Real-time gaming and fan intelligence
In a live game, the objective you give the model decides who stays. Tune matchmaking and live-ops for the whole skill range and retention follows; tune them for the most vocal players, or for spend with children in the audience, and you lose players or invite enforcement.
Part of AI for Telecom, Media, Entertainment & Gaming. 6 sources, checked .
- Americans who play video games every week (ESA survey, 2026)
- 212.3M
- Americans who play video games every week (ESA survey, 2026)
- Entertainment Software Association, 2026(opens in a new tab)
- Share of players whose 14-day return rate fell significantly when skill matching was loosened in Activision's 2024 Call of Duty test
- 90%
- Share of players whose 14-day return rate fell significantly when skill matching was loosened in Activision's 2024 Call of Duty test
- Activision Research, 2024(opens in a new tab)
- Epic Games' 2022 FTC settlement: a $275M COPPA penalty plus $245M in consumer refunds
- $520M
- Epic Games' 2022 FTC settlement: a $275M COPPA penalty plus $245M in consumer refunds
- Federal Trade Commission, 2022(opens in a new tab)
The problem
The audience is large and it includes children. The ESA's 2026 survey found 212.3 million Americans play video games every week, up 3% on 2025, and 67% of Americans aged 5 to 90 play at least an hour a week.1 That means player telemetry, chat and purchase flows in many games fall under children's privacy law, and the models built on them inherit those limits.
Matchmaking is the clearest case of a model objective shaping retention. Activision reports that in an early 2024 Call of Duty test that loosened skill matching for half of North American players, the 14-day return rate fell significantly for 90% of players, while only the top 10% came back more. Its broader finding is that when skill is used, 80 to 90% of players place better, stick with the game longer and quit matches less often.2 Connection quality remains the most heavily weighted factor.
Regulators look at the same systems from the safety side. In 2022 Epic agreed to a $275 million COPPA penalty and $245 million in refunds. The FTC alleged that on-by-default voice and text chat, and matching children and teens with strangers, harmed young players, and that design tricks led to unwanted purchases. The order requires voice and text communications to be off by default.3
The rules have tightened since. The amended COPPA Rule took effect on June 23, 2025, with most obligations due by April 22, 2026.4 It requires separate parental consent to disclose children's data to third parties and bars keeping it indefinitely.5 In the EU, the Commission's July 2025 guidelines for platforms accessible to minors ask for private-by-default accounts, explicit signals over behavioral ones, and protection from manipulative virtual currencies and loot boxes.6
What we build
- 01
Multi-objective matchmaking
Balances connection quality, wait time, skill and party makeup, with every rule change A/B tested on return rate and quit rate broken out by skill decile.
- 02
Churn prediction for live-ops
Scores lapse risk from session telemetry and tests event, reward and content schedules on retention, with age signals routing child accounts to limited, non-profiling treatment.
- 03
Real-time chat and voice safety
Classifies harassment, threats and grooming signals in text and voice within the session, queues cases for human moderators, and enforces default-off communication for younger players.
- 04
Cheat and payment fraud detection
Flags anomalous input patterns, impossible statistics and suspicious purchase or refund sequences from streaming telemetry, with appeal paths for flagged players.
- 05
Live fan and audience signals
Tracks concurrency, chat sentiment and engagement during live events and esports broadcasts in real time, so production and community teams can respond while the moment is still live.
Where it goes wrong
The failure modes we design against from the first week, and the control for each.
Tuning for the loudest cohort
In Activision's test, looser skill matching helped the top 10% and hurt return rates for everyone else.2 Report every experiment by skill decile, set guardrails for the bottom and middle of the distribution, and run tests long enough to see return behavior.
Profiling children under the internal operations exception
COPPA lets you use persistent identifiers without consent for internal operations such as personalizing content, but not to amass a profile on a specific child or for behavioral advertising.5 Keep child-account data in separate pipelines with purpose limits enforced in code.
Unsafe defaults in social features
The FTC alleged that on-by-default chat and matching children with strangers harmed young players.3 Make communication off by default for minors, and add a safety review to the release gate for any feature that connects players.
Models that optimize spend from minors
The FTC alleged dark patterns drove unwanted purchases, and EU guidance names manipulative virtual currencies and loot boxes as practices to keep from children.3,6 Exclude minor accounts from spend-optimizing models and review purchase flows like a regulator would.
Rules that apply
COPPA Rule, 16 CFR Part 312
Primary text on ecfr.gov(opens in a new tab)Operators need verifiable parental consent before collecting, using or disclosing children's personal information, separate consent for third-party disclosure, and a written retention policy; data may not be kept indefinitely.
COPPA Rule amendments, 2025
Primary text on federalregister.gov(opens in a new tab)The amended rule took effect June 23, 2025, and most covered operators had to comply by April 22, 2026.
EU Digital Services Act, Article 28 guidelines on minors
Primary text on digital-strategy.ec.europa.eu(opens in a new tab)Platforms accessible to minors, other than micro and small enterprises, should set minors' accounts to private by default, prioritize children's explicit signals in recommendations, and turn off features such as streaks and push notifications by default.
How to measure it
- 14-day return rate by skill decile against a control group
- Match quit rate and blowout rate
- Time to match and connection quality (delta ping)
- Harassment reports per 1,000 matches and time to moderator action
- Share of child accounts with verified parental consent and default-off communication
Questions buyers ask
Does skill-based matchmaking actually improve retention?
Activision's published tests say yes for most players. Loosening skill matching in a 2024 Call of Duty test lowered 14-day return rates for 90% of players, with only the top tenth gaining. Tightening it beyond the current setting hurt high-skill players instead. The answer for your game comes from your own controlled tests, read by skill decile rather than as one average.
Can we personalize a game for players under 13?
Within limits. COPPA allows persistent identifiers without parental consent for internal operations, which includes personalizing content, but that data cannot be used to build a profile of a specific child or for behavioral advertising. Sharing children's data with third parties now needs separate parental consent. Build age signals into the data pipeline so child accounts are handled differently from the start.
How real-time does player analytics need to be?
It depends on the decision. Matchmaking and chat safety act within a session, so they need results in seconds or less. Live-ops decisions such as event tuning or offer timing can run on hourly or daily data. Building everything for sub-second latency is expensive; match the pipeline to the decision, and keep a shared definition of each metric across both.
Sources
Every figure on this page comes from the sources below. We checked each link on .
- 1.Two-Thirds of Americans Play Video Games Every Week According to New Report from the Entertainment Software Association(opens in a new tab)Entertainment Software Association, June 3, 2026 · theesa.com
- 2.Matchmaking Series: The Role of Skill in Matchmaking(opens in a new tab)Activision Research, July 2024 · activision.com
- 3.Fortnite Video Game Maker Epic Games to Pay More Than Half a Billion Dollars over FTC Allegations of Privacy Violations and Unwanted Charges(opens in a new tab)Federal Trade Commission, December 19, 2022 · ftc.gov
- 4.Children's Online Privacy Protection Rule (final rule)(opens in a new tab)Federal Trade Commission, Federal Register, April 22, 2025 · federalregister.gov
- 5.16 CFR Part 312: Children's Online Privacy Protection Rule(opens in a new tab)Electronic Code of Federal Regulations, 2026 · ecfr.gov
- 6.Commission publishes guidelines on the protection of minors(opens in a new tab)European Commission, July 14, 2025 · digital-strategy.ec.europa.eu