Criteo’s Overhauled Prediction Engine Now Factors Actual Conversions, Not Just Clicks
Personalized display advertising firm, Criteo, today announced a major update of its prediction engine, three years in the making.
The Criteo Engine has been retooled to automate impression bidding based on the likelihood that a user will both click on an ad and make a purchase. Up to now, bidding was based on click predictions.
Criteo says the engine’s ability to close the prediction loop through to purchase has lifted sales volumes 38 percent while maintaining a stable cost-of-sale in A/B testing across billions of impressions and millions of users and clicks.
“We’ve solved a lot harder problem. The new engine predicts not just the click but whether the person is going to convert on the site,” Jonathan Wolf, Criteo’s Chief Product Officer, told me by phone. “In order to predict the one-in-ten-thousand impressions that would give the conversion, you need more data and a better algorithms … This new engine is a result of years of work. We dramatically improved the technology to do more volume at the same level of cost-of-sale and profitability targets.”
The re-engineered Criteo Engine can process up to 15 million predictions per second and respond within 20 milliseconds. Criteo says it now stores more than 20 Terabytes of data daily, including actual purchase behavior.
The engine influences what products are shown in retargeted ads, says Wolf, and ads can be displayed across all networks including mobile web and in-app. Most current advertisers will be transitioned to the new engine by the end of the quarter.
Top news and emerging ideas in display advertising, delivered every Monday.