Skip to main content
Industries · Mobile Games

Autonomous AI discovery built for mobile games.

Continuous AI discovery across your install, session, monetization, and live-ops data. Every finding ships with the underlying queries, the cohort definition, and the cost of the run. The cohort-level questions that never make it past the weekly KPI deck, answered before your next live ops review.

Built for Heads of Live Ops, Monetization Leads, Heads of Analytics, and Senior Live Ops Producers at mobile and F2P studios.

CRITICALRetention cliff
D7 retention from one ad network collapsed to 6% over six weeks.
~$340k UA spend on the affected source over the…
From a real DecisionBox run
Why DecisionBox

The cohort-level questions that never make it past the weekly KPI deck, running overnight against your data.

Your team owns the event calendar, the offer design, the UA mix. The agent runs unattended against your install, session, spend, and content data — and ships findings into a queue your live ops team triages. Every claim carries the underlying queries and the data-validated number, not the MMP-reported or SDK-reported one.

See how the agent reads cohort retention and LTV
The cost of late signals

The KPI deck looks fine. The cohort underneath is where the game actually moved.

Three benchmarks every live ops and monetization leader recognizes. Each points to the same gap: your data already holds the signals — install-source LTV, whale behavior shifts, event payback math — but they only surface when somebody asks the right question. Most weeks, nobody does.

D30 retention
4%
Median Day-30 retention across mobile games. The aggregate hides which install sources, cohorts, and content paths return the players you actually keep.
GameAnalytics, Mobile Gaming Benchmarks
Revenue concentration
50%
Of F2P revenue typically comes from the top 0.15% of players. When the whale cohort drifts, the dashboard ARPU stays flat — and the data already knows.
Swrve, Mobile Monetization Report
Paying conversion
2%
Of installs convert to a first purchase. The narrow funnel makes cohort-level drift expensive — small shifts in conversion or first-purchase ARPU change the LTV math.
Liftoff, Mobile Heroes Report
From real runs

Three findings the agent surfaced, paired with the recommendation it shipped.

Three shapes the live ops team sees every quarter: a single-channel retention cliff the aggregate concealed, a cross-system whale drift the dashboard ARPU averaged away, and an event lift that hid an LTV mirage. Same auditable trail every time — severity, the underlying queries, the cohort definition, the cost of the run.

CRITICALRetention cliff
D7 retention from one ad network collapsed from 14% to 6% over six weeks.
period: 6-week rolling · affected source D7: 14% →…
Cross-checked against installs.by_source…
single-channel retention cliff
P1Recommendation
Pause source + reweight UA
Pause the affected source pending creative review;
Expected impact
target: stop paying for installs that don't retain…
Target: Pause source +…3 concrete actions
CRITICALWhale drift
Top 1% spender ARPDAU fell 22% over six weeks.
period: 6-week rolling · top-1% ARPDAU: $14.20 →…
Cross-checked across iap.transactions…
cross-system pattern
P1Recommendation
VIP outreach + offer-tier audit
Surface the affected whale cohort to the VIP team this week;
Expected impact
target: defend whale ARPDAU before churn lands · effort…
Target: VIP outreach +…2 concrete actions
HIGHEvent LTV mirage
Last event drove +18% revenue on launch day.
launch-day revenue lift: +18% · event-acquired…
Cross-checked across events.calendar…
launch-day lift hid the LTV
P1Recommendation
Switch event KPI + adjust next-event offer mix
Re-evaluate the event's payback against 14-day LTV, not launch-day lift;
Expected impact
target: stop borrowing spend from the next 14 days…
Target: Switch event KPI +…2 concrete actions

Point the agent at your live ops data for a week. Triage what it finds Monday morning.

Use cases

Five surfaces the team already owns, one continuous discovery.

Same engine, same data, five categories of work your team is already doing — or wants to do, if the queue ever clears. Pick the surface that matters this quarter.

The install cohort that suddenly stops returning, and which step they fell out at.

D1 / D7 / D30 retention drift by install source, country, device tier, and onboarding cohort. The agent reads installs, sessions, tutorial events, and content milestones together — and surfaces the cohorts whose retention shape changed before the weekly KPI deck shows it. The deep dive lives in how the agent decides what to investigate.

Sample analysis areas
  • D1 / D7 / D30 retention drift by install source, geo, and device tier
  • Onboarding-step drop-off and tutorial completion by cohort
  • Build-version retention impact (regression catches per release)
  • Content-progression milestones and where cohorts stall
  • Pre-churn behavior patterns and early-warning windows
Why DecisionBox

Built for studios who want a cohort analyst running overnight, not three dashboards to reconcile.

Three things every finding from DecisionBox carries. None of them changes how your team works in your MMP, your analytics SDK, your event tooling, or whatever data tools and BI you already standardized on.

Sees what nobody had time to query for

The agent picks which cohorts, install sources, spender tiers, and events to investigate from your install, session, monetization, and content data. Findings ship into a queue your live ops team triages in the morning. The cohort-level question stops sitting in the data waiting for the next QBR.

Every number you'd defend at the live ops review

The underlying queries the agent ran, the cohort definition, the data window, the cost of the run, the validation checks. Your team re-runs anything, extends anything, overrides anything. No black boxes, no MMP-reported or SDK-reported numbers the data can't validate, no "trust the AI."

Reads all your data, not just the MMP and the analytics SDK

Install attribution plus session, monetization, content progression, and live-event data — read together. The patterns Adjust, AppsFlyer, or Singular can't see (because they only see attribution) and the patterns Amplitude, GameAnalytics, or deltaDNA can't see (because they only see events) surface when your data is read end-to-end. Snowflake, BigQuery, Databricks, Redshift, PostgreSQL. Open source, AGPL v3. The agent meets you on the stack you already run.

The next live ops review, with the cohort shifts already mapped.