TMA Growth Engine — Decision Matrix & Operating Plan (Jul 2026)

§0What this is

This is our sprint to grow MRR, run through the two funnels that produce it — plus the email engine that converts them and the organic offence engine that fills them. Two numbers tell us it's working: lead → trial per funnel and month-on-month new-MRR growth. Retention now has its own plan (post-first-payment); this doc is top-of-funnel only — stranger → first payer.

The two funnels + the backend

Funnel	The path	Why it matters
1 · Web	quiz → paywall → email → nurture → paid (two paths at the paywall: carded trial-start OR direct ~50%-off purchase — gate metric = Lead→Paid, corrected 16 Jul)	Fast to change (no app release) — we learn here first
2 · App (iOS-first)	install → onboarding + assessment → paywall → 7-day trial → subscribe	Where 191 of our 763 paying members live (572 are web/Stripe)
Backend · Email	lead→trial nurture · trial→paid · churn-defence	Converts both funnels; moving ActiveCampaign → Customer.io (Intercom already runs our in-app messages)

Where we are today

MRR $11,176 (all figures USD — one currency, Aga's rule) — Stripe web $7,771 (538 subs) + RevenueCat app $3,405 (240 subs) = 778 paying subs (auto-refreshed 12 Aug). The "£10K sub-sum" over-read by counting 110 already-cancelling subs. Early signal of ~−6% MoM — needs 2–3 months to confirm as a trend. Break-even = $15K/mo — canonical (the survival gate; same figure as the old "£12K"), but the goal isn't just break-even, it's a growth rate.
We can see that sales happen, not yet which step made them. The platforms show ~761 paying subs (565 web + 196 app, 10 Jul), but the analytics layer recorded the trial event only a handful of times in 90 days — the funnel is real but not readable end-to-end. Nic's already wiring the fix (~10 days) — full spec in the analytics brief.
Paid runs to the web quiz now — Meta + Google → the free 8-min assessment (a matched offer, never £297-cold), measured on cost-per-lead (which GA4 shows today). Nic runs the ads. iOS ads follow once the app funnel's instrumented. July's lesson wasn't "don't run paid" — it was "don't send cold traffic to a £297 lifetime offer you can't measure."
Paid guardrails (so July can't repeat): in-platform budget ceiling ($3K/mo — ramp, Mo.1 ≈$40/day; SoT amended 9 Jul; spend still gated on lead→trial visibility) · weekend lock · a daily spend alert to both of us · and lead quality reviewed weekly by source (email opens/clicks of new leads) — cheap junk leads are the trap when optimising on cost-per-lead.
The plan in one line: turn on free organic growth now → make both funnels measurable → fix the biggest leaks → then scale paid, gated on proof it's incremental.

★ My recommendation (read this, then unfold the rest)

You don't have a traffic problem. You have a visibility problem and a focus problem. The engine is dark — RevenueCat holds ~190 paying subs; Amplitude sees 3–5 — so every fix today is a guess, and July proved what guessing costs. And the business is spread across 10+ workstreams while MRR sits flat-to-declining. The whole point of this matrix is to say no to almost everything and yes to one chain.

Two things run at once:

① Always-on — the offence engine. Not a task, it just runs: organic content (Lee + Ritesh) + paid ads (Nic — Meta/Google, small and capped) → the free assessment → email, plus the always-on quiz-email retargeting. This is how we grow while the build below happens — it never waits on the wiring, and it compounds week over week.

② In sequence — the build (mostly Nic):

This week — make the numbers real. Nic wires RC→Amplitude, fixes the checkout metric, captures UTMs. ~1.5 days, no app release. Everything below is unreadable until this fires.
Then — instrument both funnels end-to-end (canonical user_id = the keystone), and fix the biggest leaks — web quiz first (it iterates without an app release).
Prove paid works, then scale it. Paid's already running small to the quiz (cost-per-lead is measurable today); once we can see lead→trial, scale what works and open iOS.

The numbers you and Nic watch: Lead → Paid, per funnel & source + net MRR Δ (new − churn) + MoM new-MRR growth — $15K break-even is the gate, the growth rate is the prize — reconciled to Stripe + RevenueCat. If work didn't move them, it didn't count.

Funnel model (locked 16 Jul, auto-synced): Lead → Paid, by source — two paths: carded free trial (trial-start IS the purchase) + direct ~50%-off purchase (no trial). Not "trial → paid".

Two things worth knowing up front

Retention-Now lives IN this doc (amended 8 Jul — approved by Aga). The deeper retention programme (habit loops, product quality) lives in the Retention Plan, but the three immediate churn actions are in Phase 0 below, because MRR = new − churn and the churn side is currently the bigger number. A cockpit that only watches top-of-funnel watches MRR fall without knowing why.
We may not have the volume for classic A/B tests. A proper split-test needs a few hundred trials per version; at our current size we may not hit that quickly. So some "tests" will be bigger, bolder changes we judge by direction — not tiny split-tests. We'll check the real numbers before committing.

§0.5The Sprint Contract — the maths, the targets, the finish line (amendments approved 8 Jul)

The maths, run backwards (USD)

Gap: MRR $11.1K → break-even $15K = $3.9K to close. Blended ARPU ≈ $14.5/sub/mo → ~270 net new subs.
The bleed: web rail is running −29 subs net/week (33 cancels vs 4 new) ≈ −$1.8K MRR/month. Standing still costs ~130 new subs/month.
What reaching $15K in ~4 months requires: ~200 new subs/month = ~48/week vs today's 4 (12×) — or cut churn in half (~16 saves/week ≈ +$1K MRR/mo), which is worth roughly the same as tripling lead flow and is achievable this month. That's why churn defence is Phase 0, not "another doc".
Volume chain at target rates: ~4 trials/day (at 40% trial→paid) → ~350 leads/week (vs ~102 today). Email + organic must carry this while paid is gated.

