Supriya Subudhi
Product Manager at Bumble
About Supriya Subudhi
Supriya Subudhi is a product manager at Bumble, the dating app with over 20 million monthly active users. She worked on Bumble's AI photo feedback feature, which gives users AI-generated suggestions for improving the photos in their dating profiles.
In her AI Engineer talk, she explains why a positive A/B test is only the start for generative AI features. She covers five problems: adoption, results in the long tail, cost, model dependence and regional regulation.
Career Highlights
- Product manager at Bumble working on generative AI features
- Speaker at AI Engineer (2026) on A/B testing AI products
Notable Positions
On Stated Demand vs Real Usage
About 78% of users shown examples said they wanted AI profile feedback. Only 10% to 15% of those who got the feature used it. Adoption improved after the feature was placed inside existing journeys, such as adding photos.
On Reading Real Outputs
Positive averages hid bad suggestions, like telling users who were already smiling to smile more. Bumble added guardrails and evals and measured results by segment, including gender, language and country. See Agent Evaluation.
On Cost and Model Dependence
Generative AI costs grow with every use. When Bumble switched from the best model to a cheaper one, accuracy and impact dropped, which showed that the A/B result came from the model rather than from the feature concept.
Key Quotes
- "Openness to AI doesn't actually predict whether people are going to use the feature." (on adoption)
- "A positive result from an A-B test is actually just where the journey begins." (on AI product testing)
Related Reading
- Agent Evaluation - Reading real outputs instead of averages
- Model Routing - Matching model cost to the task