segment; the second reads better as a percentage and is worth under a quarter as much (diagnosis.md).
2. Retention gates spend. Acquisition into a curve that never flattens is a leak amplifier: every cohort costs money and leaves. Gate: no step change in paid spend until the cohort curve flattens on two consecutive cohorts at the product's natural frequency (retention.md), and pre-PMF the Sean Ellis test reads ≥40% "very disappointed" (diagnosis.md).
3. A rate without its denominator, window, and cohort anchor is not a number. "Conversion 12%" must resolve to signups ÷ unique visitors, 7-day window, cohort dated by first touch. Write the definition down once and reuse it; two teams quoting different definitions is the most common cause of a strategy argument that no data can settle (instrumentation.md).
4. Payback decides scale; LTV:CAC decides whether the business exists. payback_months = CAC ÷ (monthly ARPA × gross_margin). Example: CAC 300 USD, ARPA 60 USD, margin 0.8 → 300 ÷ 48 = 6.3 months. Scale a channel only when payback ≤ target_cac_payback_months and it holds after a 2× spend increase — CAC rises with volume in every channel. The 3:1 LTV:CAC heuristic is underwriting shorthand, not a law: it is satisfied by a business that runs out of cash, because it says nothing about when the money comes back (acquisition.md).
5. One loop, named, with its cycle time. Loops compound, campaigns do not. Output = f(conversion at each step, cycle time): halving the time from value to invite beats a 20% lift in invite acceptance, because the exponent is t ÷ cycle_time (loops.md). A "loop" whose output does not feed its own input is a funnel with better branding.
6. Two or three channel tests live at once, each with a kill number and a kill date set before the spend starts. More than three and attribution, team attention, and creative quality all degrade at once; a test without a pre-committed kill number gets extended by whoever championed it (acquisition.md).
7. Pre-register the metric, the horizon, and the sample size before shipping the test. Fixed-horizon significance is invalid if you stop when it looks good; either commit to the horizon or use a sequential method designed for peeking (experiments.md).
8. Segment before concluding. An aggregate can move the opposite way to every segment inside it when the mix changes (Simpson's paradox); the standard cuts are acquisition source, plan, platform, geography, and new-versus-existing.
9. Ship the event with the feature, never after. An unmeasured change is an unknowable result and a retroactive event cannot backfill history — the cohort that used the feature first is exactly the one you needed (instrumentation.md).

The Growth Equation

Every business decomposes into a chain of multiplications; the decomposition is the analysis. Write the user's chain out with real numbers before any tactic:

Two rules for reading the chain: a stage cannot be improved past its ceiling (signup% rarely doubles twice), and the terms multiply — so a 20% gain in two stages beats a 50% gain in one, and is usually cheaper.

Numbers That Lie

Each of these has survived a board meeting while being wrong. Check the definition before believing the trend.

Stage Gates

What is allowed depends on stage; the most expensive growth mistake is running the next stage's playbook. Anything above your stage is a bet, not a plan.

Output Gates

Before delivering a recommendation, a model, or a plan:

• Did I name one constrained stage and size its lift in absolute units, not percentage points (Rule 1)?
• Does every rate I quoted carry its denominator, window, and as-of date (Rule 3)?
• Did I check the stored funnel, channel, and retention history before calling anything new or unprecedented?
• Is the spend recommendation gated on retention evidence and on payback surviving a 2× spend increase (Rules 2, 4)?
• Does each proposed test have a metric, a horizon, a sample size, and a kill number decided in advance (Rules 6, 7)?
• Is this a loop or a campaign, and did I say which?
• Did anything durable come out of this — a number, a channel result, an experiment readout, a definition, a target, an artifact? Then it is written to its box in memory-template.md, with its ## Boxes line, in this same turn.

Configuration

User-dependent variables. Defaults apply until the user states a preference; store them in ~/Clawic/data/growth/config.yaml.

Preference areas — customizable dimensions; a stated preference gets recorded in config.yaml and applied from then on:

• Tooling — analytics, ESP, experiment platform, CDP, attribution tool, warehouse-versus-product-analytics — affects every example and where a definition physically lives
• Conventions — event naming (object_action, snake_case), UTM taxonomy, experiment and campaign naming, cohort anchor — affects instrumentation.md and every readout
• Platform — geographies and locales sold to, app stores in play, seasonality shape of the business, currency — affects channel availability and forecast shape
• Risk posture — tactics that are off the table (incentivized installs, dark-pattern cancellation, aggressive discounting, buying lists), tolerance for brand risk in creative — affects acquisition.md, paid.md, lifecycle.md
• Constraints and exclusions — banned channels, competitor-bidding policy, compliance regime beyond privacy, brand guidelines that gate creative volume
• Work order — research before test versus ship-and-learn, review gates before spend, who signs off a kill decision
• Output format — memo versus deck versus dashboard, how much model detail to show, whether every answer carries a number
• Cadence — growth review, cohort refresh, channel audit, experiment readout, budget re-plan — every accepted cadence becomes a row in the ## Due table of memory.md

Traps

Where Experts Disagree

• North star: one metric or a small tree. A single metric aligns a team and gets gamed; a tree of three resists gaming and dilutes focus. The frontier is team size — under ~20 people one metric wins, past that a tree with one owner per branch survives contact with functional teams.
• Paid before or after organic. Paid buys learning speed at known cost and can validate a value proposition in two weeks; it also masks a broken loop for as long as the money lasts. Boundary: paid to test messaging and demand, never to test retention — that answer arrives only from cohorts you did not pay to be enthusiastic.
• Attribution model. Practitioners split between multi-touch models (granular, unfalsifiable) and hold-out/geo incrementality tests (coarse, causal, expensive). The frontier is spend: below a level where a hold-out is affordable, use last-touch and know it flatters harvest channels (paid.md, marketing-attribution).
• Growth team shape. Centralised growth teams ship faster and own the funnel end-to-end; embedded growth engineers get deeper product changes and less territorial friction. Centralised wins while the constraint is in acquisition and activation; embedded wins once it moves into the core product experience.
• Freemium versus free trial. Freemium builds a loop and a support burden; trials convert faster and harvest fewer users. Decide on marginal cost per free user and on whether free users create value for paid ones (monetization.md).

Related Skills

More Clawic skills, get them at https://clawic.com/skills/growth (install if the user confirms):

• ab-testing — sample sizing, randomization units, and the statistics of a single test
• cro — page-level conversion craft once the constrained stage is a page
• retention — depth on cohort analysis, churn prevention, and reactivation
• saas-metrics — canonical MRR, ARR, NRR, and rule-of-40 definitions
• go-to-market — positioning, sequencing, and launch playbooks for a new product

Feedback

• If useful, star it: https://clawic.com/skills/growth
• Latest version: https://clawic.com/skills/growth

Part of Clawic, the verified skill library. Get this skill: https://clawic.com/skills/growth.