Roblox
ML discovery engine plus ML economy (personalized pricing and storefronts)
Roblox runs an in-house discovery engine (sequence modeling, self-supervised) that concentrates most traffic on the home page and an Economy ML team that optimizes pricing and storefronts, with automated regional pricing launched in April 2025 and $923 million paid to creators in 2024.
Key points
- In-house discovery engine plus ML economy (smart pricing and personalized storefronts).
- Proprietary models (sequence modeling, self-supervised, counterfactual) and automated regional pricing launched in April 2025.
- $923M paid to creators in 2024 (+25%), with most traffic flowing through the ML-driven home page.
- Evidence level B, confirmed status.
Objective
Maximize the discovery of relevant content, hence engagement and Robux spend, and optimize the economy (price by market, personalized storefronts) to strengthen monetization over time.
The deployment
Roblox runs an in-house discovery engine where the home page concentrates most of the traffic. The teams train proprietary models rather than off-the-shelf LLMs, with sequence modeling on play history, self-supervised representation, and counterfactual evaluation, on infrastructure rebuilt to stay real time without blowing up the cost of service. The Recommended For You row relies on engagement signals such as the qualified play-through rate, 7-Day Playtime, and Robux Spend, with the addition of a social co-play dimension. An Economy ML team builds the machine learning layer for the marketplace, creator monetization, and payments, with smart pricing and personalized storefronts. In April 2025, Roblox launched regional pricing that automatically optimizes the price of each item to the user's local economy, with an item at 199 Robux in the United States shown at 139 Robux in Brazil, the stated goal being to capture 10 percent of global video game spending. Creators earned 923 million dollars in 2024, up 25 percent, on a daily active user base growing 21 percent.
Results Proof B
Two official Roblox releases document the ML discovery and economy systems, established press confirms the automated regional pricing, and the financial scale ($923 million to creators) is public. The isolated impact of AI is not a single financial line, hence B.
How it works
Documented architectureThe stack in detail
- outil Moteur de decouverte proprietaire Roblox In-house recommendation models: sequence modeling on play history, self-supervised representation, counterfactual evaluation; no off-the-shelf LLM.
- outil Economy ML (Roblox) Marketplace ML layer: smart pricing, personalized storefronts, creator monetization, and payments.
- outil Pricing regional automatise Automatic optimization of each item's price to the user's local economy (launched April 2025), with VPN detection on the anti-fraud side.
- infra Infrastructure de serving temps reel in-house Infrastructure rebuilt to serve recommendations in real time without blowing up the cost of service.
How it runs, concretely
For ops teams-
1Signal ingestion data / AI team
Play history, engagement, and Robux spend feed the proprietary models.
-
2Discovery recommendation AI (Search and Discovery)
Sequence modeling and self-supervised representation rank the experiences on the home page.
-
3Economy optimization AI (Economy ML)
Smart pricing, personalized storefronts, and regional pricing by local economy.
-
4Anti-fraud and control AI / trust team
VPN and location detection to prevent circumvention of regional pricing.
The engagement signals (qualified play-through rate, 7-Day Playtime, Robux Spend, co-play). If these signals degrade or are poorly attributed, recommendation relevance and spend optimization decline.
How your customers perceive this type of use
Sourced studiesLe paradoxe est documente des deux cotes : 71% des consommateurs attendent des interactions personnalisees et 76% sont frustres quand elles manquent (McKinsey, 2021), mais 75% declarent ne pas acheter aupres d'organisations auxquelles ils ne confient pas leurs donnees (Cisco, 2024). La « creepy line » est localisee : messages recus quelques secondes apres une recherche et suivi de localisation sont les pratiques qui mettent le plus mal a l'aise (Periscope by McKinsey, 2019).
Acceptance conditions
- La confiance dans le traitement des donnees precede l'achat : 75% ne achetent pas sans elle (Cisco 2024)
- Un cadre legal protecteur rassure : 59% des consommateurs disent que des lois fortes sur la vie privee les rendent plus a l'aise pour partager des informations dans des applications IA (Cisco 2024)
- La personnalisation elle-meme est attendue quand elle est consentie : environ la moitie des consommateurs (US 55%, UK 52%) disent s'inscrire souvent ou parfois a des services personnalises (Periscope by McKinsey 2019)
Red lines
- Le message declenche quelques secondes apres une recherche ou un achat : deuxieme ou troisieme cause de malaise selon les pays (Periscope by McKinsey 2019)
- Le suivi de localisation percu comme de la surveillance : 40% de malaise en Allemagne et au Royaume-Uni (Periscope by McKinsey 2019)
- Le mesusage des donnees personnelles par l'IA, devenu la premiere inquietude des consommateurs, a 53% et en hausse (Qualtrics 2025)
Sources: McKinsey & Company 2021 · Periscope by McKinsey 2019 · Cisco 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- per-user engagement and play history
- spend signals
- a catalog of content and items
- reliable location
Org prerequisites
- an ML recommendation team
- an economy or pricing team
- cost-controlled real-time infrastructure
- governance over minors and pricing
Possible stack
- custom recommendation models
- sequence modeling
- pricing engine
- real-time serving infrastructure
The plan, step by step
- Step 1Define the reference engagement signals (qualified play-through, playtime, spend) and instrument them.Deliverable: Signal taxonomy measured in production.
- Step 2Train a first recommendation model on the highest-traffic surface and A/B test it.Deliverable: v1 model with measured engagement uplift.
- Step 3Industrialize cost-controlled real-time serving (latency and cost per request tracked).Deliverable: Recommendation engine in production at scale.
- Step 4Launch the economy layer (pricing by market, personalized storefronts) on a pilot scope, with anti-fraud guardrails (location, VPN).Deliverable: Controlled pricing test with spend measurement.
- Step 5Generalize and set governance (profiling of minors, consumer law on pricing).Deliverable: Monitoring setup and operational compliance rules.
First step: Define the reference engagement signals (play-through, playtime, spend) and launch a first recommendation model on the highest-traffic surface.
The same pattern, proven in other industries
DraftKings
ML personalization and optimization of promotion and pricing across player cohorts
FanDuel
personalization of the betting offer driven by AI pricing, with a genAI discovery assistant
Sources
- S1 Inside the Tech - Solving for Personalization Primary archive pending
- S2 Unveiling the Future of Creation With Native 3D Generation, Collaborative Studio Tools, and Economy Expansion Primary archive pending
- S3 Roblox rolls out automated regional pricing tools for developers Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.