The dated targets — the sprint has a finish line

Sprint 1 (14 days, ends 22 Jul): T2 live & reconciled ±5% · 110-queue save campaign sent · cancel-flow intervention live · retarget flow live (bot-purged list only) · D6 pre-bill nudge live · UTM capture live. "Done" = a number moving on the daily cockpit.
By 31 Aug: churn <15/week · leads ≥250/week · lead→trial readable per funnel · net-MRR Δ positive 2 consecutive weeks.
By 1 Nov: MRR ≥ $15K (break-even) — recurring only, one-time cash never counted. Then the growth-rate phase toward $25K+.

Retention-Now — three actions, in Phase 0 (the fastest MRR in the plan)

Action	What	Owner	Why now
R1 · Save the 110 queue	The 110 cancel-at-period-end subs are individually identifiable in Stripe, carrying ~$1.6K MRR out the door. Personal Jesse-voice email: pause option · downgrade-to-quarterly · rescue offer. Fleet writes, Aga eyeballs, send this week.	Claude + Aga	The single fastest MRR defence available — days, not weeks.
R2 · Cancel-flow intervention	When someone clicks cancel: offer pause / downgrade / rescue before confirming. Standard SaaS save-flow; ~a day of work on the web side.	Nic (web) · Claude drafts copy	Converts a % of every future cancel, forever. Compounds.
R3 · D6 pre-bill nudge	Pulled forward from Phase 2 (#16): Day-6 push + loss-framing email before the trial bills. RC knows the trial date natively — no instrumentation gate.	Nic + Claude	#1 cancellation driver in card-upfront models is billing-day surprise.

The always-on engine — quotas and owners (was two lines; now a contract)

Engine	The weekly contract	Owner	Cockpit row
Email revenue calendar	1 value email + campaign moments to engaged, bot-purged segments (43K list; July proved email drives the sales — ~£6.8K of ~£9K). Fleet writes via /email-fleet, Aga approves. Monthly offer moment to warm segments only.	Claude fleet + Aga	Email-attributed revenue /wk
Organic content	Upside-only per the SoT (social reach ~dead — not a pillar): cheap ManyChat "comment ASSESS" capture + one creator/affiliate shot, everything pointing at the free assessment; ManyChat→quiz UTM link verified. Paid small-&-capped is the reach engine.	Ritesh (make) · Lee (post)	Organic leads /wk (upside)
Quiz-email retargeting	Ongoing nurture + small warm Meta custom audience — gated on the bot purge (engaged-only seed, per the 00e send-plan rule), matched offer, never lifetime-to-cold, holdout measured.	Nic (Meta) · Claude (emails)	Lead→trial, re-engaged cohort

Owner split — Nic-only vs Claude-builds (capacity fix)

Nic is the constraint (~85–145h backlog) — so everything buildable without him moves to Claude (built in-package, Aga eyeballs before ship, per the standing rule): quiz T6 web events · exit-intent capture · paywall/checkout copy · all email sequences & campaigns · R1 save campaign · A/B variant builds · this cockpit. Nic-only: T2 RC→Amplitude · T3 PIs · T4 user_id (+ RN review) · Meta/CAPI · CIO wiring · cancel-flow backend. RN-team question resolves by Fri 11 Jul — can they build from the prod branch, hours + rate — or Phase 1 native re-plans around Nic alone.

Offer architecture — added to the test queue (Phase 2)

July's root cause was offer×audience — so the biggest untested conversion lever is what cold traffic's first paid step should be: 7-day free trial vs the existing $9.99/30-day trial (/30trial) vs annual-first. This is now test #17b. Trial = card-on-file = the buying decision; the entry offer sets everything downstream.

§0.6The spine — Measure → Fix → Scale, and what we push while Nic measures

The whole sprint runs on one sequence: Measure → Fix conversion → Fix retention → then scale paid. Nic already owns Measure and is building it now. That stage does not block the others — while the instrumentation lands, Aga + the Claude fleet run the conversion and retention sprints in parallel (none of them wait on Nic). Only paid is truly gated. Each stage below carries its live to-do; every task names the MRR lane it moves (per §7's rule — no lane, it waits).

① MEASURE · 🔧 Nic, in flight→ ② FIX CONVERSION · ▶ now→ ③ FIX RETENTION · ▶ now→ ④ SCALE PAID · 🔒 gated

Stage ① · Measure — see the funnel 🔧 in flight · Nic

Nic's lane — we track it, we don't duplicate it.

☐ T2 · RC → Amplitude wire — the one constraint Nic
☐ T3 · Fix lifetime-cart page-load PIs — begin_checkout on real intent, not page-load Nic
☐ T4 · Canonical user_id — the keystone that joins channel data Nic + RN
☐ T5 · AppsFlyer conversion events → cost-per-trial becomes readable Nic + RN
☐ UTM capture live — every lead ties back to its source Nic

Done = cost-per-trial + trial→paid readable on the daily cockpit. Until then, paid stays gated.

Stage ② · Fix conversion — more trials from the traffic we already have ▶ push now

Parallel, no Nic gate · £0 ad spend · lane ① trial starts

☐ Quiz screen-1 fix + confirm organic completion >25% (200+ sessions) Claude → Nic ship
☐ Landing v5 — pick direction (blind desire-pick) → build → ship Claude + Aga
☐ Checkout safe wins — annual-first paywall · remove cold-lifetime · 60-day guarantee by the card fields · named charge-date (trial only) · mobile hygiene Claude copy · Nic ship
☐ Capture — earlier email capture + exit-intent "save progress" Claude → Nic
☐ Build the entry-offer test (7-day free vs $9.99/30-day vs annual-first) — build now, launch when T3 reads Claude

Stage ③ · Fix retention — stop the bleed ▶ push now

The fastest MRR in the plan — a halved churn ≈ tripling lead flow · lane ② save churn

☐ R1 · Save the cancel queue — 113 subs · $1,612/mo walking out (live 14 Jul). Jesse-voice email: pause / downgrade / rescue. SEND THIS WEEK Claude writes · Aga sends
☐ R2 · Cancel-flow intervention — pause / downgrade / rescue before confirm; converts a % of every future cancel, forever Nic web · Claude copy
☐ R3 · D6 pre-bill nudge — push + loss-framing email before the trial bills Nic + Claude
☐ Renewal sequence — currently MISSING; build it Claude
☐ 14-day no-workout churn trigger — pulled forward Nic + Claude
✅ Win-back lapsed — live in AC done

Stage ④ · Scale paid — only after the funnel can be seen 🔒 gated · do not start

Opens only when Stage ① reads and a holdout proves the spend is incremental.

