Optimizing Discovery: How Great Games Reach Millions of Players on Roblox
Refining Discovery Signals to Reward Long-Term Retention
Great games should be easy to find. With millions of games and 132 million daily active users on Roblox1, connecting players with games they’ll love—and giving creators the transparency they need to make good development decisions—is more important than ever.
That’s where Roblox’s discovery systems come in. Our Recommended For You algorithm surfaces relevant games on Roblox’s homepage based on objective metrics and each player’s unique preferences. When players come back to a game over time, bring their friends, and spend money, that’s strong evidence that the game is delivering a meaningful experience that more players are likely to enjoy too.
Today, we’re expanding the Recommended For You algorithm from its previous 7-day view to a 28-day view that directly measures long-term retention. We’re also sharing the full list of signals and their relative importance with creators so they understand how their games are surfaced by the algorithm. In testing, the updated Recommended For You algorithm did a significantly better job surfacing games that retain players over time, leading to an increase in daily active users and engagement across the platform.
Algorithm Transparency: The Signals Powering Recommendations
When a player comes to the Roblox homepage, Recommended For You is designed to surface games they’re likely to enjoy long term rather than games that win a quick click but don’t offer deeper substance.
Long-term retention was always an important signal in our algorithm, but we previously used 7-day proxies to get an early signal of retention. We've found in testing that if short-term engagement is overvalued, the system could disproportionately favor games that win attention with exciting thumbnails but don’t deliver long-term value for players. These types of games were displacing games that players value for the long term, actually harming long-term retention and engagement.
While early retention signals are still important, our new signals cover day 1, day 2–7, and day 8–28 time periods to capture a clearer picture of player behavior, and allow us to properly prioritize long-term value. We’re also breaking our previous qualified play-through rate signal into more granular signals that allow us to measure play-through behavior, session quality, and spend separately. See more details about the changes in our Devforum post.
We give creators visibility into our signals and their relative importance as well as their own individual game performance data through the Creator Analytics dashboard so that they know where their game is strong and where it could improve.
