AI and Growth in Regulated Markets

Context: on Meta, credit/fintech ads can't use lookalikes, age, gender, ZIP or specific interests. Funnel was install -> sign-up -> KYC -> bank link -> first transfer (revenue event). Sign-up was too loose (people gaming cashback/referral rewards); first transfer was too deep (three high-friction steps). At sign-up, data was routed to the internal database and the first and last name run through a name-origin API; anyone with 70%+ probability of being from South Asia (the Indian expat target) triggered a synthetic 'HQ' event sent back to the ad network. No financial data or PII was sent. The event met the three criteria: enough volume (Facebook 50/7 days, Google 10/campaign/day), recency (within 24-48h) and a good proxy for first transfer. Guest notes such scoring could now be vibe-coded instead of using an API.

For a 7-day free trial that auto-renews into a subscription, analyzed how soon after starting the trial people cancel: 25-30% cancelled within the first 12 hours. Instead of firing trial-started immediately, sent an adjusted trial-started event to Meta after ~12 hours (host: '12 hours and 5 minutes') only for users who hadn't cancelled, so the algorithm finds users more likely to pay. Done around 2023 or earlier; guest says it was an industry first then and is now commonplace.

Telehealth intake/qualification forms run 25-35 steps, mostly mandatory. Meta does not allow purchase as a feedback event for this category (close proxies like a checkout-page view are allowed). So start by optimizing to 'first question answered'; once there is enough data, move to 'five questions answered'; then to 'email entered'. Accept a 'learning cost' paid to Meta for any new product.

Health data must never reach the ad network. Standard Facebook events are not used; only custom events. Before events are sent back, name, email and medicine information are stripped so no patient data is passed.

Works only with US-based pharmacies and got LegitScript approval, which lets a prescription telehealth company advertise on digital ad platforms.

Built three AI skills from outside research and own experience that review all messaging: (1) platform guardrails - does the channel's policy (Meta, Google) permit it; (2) category rules - e.g. never call compounded drugs FDA-approved; (3) operating reality - claims like 'same-day shipping' must match what the pharmacies deliver. The skills flag issues; a human reviews each flag and either changes the copy or marks a false positive and runs as is. Run before uploading to the ad manager to avoid Facebook disapprovals and account suspensions.

Anything after someone becomes a patient lives in the EHR with its own lifecycle tool (e.g. 'your refill is coming up'). Marketing lifecycle (e.g. a nutrition guide) runs in a separate tool with no patient information and must not mention the medication (e.g. tirzepatide) as the reason for the message. A controlled data boundary keeps PHI from crossing over.

Using Codex agents that operate their own computer: 'go to these five competitors, fill out their medical intake form and funnel, fill out mine, tell me where I do better and give your top five recommendations to reduce friction', then have the AI implement them. Next step: connect real data (Google Analytics, AppsFlyer, PostHog, free Microsoft Clarity heatmaps and recordings) to the agent and ask which step loses users, what they scroll on, what to double down on or remove.

Originally a 7-day trial on the yearly plan: people started trials but nobody converted. Moved the trial to the weekly plan as a 3-day trial ($6.99/week). Host recommends the same to an audited app whose weekly plan sells best: 3-day trial on weekly, yearly by itself.

A/B test against $6.99/week with a 3-day trial: variant shows $9.99/week with a 3-day trial plus a $50/year plan with no trial. Users seeing '$10 a week' opt for $50 a year. Paywall built in-house, tested via RevenueCat Experiments.

Created a website for his app with roughly a 10-minute AI prompt and started running Google Ads to it to drive more app installs and conversions.