☐ Gate: cost-per-trial readable + holdout proves incrementality (retargeting often isn't)
☐ Cap $3,000/mo — ramp, not splash (the $30–40/day cap is retired, per the SoT)
☐ Prove one channel cold (Meta) first before opening Google category search
☐ Meta CAPI repoint Lead → Purchase

Running underneath all four, every week (doesn't stop for measurement)

The email revenue calendar (1 value email + campaign moments to engaged, bot-purged segments — July proved email drove ~£6.8K of ~£9K) and organic capture (ManyChat "comment ASSESS" → free assessment). See The always-on engine table above. The filter for anything new joining this list: name the MRR lane it touches in 30 days — no lane, it waits (§7).

§1The four decisions you asked

Clear yes/no first; unfold each for the full reasoning, the data, and the skeptic's correction.

1 · iOS ads — early results look promising. What's the framework to scale?scale on trial dataframework below

Where we areThe iOS results look good and it's worth leaning in. The one thing we need before we scale spend with confidence: today we can see installs, but not yet which installs become trials and subscribers — that's the RC→Amplitude + AppsFlyer link Nic's already wiring. Once it's live we can read cost per trial, which is what tells us the ads are truly working, not just cheap. The framework to scale (what has to happen)① Trial-level read live — T1 (RC→Meta CAPI), T5 (AppsFlyer conversion events), T4 (canonical user_id) so a trial ties back to the campaign that drove it. ② A few days of data so we have a real cost-per-trial number. ③ Then a small, measured step-up in spend, judged on cost per trial — scale what proves out. This is how we turn "the installs look cheap" into "we know these ads make subscribers." Two things worth doing alongside• A tiny Apple Search Ads brand test — worth it if a competitor is bidding on our brand terms (quick App Store check first). • Nic can confirm AdServices.framework is linked (10 min) so ASA needs no app release. One number to hold ontoCost per trial, net of Apple's IAP cut (~30% yr 1) — the real unit economics, and what makes the scale decision safe rather than a guess.

2 · Google Ads → web quizGO — measured on cost-per-lead

VerdictYes — but Category Search only, and not until four gates clear. No to Demand Gen / YouTube now. This is a genuinely better play than the failed Google UAC app-installs (it feeds the owned email list and the most-iterable funnel). Why the "cheaper" claim is half-trueCPC is cheap — calisthenics category keywords run £0.80–1.80 UK. But with today's broken quiz screen-1 and dark email nurture, that clicks out to a Cost per Trial of ~£41–66 (MODELLED) — 2.5–4× the ~£16 ceiling. Cheap traffic into a leaky, blind funnel is the exact July mistake. July's Google result (1,809 page-loads, 0 sales) wasn't proof Google can't work — it was proof that spending before conversion events are wired produces no usable information. The four gates (in order)① T3 live (begin_checkout fires on real intent, not page-load). ② Quiz screen-1 fixed and organic completion confirmed >25% from 200+ sessions. ③ Meta restarted and proven first — prove one channel cold before opening the next; if Meta is under £16/trial and Google is unproven, scale Meta, don't diversify. ④ T6 + Google offline-conversion import wired (bid to trial_started, not page-views). Free bonus testOnce T3 is live, run a 14-day brand-keyword pause test (£0). If organic branded traffic stays flat, branded Google spend is pure harvesting — cut it, recover ~£100–150/mo. If it drops >20%, it's incremental. Highest information-per-pound decision in the set.

3 · How we split the analytics buildNic + RN team~10 days, best case

The splitNic's already executing Phase 0 (the RC→Amplitude wire, the checkout metric fix, UTM capture) — the highest-leverage work, no app release needed. To hold the ~10-day path, the RN team can take the native Phase 1–2 (T4–T7, the SDK-level user_id + conversion events) in parallel, with Nic reviewing the two attribution-critical merges (T4 user_id, T5 purchase callback) before they land. Timeline — honest~10 working days for T1–T6 is best case, if the user_id schema's locked on Day 2; realistically 15–25 days. T7 adds App Store review (+12–17 calendar days), outside our control — so it's not a sprint gate. The one thing that sets the timelineConfirm the RN team can build from our production branch quickly. If yes, the 10-day path is real; if it needs environment setup first, we add that time up front rather than discover it mid-sprint. Day-2 schema lockAgree in writing: the canonical user_id + pre-auth fallback, the exact SDK call order, and the web→app pass-through — the one step that lets all the channel data join up later.

4 · Rebuild the Django checkout?NO rebuild nowSafe wins + gate

VerdictDon't rebuild now. Ship a few directionally-safe wins; gate the full rebuild on data. WhyToday's autopsy proved the payment step works (72% of card-entries succeed) — so "it's broken, rebuild it" isn't supported, and the diagnosed dominant leak was offer×audience, not checkout UX. A full rebuild is 7–10 of Nic's days on an unverified hypothesis. Critically, that 72% only measures card-entry→success; it says nothing about arrive→card-entry — the part that could actually leak — because page-load ghost PIs erase it. We literally can't see whether the page is the problem yet. The subtle trap the skeptic caughtDon't bundle CRO changes into the Rank-2 measurement fix — that contaminates the clean baseline Rank-2 exists to create (you'd get lift you can't attribute — the 8-Jul mistake in a new costume). Only the paywall-coherence fix rides along (annual-first, remove cold-lifetime) because it's a correctness fix, not an experiment. Ship now (safe, after cart closes ≥13 Jul)Paywall/checkout coherence · restate the 60-day guarantee by the card fields · named charge-date copy (trial plans ONLY) · mobile hygiene (viewport, numeric keypad, sticky CTA). ⚠ "£0 today / Start Free Trial" is a false statement on the £297 lifetime path — dark-pattern/chargeback risk. Scope to trial checkout only. Wave-B (after ~2 weeks of clean baseline)Apple/Google Pay, trust-stack, per-day framing, button A/B — all measurable, so ship them dated with a "before" snapshot. Rebuild unlocks only if≥2 weeks of clean arrive→card-entry data shows the rate is genuinely bad (<50–60% mobile) and session recordings localise where users drop. If it's ≥60%, the page isn't the leak — reallocate Nic upstream.

