Byte
Reducing Case Times by 50% by Training an ML for Image Detection
AIProfitabilityMobile App
Background
With a high-performing treatment companion app already playing a major role in reducing servicing costs, it was clear that the gap to get Byte to as profitable as possible would require a thoughtful approach to bringing AI into the app ecosystem. We needed to train an ML on fit assessment to create a truly magical UX in the palm of our users' hands.
What I Did
- Aligned stakeholders across engineering, data, design, regulatory, legal, marketing, sales, and clinical support on the scope and ML requirements.
- Managed the internal tooling and processes required to effectively annotate 600k photos that would be used to train the model.
- Ran user acceptance testing sessions both internally and in customer interviews, distilling valuable insights that translated to actionable adjustments in the end feature.
- Worked with marketing, data, support, and engineering to ensure we correctly cohorted users during a phased rollout.
Outcomes
- Reduced CPC (cases per customer) by 15% in the cohort of users with access to the fit scan feature.
- Clinical team approval of initial ML recommendation rose to over 80% during first phase of rollout.
- Cases derived from fit scan users were closed at a 51% faster rate.
- Users with the fit scan flow had over 40% less dropoff than users in our traditional flow.