AdPanel: fixing activation for an AI UGC ad platform

AdPanel | Product UX for an AI UGC ad platform (Activation & Retention)

AdPanel turns a product photo into a ready-to-run UGC video ad: drop your product image, and an AI creator presents it in a TikTok-native vertical video with voice and motion. I joined as embedded product UX for activation and retention, working directly with the founder on an ongoing weekly contract: my job is to find where users leak out of the funnel, fix it, and prove the fix with data.

The product was capable but the funnel was invisible and leaky. There was no analytics foundation, so every decision was a guess. New signups were redirected to a payment form seconds after registering, before seeing anything the product could do. And the platform's core action, rendering a video, silently failed 64% of the time: renders hung on an infinite spinner with no error, no email, and no retry, because the whole pipeline depended on the user keeping their browser tab open.

Built the analytics foundation from zero in PostHog: 15+ product events, conversion funnels, dashboards, deploy annotations and internal-account filtering, so every change ships with a measurable before and after

Diagnosed the render failures with data (event analysis plus database forensics), found the root cause, and rebuilt the pipeline: server-side reconciliation crons that finish renders even with the tab closed, a fallback cascade across three AI video models, render-complete emails, and a safety-net sweep. Failures went from 64% to zero, and the fix absorbed a 3x traffic spike its first week

Redesigned the top of the funnel: a homepage where visitors drop their product image before signing up (the image survives signup and lands preloaded inside), and an open gallery anyone can browse without an account

Shipped the retention sprint: an automation onboarding wizard with a post-first-ad nudge, activation events that measure real usage instead of saved drafts, and classification guards so paying customers can explore freely

Designed and built a preview-first paywall: new users watch their own product being placed in a creator's hands (a real frame of their ad, generated in seconds for cents) and the payment form renders next to that preview via embedded Stripe checkout. After paying, the video starts rendering on its own. Live behind a feature flag, ready for a 50/50 A/B against the old gate

Delivered weekly data-driven reports to the founder: what changed, what the numbers say, what's next

Video render completion went from 36% to 100%, sustained across weeks including traffic spikes

The funnel went from unmeasured to fully instrumented: we now know 40% of visitors who drop a product complete signup, and exactly where the rest leave

18 real paid subscriptions in a single week on the newly measured funnel, giving the founder a clean baseline before launching paid campaigns

The paywall experiment is built, validated end to end, and ready to flip on, with a 15% gate conversion baseline to beat

The engagement is ongoing: next up is running the A/B and scaling what wins.