TL;DR
After three acquired companies merged into one product, the homepage — where every multi-million-dollar service starts — saw conversion fall. I led an AI-personalized, performance-tuned homepage that drove conversion to an all-time high across 24M members.
The Problem
The newly unified homepage was the digital front door for the whole portfolio, but it had no real owner and conversion was sliding. Diagnosis with engineering, data, and research (plus competitive mystery-shopping) surfaced three issues: latency was causing drop-offs, members were overwhelmed by service choices and unclear messaging, and the assistant looked like basic tech support.
The Approach
I co-authored the strategy rather than selling it — elevating to org-level KPIs so six vertical product teams each owned a piece, instead of defending their own real estate. I sequenced the work to build trust: the latency fix first (no alignment needed), then dynamic personalization, then the agentic assistant rebuild. I navigated real pushback — vertical PMs worried about losing visibility, legal nervous about PII through LLMs — with guaranteed top slots for cash-cow products, a revenue-weighting coefficient when scores tied, HIPAA-compliant infrastructure, and cohort-based rollout.
The Solution
For relevance, I used two recommendation approaches together — collaborative filtering (learning from similar members' choices) and attribute-based filtering (learning from a member's own history, biometrics, and unfinished journeys) — to reorder cards per user. For performance, I partnered with engineering to prioritize what loads first: lazy-loading and CTA-first rendering on the front end, and caching plus breaking a monolith into parallel microservices on the back end, so members could start care while secondary data loaded behind it.
The Impact
Conversion hit a record 3.7% (up from 3.2%) across a ~24M active population — a ~$131M revenue impact — and the personalization framework became a reusable primitive for the portfolio. The key lesson I carry forward: in ambiguity, the temptation is fast tactical fixes; the real value comes from slowing down, then moving with conviction backed by data.