20+ years growing products millions of customers keep using.

Turning retention into one adaptive system

I lead a team of three product managers covering both sides of churn: customers who choose to cancel, and customers who disappear because a payment failed and never got fixed. The work runs across North America, Latin America, Europe and Asia Pacific, on every platform Max is sold on.

• Replaced static save-offer rules with a learning system. Offer and content in the cancel flow are now selected by predicted customer value rather than a fixed rule table, with reinforcement learning closing the loop.
• Moved payment failure from reactive to predictive. Models identify accounts at risk before a charge fails, and reach those customers across email, push, in-app and connected TV, so the first thing a customer hears isn't that they have lost access.
• Ran retention as one orchestrated layer across product, payments and ML teams on shared metrics, instead of as disconnected interventions owned by whoever happened to touch the customer last.

Personalization and search for the tiers most customers actually use

I led teams of engineers and applied scientists on discovery for Amazon Music's Free and Prime tiers, the ones that come with a Prime membership rather than a separate music subscription. The work covered iOS, Android, desktop, web, Fire TV and Alexa devices.

• Ranking and selection for the home experience. Which playlists a customer sees, in what order, and how quickly their listening actually changes what's shown.
• Search relevance under real constraints. Free and Prime are built around stations and playlists rather than unlimited on-demand play, so search has to satisfy a customer hunting one exact song and a customer who will happily take a good shuffle, without punishing either. That constraint is the whole problem, and it's what makes the work different from search on a full-catalog service.
• Scale that changes the math. Free and Prime reach everyone who already has a Prime membership, so a small improvement in ranking or search is worth a great deal, and any friction you add is expensive at the same rate.

Also at Amazon: a new ad surface, zero to one

Before Music, I launched display advertising for Alexa multimodal devices, the ads that run on the Echo Show home screen. Advertisers supply components (image, logo, headline) rather than finished creative, and the device assembles the ad. It's proximity-aware: beyond about four feet it renders full-screen in rotation with weather, recipes and news; up close it collapses into a card in the home-screen grid. The surface now runs across five regions and roughly two dozen countries →

One listening product across radio, podcasts and playlists

I led six product teams across personalization, subscription, search, live radio, playlists and podcasts, on iOS, Android and web, for a platform with more than 150 million registered customers.

• Shipped the features and the discipline behind them. Defined the customer segmentation the team planned against: by engagement, by content type, by platform, by customer tier. Drop-off analysis then decided what got fixed first.
• Reorganized the teams around products instead of platforms. Engineering had been split by iOS, Android and web, so a single feature shipped three ways on three timelines. I moved the group to project-based squads, corrected a badly skewed PM-to-engineering ratio, and put a regular release cadence and feature flags in place. Delivery got faster and far more predictable.
• Built iHeartRadio into a top-three podcast platform. Took podcasts from a secondary feature to a full catalog alongside live radio, artist radio and playlists, with hundreds of thousands of titles.