ebselman
commented
1 year ago To dig deeper into this issue at Criteo we performed a study on the 6 product categories that generate the most publisher revenue and studied the probability that a given audience comes back to visit a product category after they already saw that category x days ago. This is what we call the relationship between recency of the signal and the relevancy of the audience towards a given commerce category.
As suggested in a recent article published by Criteo regarding our testing of the Topics API , our study shows a negative relationship between recency and relevancy for all the studied categories. On average the relevancy decreases by 15% after 7 days and by 35% after 21 days.
This may suggest that having smaller epochs lengths in topics, may indeed improve the relevancy of the signals, and therefore the utility of the API.
Topics signal can be based on up to three weeks old data, which is quite an old signal for web advertising, as users buying patterns are usually much shorter than that.
Reducing the period on which topics are being assigned, from one week to 2 or 3 days, should improve the way topics can inform marketers of the users buying intents.