AI Optimization Platform
Engine A · AI Advisor

Know exactly what to fix next — and why.

The Advisor turns your event stream into diagnoses you can defend: churn risk, revenue leakage, drop-offs and monetization gaps, each explained with evidence and a measurement plan. Nothing ships until a developer approves it.

What it analyzes

Every lever of engagement and revenue

Journeys & funnels

Ordered funnels from your events, step conversion and drop-off points.

Retention & churn

Day-N cohorts, 14-day churn and a churn model with reported holdout AUC.

Revenue

IAP, subscriptions and ads; ARPU, ARPPU, payer conversion, concentration risk.

Segmentation

Behavioural segments (recency, frequency, spend) with readable names.

Feature adoption

Which features are discovered, and by whom.

Game progression

Per-level completion and fail rates; difficulty spikes relative to neighbouring levels.

Anomalies

Robust detection of unusual revenue or activity drops.

A/B test suggestions

Each recommendation can carry a ready-to-approve experiment for Autopilot.
The recommendation contract

Every recommendation explains itself

We don’t surface a suggestion unless we can show why. Each one carries the same eight fields, so product, engineering and finance can evaluate it the same way.

  • Developer approval required before anything is implemented
  • Approve, reject or mark implemented — every decision is audit-logged
  • Approved recommendations can become Autopilot experiments in one click
Problem identified
What is wrong, in plain language.
Evidence
The metrics and values that triggered it.
Affected users
How many, and which segment.
Recommended action
A concrete next step.
Expected impact
A range, labelled as an estimate, with assumptions.
Confidence
Low / medium / high, plus a numeric score.
Implementation difficulty
So you can prioritise.
Measurement method
How to prove it worked — usually a controlled test.
How it thinks

Deterministic where it matters, AI where it helps

Deterministic metrics

Revenue, conversion, retention and statistics are computed with exact, reproducible code — never by a language model.

Classical ML

Churn prediction (logistic regression with holdout AUC), K-means segmentation and LTV estimates. If there isn’t enough data, it says so and uses a transparent heuristic.

Language models for narrative

LLMs only write explanations and reports, using aggregates — local models by default, cloud only if you opt in. If no model is available, a deterministic template is used and labelled.
Use cases

Where teams start

Mobile games

Find the level that ends careers, the offer nobody accepts and the players about to lapse.

Subscription apps

Quantify cancellations, trial-to-paid leakage and onboarding that never reaches value.

E-commerce

See which funnel step bleeds buyers and which segments respond to which offers.

SaaS

Feature adoption gaps, accounts at churn risk and expansion signals.

Integrations

Bring data the way you have it

  • Web, React Native and Unity SDKs
  • Server-to-server REST ingest with a secret key
  • Automated performance reports
  • Webhooks when new recommendations are created
Honest limits
  • • Models need history: churn modelling requires roughly 200+ users with two weeks of activity.
  • • Impact ranges are estimates from stated assumptions — only experiments prove impact.
  • • Thresholds that flag issues are internal heuristics, not industry benchmarks, and are configurable.
  • • Third-party analytics connectors (GA4, Mixpanel, Amplitude…) are on the roadmap and need your credentials.

See what the Advisor would tell you.

Run the free audit on an export you already have — it uses the same diagnostics engine.