Marsh
Mallow, Roblox changed the "Recommended For You" algorithm. That's the discovery feed, right? Why should a brand care which games get recommended?
Mallow
Because Recommended For You — RFY for short — is the personalized sort at the top of every player's Home page, and it's the single biggest organic discovery surface on Roblox. It decides which experiences get pushed in front of which players. If your experience climbs that feed, you get free, high-intent traffic. If it doesn't, you're paying for every visit. So a change to how RFY ranks is a change to your acquisition economics.
Marsh
Okay. So what actually changed?
Mallow
The core of it: Roblox expanded the evaluation window from 7 days to 28 days. The old algorithm largely judged you on the first week of a player's relationship with your experience. The new one watches a full four weeks. The whole point is to reward experiences that keep people coming back over the long term, instead of ones that spike on day one and then evaporate.
Marsh
What does it look at across those 28 days?
Mallow
It splits the month into three buckets — Day 1, Day 2–7, and Day 8–28 — and measures familiar signals in each: playtime, play days, qualified play sessions, intentional co-play days, spend days, and Robux spent. Same signals you already know, just measured across a much longer horizon. That new Day 8–28 bucket is the real story — it's where "do players still care about you three weeks later?" gets scored.
Marsh
Some of those terms lost me. "Qualified play sessions"? "Intentional co-play days"?
Mallow
A qualified play session is a session substantial enough to count — not a five-second bounce, an actual visit where the player engaged. Intentional co-play days are days when a player deliberately played with other people — friends, groups, social play. Both reward depth and stickiness: a real session beats a drive-by, and social play is one of the strongest retention drivers on the platform.
Marsh
You said there was a second change, about play-through rate?
Mallow
Right. Roblox is removing qualified play-through rate — QPTR — and adding plain play-through rate (PTR) plus first-play bounce rates. This zooms in on the moment right after a player taps your icon: did they actually load in and stay, or did they bounce out almost immediately? It's a sharper read on the quality of that very first impression.
Marsh
So a clickbait thumbnail that wins the tap but loses the player…
Mallow
…now works against you. First-play bounce rate measures exactly that failure. A gorgeous icon that oversells gets the click, but if players pile in and bail in the opening session, the algorithm reads it as a bad match and pulls back distribution. The winning move flips: an honest, high-converting first impression that delivers on what the icon promised beats a clickbait icon that doesn't.
Marsh
And the third change — transparency?
Mallow
This one's genuinely unusual. Roblox is publishing the relative importance of the different signals — a chart showing roughly how much each one contributes — so creators know where to focus. Recommendation algorithms are normally black boxes. Telling you which levers matter most means you can prioritize the high-weight ones instead of guessing.
Marsh
Where do we see our own numbers for all this?
Mallow
In Creator Analytics ⟩ Acquisition ⟩ Home Recommendations. And here's a crucial caveat: every one of these signals counts only players who joined organically from the RFY sort — not ads, not sponsored placements, not referral links. So you can't buy your way into a better RFY score. Paid traffic grows your DAU, but it doesn't move this particular needle.
Marsh
Why is Roblox doing this now?
Mallow
To stop rewarding flash-in-the-pan hits and clickbait, and start rewarding experiences that genuinely retain. Long-term retention means more total session time, more spend, and a healthier platform overall — so Roblox is aligning discovery with quality. It's the same maturation arc as the economy-transparency and analytics moves: push the ecosystem toward durable value, not short-term spikes.
Marsh
For a brand experience, what should actually change in our plan?
Mallow
Three shifts. (1) Stop optimizing only for the launch spike — build explicit reasons to return in the Day 8–28 window: content cadence, live events, progression, seasonal drops. (2) Make the first session deliver on the icon's promise, so first-play bounce stays low. (3) Lean into social — design for intentional co-play, because friends-playing-together is now an explicitly weighted retention signal. Then watch the Home Recommendations tab weekly to see which bucket is leaking.
Marsh
Anything time-sensitive we should flag?
Mallow
Roblox scheduled a webinar for June 17 at 11am PT (registration required) to walk through the implementation. If discovery is core to your strategy, send whoever owns analytics. Expect them to clarify exactly how the weightings are presented and how the new bounce metric is computed.
Marsh
Executive summary?
Mallow
(1) RFY now evaluates over 28 days in three buckets (Day 1 / 2–7 / 8–28), rewarding sustained retention over the launch spike. (2) QPTR is out; PTR + first-play bounce are in — the first impression and session quality now count directly. (3) Signal weightings are being published; your data lives in Creator Analytics ⟩ Acquisition ⟩ Home Recommendations, and it's organic-only. (4) For brands: invest in the four-week return loop, an honest first session, and intentional co-play. Webinar June 17, 11am PT.
Marsh
So the algorithm finally rewards "did they still love it a month later" instead of "did they click once." That's… exactly the kind of value a brand should be building anyway.
