AI live-ops and player retention platform for a game studio. D30 retention up 41%, ARPDAU up 28%

The studio operated two live free-to-play titles in adjacent genres. Both were generating strong launch numbers and both were losing players in the same shape. a sharp drop between day 7 and day 30, a long tail of players who installed and never converted, and a high-value cohort that was monetising hard but burning out fast because the offer cadence was wrong for them. The live-ops team was talented and engaged but the tooling was unfit for the scale. Player segmentation was manual and lagged by days. Targeted offers were technically possible but operationally impractical at the speed the game economies needed.

Toxicity and anti-cheat were reactive. The community team was learning about toxic behaviour from player complaints rather than from the in-game telemetry that had been showing the pattern. The economy team was tuning prices on a weekly cadence using week-old data, which meant the team was always answering last week's question instead of this week's.

We built a live-ops intelligence platform that fused player behaviour, monetisation, social signals, and in-game economy data into one continuously updated layer running on a real-time event pipeline. The player segmentation engine clusters players in real time across four dimensions. behaviour, spend, social engagement, and skill progression. and the cohorts update continuously as players move between segments. The personalised offer engine targets every offer to the right player with the right discount, at the right moment in their session, against an in-engine surface that respects the design team's constraints.

The churn prediction model scores every active player on a rolling basis and surfaces at-risk players inside the intervention window where action still works. The toxicity and anti-cheat layer detects abuse, harassment, and exploit patterns in near real time, with a tiered response policy from automated warnings up to ban referrals reviewed by the community team. The economy intelligence module gives the design team a daily view of price elasticity across every offer SKU, currency flow across the in-game economy, and the leading indicators of inflation or sink imbalance.

The studio's pre-platform offer system was the same offer for every player in a coarse cohort. That meant high-value players were being given discounts they did not need. straight margin erosion. and at-risk players were being shown offers calibrated for the median player, who was not the one churning. The personalised offer engine flipped that. High-value players see scarcity and exclusivity offers that lift average revenue per paying user. At-risk players see targeted retention offers calibrated to the specific reason they are at risk. Median players see the offers that the design team has shown best convert their cohort.

ARPDAU climbed by 28 percent across both titles inside the first 60 days. Win-back conversion on lapsed players tripled against the prior baseline. The design team retained full control of every offer in the catalogue. the engine targets within constraints the team owns, it does not invent offers the design team has not approved.

Day-30 retention climbed by 41 percent across both titles. ARPDAU grew by 28 percent. Lapsed-player win-back conversion tripled against the prior baseline. Live-ops manual operations time per week dropped by 62 percent, freeing the live-ops team to focus on creative interventions and event design instead of segmentation spreadsheets. Toxicity-related player reports dropped sharply once the detection layer caught the pattern at telemetry level, and the community team's average response time on confirmed abuse cases dropped from days to hours.

The platform has since been extended to a third title in the studio's pipeline and the foundation supports the studio's planned move into a publishing model where the same live-ops infrastructure can be offered to partner studios as a managed service.