Ergatta - Coach AI
The Opportunity
This started off as the founders pet project that had the potential for exclusive intellectual property, positioning Ergatta as a leader in the fitness industry. The hope was that this positioning in the industry would translate to increased growth and win rate against our competitors. Rowing is a pretty niche exercise modality. As Ergatta transitioned from early adopters with rowing experience, to the mainstream members with no rowing experience. This created a new education and support gap in our experience.
The Audience
Members who are new to fitness or new to rowing were concerned about their potential for injury without proper form.
The Role
Director of Exercise Science:
I acted as the biomechanics expert leading Research and Development with our Data Science Engineer.
Led a mass data collection effort for capturing rowing technique at scale across a variety of body types.
Led a strategic pivot shifting our machine learning model from power curve data to computer vision data.
Led model development with our Data Science Engineer.
Led model validation with testing and spot checking model outputs against real world rowing inputs.
Transitioning from R&D to product execution, I acted as a consultant for the product team.
I gave the product manager enough context to run with for implementation of the feature set.
Key Performance Indicators
Success Criteria
Users are able to identify and address their issues via companion app form analysis functionality
Marketing is able to leverage the feature for marketing purposes in winning competitor match-ups leading to new rower sales
Metrics
Feature usage
X% of companion app users use form analysis
X% of users use form analysis multiple times
Form improvement
X% of users see a reduction in form issues over time
The Outcome
Increased TAM and our head to head comparison with competitors, but did not move the needle on new customer growth.
Added perceived value to our paid subscription
We did see an initial usage spike at launch, with quick drop off.
Adoption = 2%
Repeat Usage = 30%
Repeat users saw their Average Form Score increase from an initial 85% to 95%
Lessons Learned
The founders were very attached to this project and with sunk cost fallacy were determined to make it work at all costs. Short term sales growth was made the priority over long term retention, but did not move the needle for the business.
Power curve data is noisy and led to a lot of conflation among different form errors. Initially we thought maybe we just need more data and better quality data, but in the end we found power curves weren’t a great proxy measurement for biomechanical feedback in our application.
Pivoting to computer vision ended up saving us like 6 months of R&D time.
Key Skills
Product Management for Machine Learning
Data Collection, Tagging, Cleaning
Biomechanics
Computer Vision