§2The effort × impact map — what to do first

Where every major action falls. Do-first = high impact, low effort. Everything gated on measurement is marked 🔒 gated.

◀ Lower effort ———— Higher effort ▶

▲ HIGH IMPACT · LOW EFFORT — DO FIRST

RC→Amplitude wire (T2) — the one constraint
Retarget the quiz-email list — ongoing, proven offer web-quiz £
Annual-first paywall + remove cold-lifetime web-quiz £
Fix the lifetime-cart page-load PIs (T3) — before the next promo (the evergreen sub checkout is already clean)
Win-back lapsed subs — ✅ done (live in AC)
Confirm retention is the leak (already analysed)
iOS AHA-before-paywall walkthrough (Nic)
Churn trigger (14-day no-workout) — pulled forward

▲ HIGH IMPACT · HIGH EFFORT — BIG BETS (gated)

Canonical user_id (T4) 🔒 keystone, error-prone
CIO migration 🔒 gated on event wire + deliverability
Trial→paid 7-email + lead→trial nurture 🔒
RC Experiments paywall tests 🔒 if trial volume allows
Meta CAPI repoint Lead→Purchase 🔒
Checkout rebuild 🔒 only if data proves it

▽ LOWER IMPACT · LOW EFFORT — QUICK WINS

AdServices.framework check (10 min, changes calendar)
Google brand-keyword pause test (£0 revenue decision)
Checkout safe copy (guarantee, charge-date, mobile)
ASO keyword refresh (Aga, App Store Connect)
Google Ads API token application (external lag)

▽ LOWER IMPACT / UNVERIFIED · HIGH EFFORT — DON'T (yet)

iOS Meta ads now — blind spend, wrong signal
Google Demand Gen / YouTube — low intent, unproven
Full checkout rebuild now — unverified hypothesis
Adapty migration — RC Experiments already covers it
Intercom — CIO in-app messaging covers it
Scaling any paid before the holdout

§3The lever map — what "maximising the funnel" means

Maximising isn't more traffic. MRR ≈ Traffic × %→email × %email→trial × %trial→paid × price × months-retained (+ warm-audience recovery). Every × is a lever; lift five stages 20% each and you ~2.5× the business with the same traffic. Each lever below is tagged pull now / 🔒 gated / later.

Bank	Levers	Status now
1 · Volume (more in)	Organic (YouTube/Ath, reels, SEO, ManyChat) + paid (Meta + Google) → free-assessment email capture	organic + paid ON now both point at the web quiz / free assessment, measured on cost-per-lead. iOS paid follows once the app funnel's instrumented.
2 · Capture (→ emails)	Screen-1 hook, fewer/relevant questions, earlier email capture, exit-intent "save progress"	pull now free leverage on existing traffic
3 · Convert (→ trials/paid)	Sunk-cost anchor, annual-first, AHA-before-paywall, trial mechanics, D6 pre-bill nudge	pull now (safe copy) / 🔒 (A/B) — multiplies all traffic, £0 ad spend
4 · Re-engage (warm audiences you paid for) ★	Quiz-email list → email + small Meta retarget · abandoners · installers-no-trial · trial non-converters · churned → win-back	pull now highest cheap leverage — measure incrementality with a holdout
5 · Monetise (more per converter)	Annual-first mix, price tests, lifetime/annual to warm only, expansion/upsell	🔒 gated raises LTV → raises allowable CAC → unlocks paid
6 · Retain (keep longer) ★ the wound	Onboarding/habit, streaks/skill-unlocks, 14-day no-workout churn trigger, product quality (some churn = app bugs)	→ own Retention Plan now split out (post-first-payment); still urgent — MRR's slipping.

Your instinct — "small retargeting on everyone who entered the quiz email" — is right (it's the anti-July: warm audience + matched offer). Order: email first (free, we own the list) → small Meta custom-audience retarget (matched offer, never lifetime) → measure with a holdout (retargeting often isn't incremental — you can pay to reach people who'd convert anyway). Exclude anyone who already converted.

§4The to-do list — every task, in hours

The whole backlog in sequence, grouped by phase (no calendar — just order + hours). It's overwhelmingly Nic's build, and he's already on the top of it. The detailed cards below unfold any task for the reasoning, the metric it moves, and the skeptic's caveat.

