Consumer technology · Recommendation systems
App recommendations for a mobile app store
Recommendations for about 200M mobile devices
Data points processed every day
Result 1B
The problem
An app store is only as useful as what it puts in front of each person. With a large catalog, showing everyone the same lists left most of it undiscovered.
What we built
We built the recommendation systems on the store's own usage data, with data engineering and data science at scale: collaborative filtering for personal suggestions, similar-app models built from app descriptions and behavior, and 30 variations of the recommendation algorithms tested against each other. The pipelines ran as daily jobs on the client's infrastructure, and we handed them over with a full knowledge transfer.
What changed
- Recommendation systems serving about 200M mobile devices
- 1B data points processed every day
- 30 variations of the recommendation algorithms tested
- Personal and similar-app recommendations refreshed daily in production
- Pipelines run on the client's own infrastructure
- Ran daily until the 2023 handover, with knowledge transfer to the client's team
Have a problem in this shape?
Tell us what you are trying to win. We will map your version of this build in one call.