TrainerRoad · 2020–Present
TrainerRoad AI
Took TrainerRoad from reactive to predictive. Simulation engine that runs hundreds of scenarios to find the right workout. Close to 100k athletes use it.
Challenge
Adaptive Training was reactive. It adjusted after the fact. That was a massive step forward, but we kept asking: what if the system could see ahead? What if athletes could preview how a schedule change today affects their fitness in four weeks? We needed to go from reacting to planning with foresight.
Approach
I ran the product strategy and led the team through the shift from reactive to predictive. The core of TrainerRoad AI is a simulation engine that runs hundreds of scenarios on power, heart rate, and RPE to find the right training path. Athletes get a 4-week Training Simulation Window, Predicted Difficulty before every workout, AI FTP Detection without testing, and what-if scenarios before committing to any change. The system plans ahead instead of cleaning up after the fact.
Impact
Close to 100k athletes use it. The shift from reacting after the fact to planning with foresight is the biggest upgrade we shipped. Athletes train with data, not guesswork.
Reflection
ML products are not engineering problems first. They are trust problems. Athletes had to trust that an algorithm knew better than their coach or their gut. Building that trust required transparency, gradual rollout, and paying close attention to the moments where the system got it wrong.