#	Task	Owner	Est.
PHASE 0 · Make the numbers real + stop the bleed — the constraint · ~20–28h
R1	Save the 110 cancel-queue — Jesse-voice pause/downgrade/rescue email to the identifiable Stripe cohort (~$1.6K MRR) — fastest MRR in the plan	Claude + Aga	2–3h
R2	Cancel-flow intervention — pause / downgrade / rescue offered before cancel confirms	Nic · Claude copy	4–6h
R3	D6 pre-bill nudge (pulled forward from #16) — push + loss-framing email; RC knows the trial date, no gate	Nic + Claude	4–8h
1	Wire RC → Amplitude (T2) — the one constraint · 9 Jul: source-prefix convention live (`rc_`=RC, `af_`=AppsFlyer, unprefixed=web; Amplitude AI-merge unifies, raw kept for audit) — `rc_` not yet visible in taxonomy at the 9 Jul API check; done = `rc_` flowing + ±5% vs RC	Nic	2–3h
2	Fix page-load PIs on the lifetime/promo cart (T3) — before the next promo (evergreen checkout is clean)	Nic	1–2h
3	Capture UTM / click-IDs → localStorage → Stripe → RC	Nic	1–2h
4	Confirm acquisition-vs-retention (already analysed = retention; refresh current numbers)	Aga/Nic	~1h
5	Re-fire the existing AC win-back to current lapsed subs — ✅ DONE (live in AC)	Aga	✅
6	Retarget the quiz-email list — email nurture + small warm Meta audience · gated on the bot purge (engaged-only seed)	Claude emails · Nic Meta	2–3h
7	Paywall + checkout coherence, safe copy (annual-first, remove cold-lifetime)	Claude drafts · Nic ships	2–3h
8	Nic walks the iOS onboarding flow — AHA before the paywall?	Nic	1–2h
9	Two 10-min checks: RC Experiments tier · AdServices.framework linked	Aga/Nic	0.5h
10	AC baseline audit + snapshot all baselines before Phase-0 changes	Aga	2–3h
PHASE 1 · Instrument both funnels end-to-end · ~20–34h
11	Canonical user_id (T4) + the schema lock — the keystone	Nic + RN	6–10h
12	AppsFlyer conversion events (T5) + setCustomerUserId + SKAN schema	RN, Nic reviews	3–6h
13	Quiz milestone events + in-app paywall events (T6)	Nic web · RN app	8–13h
14	Google brand-pause test (£0) + apply for Google Ads API token	Nic	~1h
15	ASO keyword + listing refresh	Aga	2–4h
PHASE 2 · Fix the biggest leaks + activate A/B · ~18–33h
16	D6 pre-bill nudge — → pulled into Phase 0 as R3 (8 Jul amendment)	Nic + Claude	—
17	Screen-1 proper A/B (goal-question-first)	Claude builds · Nic reviews	4–6h
17b	Entry-offer test: 7-day free trial vs $9.99/30-day (/30trial) vs annual-first — cold traffic's first paid step, the top untested conversion lever	Aga + Claude · Nic ships	3–5h
18	RC Experiments — first app paywall test (injury-aware headline)	Nic	3–5h
19	Exit-intent mid-quiz email capture + sunk-cost anchor at paywall	Nic/Aga	3–6h
20	Checkout Wave-B — Apple/Google Pay, trust-stack, per-day framing	Nic	4–8h
PHASE 3 · Retention (pulled forward) + email migration + paid-restart prep · ~27–46h
21	Trial→paid + lead→trial email nurtures · churn trigger → Retention Plan	Fleet	4–8h
22	Wire 4 RC events → migrate top-3 AC automations → CIO cutover	Nic	16–24h
23	LTV / retention-curve baseline — net of Apple IAP · shared, Retention Plan owns it	Nic/Aga	3–6h
24	Meta CAPI Lead→Purchase + blended-MER dashboard + geo-holdout design	Nic/Aga	4–8h
PHASE 4 · Re-open paid — gated on the holdout · ~6–10h setup + a 4-week run
25	ASA brand defence — small, gated on a competitor actually bidding	Aga	1–2h
26	iOS Meta + ASA category + Google category — small, measured	Aga + Nic	3–6h
27	Execute the geo-holdout — the incrementality verdict	Aga + Nic	2h + run

Reality check: that's ~85–145 hours of build end-to-end, almost all Nic — which is exactly why the sprint is the top ~10 items, and the rest is "not now."

The detail — unfold any task for the "why"

PHASE 0 · Make the numbers real

📋 The full instrumentation spec — every event, exactly what Nic wires, in order — lives in the TMA Analytics Brief. This section is the priority order; the brief is the step-by-step how-to.

#1Wire RC → Amplitude (T2)THE constraintSNic

WhatRevenueCat webhooks → n8n → Amplitude for all five event types (trial_started, initial_purchase, renewal, expiration, cancellation). Dedup key = RC transaction_id. Why this is #1RevenueCat holds ~190 paying subs; Amplitude sees 3–5. Until this reconciles to within 5% of the RC dashboard, lead→trial→paid cannot be computed for either funnel, no A/B result is readable, and Meta keeps optimising for the cheapest email submitter instead of a buyer. Every other item is downstream of this one wire. Metric movedtrial_started / subscription_purchased counts in Amplitude within 5% of RC dashboard, 48h after deploy. NotePre-T4 events land on partly-anonymous profiles — correct with T4, don't block T2 on it.

#2Fix the page-load PIs on the lifetime / promo cart (T3)lifetime cart1–2hNic

Scope — this is the LIFETIME / promo cart, not the evergreen checkoutThe page-load-PI / inflated begin_checkout problem is the July lifetime cart specifically. Evidence: June's evergreen subscription checkout made 169 PaymentIntents at 86% success — clean, no shells; the ~5,000 page-load shells are all the july4 lifetime cart. So the evergreen sub checkout the quiz feeds is already clean on this — its arrive→card-entry is likely already measurable. WhatOn the lifetime/promo cart, create the PaymentIntent on the "Pay" tap (not page-load) and fire begin_checkout only on real intent. PriorityNot urgent this week — that cart closes ~12 Jul. Do it before the next promo cart so that campaign is measurable and doesn't produce another 98%-ghost dataset. (UTM capture on the evergreen flow is the separate #3.)

#3Capture UTM / click-IDs on quiz load → localStorage → Stripe PI → RCMED-HIGHSNic

WhyAttacks the 97%-unknown-source problem directly, and is the prerequisite for both paid-restart fixes (Meta CAPI needs fbclid; Google loopback needs gclid). Metric movedUnknown-source % in Stripe from 97% → below 50% within 4 weeks.

