Kurly

At Kurly, I worked on recommendation and growth products during a period when acquisition costs were rising and new user growth was beginning to slow.

Rather than focusing only on bringing in more users, the team started asking a harder question: How do we increase value from the users who are already here?

I treated the recommendation tab almost like a laboratory. Instead of redesigning the entire homepage, I ran continuous experiments inside smaller recommendation surfaces, testing ranking logic, merchandising structures, thumbnail sizes, duplication policies, and behavioural signals.

Over time, I led more than 20 A/B tests across recommendation experiences.

One project involved building an automated merchandising system based on repurchase probability, allowing high-value products to surface more dynamically without manual curation.

Results: · Recommendation ARPU +77% · Duplicate product exposure -73% · Experimentation framework adopted by other teams

But more importantly, this was where I learned how much product work is really about uncertainty management.

Most experiments failed. Some produced results nobody expected. Occasionally, tiny interface changes shifted user behaviour more than major strategic discussions.

Kurly was where I became deeply comfortable with experimentation, ambiguity, and data-informed product thinking.