Pinterest Acquires Machine Learning Commerce Recommendation Engine Kosei
Facebook knows who you were. Google knows what you want now. But Pinterest yearns to know what you want next, which is “exactly” why it acquired recomendation engine startup Kosei, Pinterest's head of engineering Michael Lopp tells me.
Context & Ripple Effects
In January 2015, Pinterest bought Kosei to close what its head of engineering Michael Lopp framed as a gap between rivals: Facebook knows who you were, Google knows what you want now, and Pinterest wanted to know what you want next. The acquisition put an in-house machine learning commerce recommendation engine at the center of a company whose entire product is a feed.
The bet paid out visibly across the following years' coverage: within months Pinterest shipped its buyable pins and recommendations platform, by 2019 it made merchant-facing shopping tools like Catalogs generally available, and by 2021 the neural network curating pins alongside tailored ads was drawing 480M people to the site in a single quarter.
First-order effects
- Kosei's recommendation technology moves inside Pinterest's engineering org under Michael Lopp, giving Pinterest native ML ranking for commerce content instead of licensing or bolting on third-party ad-tech.
- The deal sharpens Pinterest's positioning against Facebook and Google as a distinct advertising surface: predicting future purchase intent rather than past identity or immediate search demand.
Second-order effects
- Merchants gain a direct pipeline into Pinterest's feed — a path that runs through buyable pins and later Catalogs, where personalized recommendations decide which products get surfaced.
- Ad buyers evaluating Facebook and Google now have a third option priced on predicted intent, pressuring the two incumbents to defend their own recommendation quality on shopping queries.
Third-order effects
- The pattern points toward consumer platforms treating owned recommendation models as core infrastructure: the Kosei team's work matures into the neural network that pairs curated pins with targeted ads, making inference capability inseparable from monetization.
- If the arc holds — visual search guides, a Messenger search bot, and catalog-scale personalization all followed — discovery platforms consolidate around whoever owns both the content graph and the model ranking it, leaving point-solution recommendation vendors squeezed between platform buyers.
The trend: Consumer internet platforms are absorbing machine learning recommendation startups in-house, converting prediction of user intent into proprietary commerce and ads infrastructure.