#4Acquisition or retention? — already analysed: it's retentionanswered→ separate plan

AnsweredThe leak is retention, not acquisition. A separate Retention Plan addresses this — it's not part of this growth doc, which stays top-of-funnel (stranger → first payment).

#5Re-fire our existing AC win-back to current lapsed subs✅ DONElive in AC

Status✅ Done — the win-back is live in ActiveCampaign. WatchReactivation rate (RC restart within 14 days) on the daily dashboard, so we can see what it pulls back.

#6Retarget the quiz-email list — ongoing, the proven offerHIGH · always-on2–3h setupNic

WhatEveryone who gave an email but didn't trial: email nurture first (free), then a small warm Meta custom-audience retarget with the matched offer that's already proven to work. Set it up to run on an ongoing basis — every new quiz-email lead flows in automatically, not a one-off blast. WhyThe anti-July — warm audience, right offer. Highest cheap leverage we have, and it compounds: every quiz lead we capture becomes a re-marketable asset forever. Hard gate (added 8 Jul)Bot purge first. The leads list is bot-polluted — seed the Meta custom audience and the nurture from the engaged-only, purged segment (the 00e send-plan rule), or we pay to advertise to scripts and poison the metrics. Metric movedlead→trial from the re-engaged cohort, weekly. Measure the paid slice against a small holdout so we know it's incremental, not just crediting people who'd have converted anyway.

#7Paywall coherence + safe copy (annual-first, remove cold-lifetime)MEDSNic

WhatAnnual-first with per-day anchor ($2.90/day), remove lifetime from the cold surface, one line stating the specific charge date. The checkout order-summary must mirror the paywall exactly. WhyEstablished pricing psychology, not a hypothesis — ships without an A/B layer. A paywall saying "annual, 7-day trial" pointing at a checkout saying "monthly/lifetime £297" actively bleeds conversion today. Skeptic caveatThe autopsy proved COLD→lifetime = 0; it did NOT prove WARM→lifetime = 0. Confirm the paywall can segment cold vs warm, or keep a warm-only lifetime path — don't kill a working warm revenue line. Metric moved% annual at trial start (target >50%).

#8iOS onboarding walkthrough — is AHA before or after the paywall?HIGH-potential1–2hNic

WhatNic goes through the current iOS install→onboarding flow himself (he hasn't yet) — count the screens install→AHA, confirm the AHA moment lands before the credit-card gate, and check whether the injury flag shows up at all. Document it. WhyResearch is clear: fitness paywalls win when placed after 2–4 goal-collecting steps, before the first workout — and our assessment-first model already fits. If AHA sits after the CC gate, every day is a compounding cancellation leak. And whoever's building the app should know this flow cold.

#9Two 10-minute unblockers: RC Experiments tier + AdServices.framework checkenablingSAga/Nic

RC tier (Aga)Confirm the RC plan includes Experiments before Nic builds to it — 15-min check that gates an entire A/B workstream. Skeptic: RC Paywalls/Experiments can be revenue-share priced — quantify the cost against £2.2K app MRR. AdServices (Nic)Check if AdServices.framework is linked (10 min). Present → ASA needs no release. Absent → 1–3 week release lag. Highest info-per-minute task in the plan.

#10AC baseline audit + snapshot all baselines before Phase-0 changesenablingSAga

WhatDocument open/click/best-available-trial-attribution for every AC automation; rank top three to replicate in CIO; drop the zero-impact ones. And snapshot RC/Stripe/AC baselines before deploying any Phase-0 change. WhySets the baseline CIO must beat, and preserves a clean "before" so Phase-0 changes are attributable (skeptic: concurrency pollution risk otherwise).

PHASE 1 · Instrument both funnels end-to-end

#11Canonical user_id (T4) + the Day-2 schema meetingHIGH · keystoneMRN+Nic

WhatOne canonical id across RC, Amplitude, AppsFlyer + web pass-through. Set in the agreed SDK call sequence before the first event fires. The non-optional gateA Day-2 written schema meeting (Nic chairs, RN attends): agree the id (authenticated app_user_id; pre-auth fallback = lowercased SHA-256 email hash so quiz leads join to app users), the exact call sequence (setCustomerUserId → setUserId → RC.logIn), and the web localStorage→checkout pass-through. If this meeting is skipped, T5/T6/T2 data can't be joined — the single largest quality risk in the project. Owner splitNic does the web side + reviews the RN native merge before it lands.

#12AppsFlyer conversion events (T5) + setCustomerUserId + SKAN schemaHIGH · unlocks iOSS-MRN, Nic reviews

WhatWire af_start_trial / af_purchase from the RC purchase callback (not the UI layer). Call setCustomerUserId on account creation. Set SKAN conversion values (cv=1 trial, cv=2 purchase), submit to Apple. WhyThis is what makes any future iOS paid channel measurable (LTV by channel). Without it, Meta/Google get no conversion signal in the SKAN window. Skeptic caveatNic must verify the callback hook location before merge — firing from the UI double-counts and pollutes with failed payments.

#13Quiz milestone events (T6 web) + in-app paywall events (T6 app)HIGHMNic web · RN app

What8 named quiz events (started, per-screen view, path assigned, assessment done, email captured, paywall viewed) + 3 app events (onboarding_completed, paywall_shown, cta_clicked). Instrument the new flow only, not old screens. Batch app events with the RC Paywalls migration to save store cycles. Why#1 gave the funnel bottom; this gives the top. Together they complete both funnels. paywall_shown is the denominator every RC Experiment needs. Metric movedBoth funnels visible end-to-end by source; the two actionable rates (email→trial, paywall→trial) become real.

#14Google brand-keyword pause test (£0) + Google Ads API token applicationfree revenue decisionSNic

WhatAfter T3, pause branded Google keywords 14 days. Flat organic branded traffic → cut the spend (recover ~£100–150/mo). Drop >20% → it's incremental, keep it. Separately, apply for the Google Ads API token now (3–5 day approval lag) for the later loopback fix. WhyHighest information-per-pound test available; the token has a multi-day lag so start it early.

