An app with a 4.8 rating
and a funnel that lost money on every user
For an AI market-analysis app (46 tools, App Store 4.8) we ran Meta and Google Ads across four countries, built a unit-economics model on real data, audited the funnel from session recordings and delivered a 14-page developer brief plus a lead magnet. The numbers showed exactly which churn reduction would make the subscription pay.
The situation
A polished app: 46 analysis tools across crypto, forex, gold and indices, 4.8 in the App Store, five languages. Paid traffic was running, installs were growing, and nobody had checked whether a subscriber ever paid back the cost of getting them.
What we did
Compliance and keywords before spending. Google’s financial-advertising rules differ by country, so eleven research agents mapped the verification gates per market. Keyword research found that Malaysia searches for a local exchange screener (46,000 searches a month) rather than “AI stock analysis”, which changed the whole campaign for that market. A shared list of 81 negative keywords cleaned the Google account.
Campaigns. Meta Ads with 21 campaigns moved from install optimisation to subscription and purchase events, SKAdNetwork configured for iOS, Google App and Search campaigns. 1,039 installs in four months across Canada, Malaysia, New Zealand and Mexico; one campaign brought installs at $3.27.
Unit economics on real data. Ad spend, installs, purchases and subscriptions joined in our own warehouse: 48 purchases, 29 active subscribers, churn 34.5% a month, subscriber lifetime 2.9 months, CAC $72, LTV $44, LTV/CAC 0.61 against a healthy 3. By country: Malaysia 0.66, Canada 0.51, two markets with spend and zero purchases. The model showed that churn down to 10% would give LTV $149 and LTV/CAC 2.4: the one lever that matters.
Funnel audit from recordings. A team of agents and a human went through session recordings, onboarding, retention and performance: day-1 retention 17.8%, day-7 6.5%; mobile largest contentful paint 3.0 to 4.2 seconds; 39 people a month failing Google and Apple sign-in; zero successful MetaTrader connections out of eleven attempts in 30 days, which meant the broker feature was effectively broken; six of eleven site pages not translated despite twelve languages advertised; a paywall converting at 13 to 15%.
Deliverables. A 14-page brief for the developers ordered by impact, a lead magnet (a guide for one market with its own capture flow), a quarterly roadmap with decision points, and a model the founders can rerun.
The result
The product team got a precise list instead of opinions, the ad budget was redirected from the two zero-purchase markets, and the business decision on the subscription versus partnership model could finally be made on numbers.
FAQ
Why publish a case where the economics were negative?
Because finding that out, with numbers, is the service. The app had great ratings and growing installs; the model showed that each subscriber cost $72 to acquire and brought $44 over a lifetime. The brief lists what changes that. Hiding it would be the real failure.