Researchers describe a system Roblox now uses to shortlist which ad creatives move on to its per-user ad ranking step. It estimates performance using data shared across audience segments, items, campaigns and creatives, so a new creative can draw on information from its parent groups. A single randomized score gives less-tested creatives a chance to be selected and learned about.
In a replay of Roblox Sponsored Ads traffic, the shortlist raised predicted play-through rate by 8.14%, compared with 2.80% for an equally sized random shortlist. The researchers say it forwarded more than 40% fewer creatives than the full pool and added 0.2% to serving time; the paper also says Roblox deployed it after a live A/B test.
Why it matters
For Roblox advertisers, the system may help newer ads get tested while reducing the number of creatives sent to later ranking; the paper does not report the live A/B test's results.