#15ASO keyword + listing refreshMED · compoundingMAga

WhatFill the 100-char keyword field (calisthenics, bodyweight workout, handstand, injury recovery, pull-up, no-equipment…), subtitle featuring assessment/skill progression, screenshots leading with the assessment-first hook. Do this before any ASA category spend — ASA is only as good as the listing it feeds. WhyZero Nic dependency; 3–4 week ranking lag means earlier compounds. Current 4.7★/363 is adequate, not exceptional.

PHASE 2 · Fix the biggest leaks + activate A/B

#16D6 pre-bill nudge (push + email) — the trial→paid leverHIGHMNic+Fleet

WhatDay-6 push (RC native) personalised to sessions/skills + the specific billing date, plus a Day-6 loss-framing email ("Your training continues tomorrow") for web trial starters. WhyIn a card-upfront model, billing-day surprise is the #1 cancellation driver. Highest-ROI single trial→paid intervention, and RC knows the trial date natively (no instrumentation gate). Metric movedD8 trial-cancellation rate. Skeptic: read the ACTUAL current D8 rate from RC before quoting a target — don't assert "15%→8%".

#17Screen-1 proper A/B (goal-question-first) — biggest web leverHIGHS🔒 needs clean data

WhatVariant B = the goal question as the landing screen (large tappable cards, plain-language framing, slim progress bar). GrowthBook or a hand-coded 50/50 split. Primary: landing→quiz_started; guardrail: quiz_started→lead. WhyResearch: first-question uncertainty is the #1 quiz drop. A 10-point lift here compounds through the whole funnel — the single largest web MRR lever. Skeptic corrections① Gate on per-screen instrumentation being live and stable ≥1 week — a start-rate win that tanks lead→trial is a loss you can't see blind. ② Keep "8-minute assessment" (a specific short-time signal reduces abandonment); test the jargon strip inside this A/B, don't ship it blind. ③ Confirm trial velocity supports "200/arm" or run a bigger directional swing instead.

#18RC Experiments — first app paywall test (injury-aware headline)HIGHM🔒 volume-gated

WhatInjury-aware paywall headline vs generic, in RC Experiments (no app release after initial SDK config). Read at 200+ trial starts/arm, 95% confidence. WhyInjury is the top stated joining motivation (46% joined for it) — test whether it lifts trial-start on the paywall. Skeptic caveatThe "5.1×" is a joining-motivation / segment-size figure, NOT a measured paywall lift — treat it strictly as the hypothesis. And check trial velocity: at ~190 subs this test could take months — if so, run a bigger swing, not an underpowered A/B.

#19Exit-intent mid-quiz email capture + sunk-cost anchor at paywallMEDS-MNic/Aga

WhatElevate the "save my progress" soft-save and add exit-intent email capture across the highest-bleed quiz stretch (must ship with honeypot + validation given the bot issue). Plus a post-assessment sunk-cost line above the paywall price selector ("Based on your assessment, training starts at [Level X]"). WhyGrows the lead numerator by recovering email-less abandoners; the sunk-cost anchor reframes card entry as continuing, not starting. Confirm assessment data reaches the paywall screen first.

#20Checkout Wave-B — Apple/Google Pay, trust-stack, per-day framingMEDS🔒 after 2wk baseline

WhatThe measurable checkout CRO changes, shipped as a dated, discrete wave with a "before" snapshot — after ~2 weeks of clean arrive→card-entry baseline. WhyDirectionally safe but they're optimisations you must be able to measure. Shipping them into the Rank-2 window contaminates the baseline.

PHASE 3 · Retention engine (pulled forward) + email migration + paid-restart prep

#21Trial→paid + lead→trial email nurtures (pre-payment) · churn trigger → Retention PlanHIGHS-MFleet

WhatSingle coach-voice email at the 14-day no-workout mark for paying subs (with a holdout group configured before first send). Plus the trial→paid 7-email sequence (fires on trial_started, NOT signup date) and the lead→trial 3-email nurture. Why (I pulled this earlier than the fleet had it)MRR is declining and June's drop was retention — catching at-risk subs pre-cancel is far cheaper than win-back after. Gated only on the 4 RC events (#22) being wired. Metric moved30-day churn of intervention vs holdout; trial→paid by email-engaged cohort. Note for Nic — some churn is bug-drivenA chunk of the cancellations are tied to app bugs, not just disengagement — so part of retention is a fix problem, not an email one. The RN team may be needed to clear those bugs alongside these emails. Flagging here so it's on the radar; the funnel work stays the focus.

#22Wire 4 RC events for CIO → migrate top-3 automations → full cutoverHIGHL🔒 event + deliverability gate

WhatWire trial_started, subscription_cancelled, workout_completed, trial_converted through n8n (same fan-out as Amplitude). Then replicate the top-3 AC automations in CIO, run parallel ≤2 weeks with a hard cutover date, then full cutover. WhyCIO's whole value over AC is firing on product behaviour — without these events it's "AC with a higher invoice." The event wire is the gate, not a parallel task. Deliverability gateBefore cutover: email deliverability / IP-domain warm-up + bot-list purge. DMARC is already fixed ✓ (SPF/DKIM/DMARC now aligned). A cold cutover can still tank inbox placement, so warm the domain and purge the bot-polluted list first — this engine is what defends MRR.

#23LTV / retention-curve baseline — net of Apple IAPshared · Retention PlanS-MNic/Aga

What (skeptic-added)Establish the real retention curve and margin-adjusted LTV — net of Apple's 30%/15% IAP cut and realistic renewal decay (£157 gross ≈ ~£110 net year one). WhyBoth paid gates ($20/trial and MER≥2×) are untrustworthy without it — you can hit <$20/trial and still lose money. This is the darkest area and must be a deliverable, not an assumption.

#24Meta CAPI repoint Lead→Purchase + blended MER dashboard + geo-holdout designHIGH · paid prepS-MNic/Aga

WhatWire RC INITIAL_PURCHASE → Meta CAPI as Purchase (event_id dedup), demote Lead. Extend monday_brief.py with blended MER (net-new MRR ÷ total spend). Write the geo-holdout design (one comparable geo pair, 4 weeks, measure new RC subs/1k population). WhyThese have no value until paid restarts but must exist before it. The MER is the attribution-agnostic cross-check that can't be gamed by a platform's own reporting.

PHASE 4 · Scale paid (web quiz first) + prove incrementality

#25ASA brand defence — small, gated on a competitor actually biddingS · gatedAga

WhatOne exact-match brand campaign (~£8/day) — only after a 10-min App Store brand search confirms a competitor is bidding on your terms. Accept install-only measurement, labelled as defence value, not acquisition ROI. Skeptic gateIf no competitor ad shows, defer — it's harvesting installs you'd get organically. And fold this spend inside the spend-freeze sign-off; nothing is exempt from the freeze without your written OK.

#26iOS Meta + ASA category + Google category search — small, measuredM · gatedAga+Nic

WhatPurchase-optimised iOS Meta at ~£30/day (read D7 AppsFlyer, never D0 Meta), ASA category, Google category search — each opened only as its gates clear, one channel proven cold before the next. GatesT1+T4+T5 confirmed green ≥7 days · web Cost per Trial is a real number · matched offer (trial, not lifetime) · all CAC math net of Apple IAP. Proof-first, not web-firstWhichever funnel first shows a confirmed Cost per Trial under ~£16 earns the next budget increment — web or app.

#27Execute the geo-holdout — the incrementality verdictthe gate on scalingMAga+Nic

WhatRun paid in the test geo only, 4 weeks, identical organic in the holdout geo. If new MRR/1k population isn't statistically higher in the test geo, paid is not incremental — pause and diagnose before any budget increase. WhyThis is the answer to Nic's attribution debate — you don't win it with a model, you settle it with an experiment. Explicit stop-loss: if a small test install batch produces zero confirmed trials, halt.

§5The daily dashboard — you + Nic, every morning

Extend tools/finance/monday_brief.py (it already pulls RC v2, GA4, Meta) — add AppsFlyer + Stripe, emit the same HTML template, deliver 06:30 to you both. Every row carries a confidence badge; the dashboard graduating from mostly-red to mostly-green is itself the visible measure of instrumentation progress.

Section	Rows	Visible today?
★ North Star	Total MRR (WoW) · new MRR this week · active paying subs · Lead→Paid by source (gate) · trials this week · trial→paid D7 (retention signal, not the gate) · Meta CAPI match quality	Now (CAPI row lights up after T1)
Funnel 1 · Web	Quiz sessions · real card-intents · succeeded charges (£) · source breakdown · checkout→subscriber rate	Partial intents/rate go red→green after T3; source after T4
Funnel 2 · App	Installs · trial starts · trial→paid D7 · Amplitude event-health · D7 workout retention	Partial retention after T7+approval
Spend Governance	COLLAPSED / PAUSED	daily spend · blended MER · Cost per Trial · Meta D7 ROAS — activates only on your written go

Skeptic fixes baked in: RC endpoint is v2 /projects/{project}/metrics/overview (already in monday_brief.py) · tag iOS installs + iOS ROAS as MODELLED (SKAN buckets, delayed) · net-of-refunds on £ rows (standing NET rule) · confirm the launchctl cron is actually loaded before promising 06:30 delivery · GA4 depends on the existing service-account credential.

§6Tooling — decided, so they stop resurfacing

Question	Decision	Why
App paywall A/B	RevenueCat Experiments — Adapty closed	RC already integrated for 190 subs; Adapty = parallel SDK + subscriber-state migration + 2–4 wks + a release, to run tests RC templates already cover
Web A/B	GrowthBook (free) or hand-coded split	No new paid tool; instrumentation is the prerequisite either way
Email platform	AC → Customer.io, gated on the event wire + deliverability warm-up	CIO's value is behavioural triggers; without events it's AC at a higher invoice
Intercom	Already in use — keep & leverage	We already pay for Intercom and run in-app messages through it. Keep it; wire the new event stream in so those messages fire on real product behaviour.
Mobile attribution	Keep AppsFlyer, wire it correctly	It's the MMP (device-level install attribution) — Amplitude can't replace it; the problem is wiring, not the tool
Event pipeline	n8n fan-out	Already in the stack; one RC event → 4 destinations in an afternoon vs a sprint of custom code

§7What NOT to do

Don't scale paid before both funnels are instrumented AND a geo-holdout proves incrementality.
Don't migrate to CIO before the 4 RC events are firing (and deliverability is warmed).
Don't add Adapty. RevenueCat Experiments covers app paywall A/B with no migration. (Intercom we already run — keep it, wire events into it.)
Don't ship the Screen-1 copy change blind — test it inside the A/B; keep "8-minute assessment".
Don't report begin_checkout as a conversion metric until T3 (it's ~75× inflated).
Don't serve lifetime to cold traffic — warm-list only, with the segment defined and locked.
Don't rebuild the checkout until the arrive→card-entry data says the page is the leak.
Don't run two paywall variables at once, and don't call any winner before it's actually powered.
Don't promise "10 days" until the RN team produces a TestFlight build from the prod branch.

§8What I need from you to execute

Confirm the RN-team logistics if we want the native build in parallel: can they build from our production branch quickly, and their hours + rate for T4–T7.
Two 10-minute checks (Nic): is AdServices.framework linked? · is a competitor bidding on our App Store brand terms?
Near-term MRR defence: since paid is ~10 weeks out, pick the one retention/win-back action that moves MRR in the next two weeks — so we're not just instrumenting while MRR slips.
The focus call (from the founder coach): the top ~10 items are the sprint; the rest are "not now." Decide what stays frozen so this doesn't become another ten-front week.

Nic's already executing the top of this — we work down it one item at a time. "Done" = a number moving on the daily dashboard.