Predict churn
before it happens.
Four-dimension customer health scoring, explainable churn risk, and playbooks that actually run themselves. Not a dashboard that tells you what you already know — a retention layer that does the next thing.
By the time they tell you, it's already over.
The customer stops logging in. Their champion leaves. A feature they relied on breaks. Three months later, a quiet cancel email lands in your inbox and everyone asks how this could have happened. RetainIQ is the layer that sees it in week one.
Four dimensions. One health score.
Each dimension is independently tracked, independently trended, and independently explainable. The weighted aggregation lives in Methodology #215 so you can see exactly how the number is built.
Engagement
How often and how deeply your customer is actually interacting. Logins, sessions, features touched, time between visits.
Usage
Breadth and depth of product adoption. Which features they have reached, which ones they have not, and how that trends week over week.
Support
Ticket volume, severity, time-to-resolution, satisfaction signal. A quiet customer is not always a happy one.
Financial
Payment health, plan changes, contract proximity, expansion signal. Failed payments are the highest-signal predictor we have.
WHAT "AUTOMATED" ACTUALLY MEANS
Real playbooks, built ground-up.
We did not bolt on Dittofeed or route your customer data through a third-party engagement SaaS. We built the executor in-house at src/empires/retainiq/executor/: an action registry pattern with typed handlers, a day-N scheduled action queue, and idempotency baked into every step.
A 30-day re-engagement playbook with actions on day 1, 3, 7, 14, 21 queues five scheduled actions the moment you launch it. The scheduler wakes each one up on its own timer. If a customer responds on day 4, Self-Refine re-plans days 7 through 21 to reflect what they actually wanted.
What's inside
Six capabilities that separate a retention platform from a health-score dashboard. Each one is production code, not a roadmap promise.
Ground-up Playbook Executor
A custom action registry built in-house. Not Dittofeed, not an external customer-engagement SaaS. Your playbooks run where your data lives.
Day-N Scheduled Actions
A 30-day re-engagement playbook queues actions on day 1, 3, 7, 14, 21. The scheduler wakes them up when it is time, not when a human remembers.
Action Registry Pattern
log_action, create_task, send_email to start — each action is a typed handler with idempotency built in. Re-running a step is a no-op, not a second email.
Constitutional Guardrails
Rate-limits interventions to at most 1 per customer per week. Refuses to upsell a customer whose health score is dropping. Written as rules, not as warnings.
Explainable Risk Scores
Every churn prediction comes with a factor breakdown. You see which dimension moved and why, not a black-box probability and a shrug.
Self-Refine Feedback Loops
When a customer opens an email, clicks, or declines, the playbook re-plans. The next action reflects the latest signal, not a week-old snapshot.
RetainIQ Pricing
Target launch pricing. Billed separately from your AI Employee subscription. Locked for the first year of any waitlist signup.
For a single CS team tracking one product.
- ✓Up to 1,000 tracked customers
- ✓4-dimension health scoring
- ✓Heuristic churn risk model
- ✓Pre-built playbook templates
- ✓Manual playbook execution
- ✓Email support
For retention leaders who want playbooks that run themselves.
- ✓Up to 10,000 tracked customers
- ✓Automated playbook executor
- ✓Day-N scheduled actions
- ✓Juno + Finn + Maestro squad
- ✓Methodology #209 + #215 wiring
- ✓Priority support + Slack channel
For multi-product CS orgs with custom retention logic.
- ✓Unlimited customers
- ✓Cox survival analysis (opt-in)
- ✓Custom action handlers
- ✓Constitutional guardrail authoring
- ✓SSO + audit log export
- ✓Named CSM
POWERED BY THE COGNITIVE MESH
Methodology #209 + #215, with 13 reasoning layers on top.
Churn prediction runs through Methodology #209 (Customer Churn Prediction). Health scoring runs through Methodology #215 (Health Score Aggregation). Both are wrapped by the same 13 cross-cutting reasoning methodologies that power the rest of LiquidCortex — Tree-of-Thought for playbook selection, Self-Refine for feedback loops, Episodic Memory for "this customer looks like a past save," Constitutional AI for intervention rate limits, Theory of Mind for modelling why they are actually disengaging.
WHAT'S REAL, WHAT'S COMING
Heuristic scoring ships now. Cox survival analysis is an optional path behind a feature flag.
REAL TODAY
- — 4-dimension heuristic health scoring (Methodology #215)
- — Explainable churn risk (Methodology #209)
- — Ground-up playbook executor with action registry
- — Day-N scheduled action queue
- — Constitutional guardrails (rate limits, upsell blocks)
- — HubSpot integration at the client layer
OPT-IN + ROADMAP
- — Cox Proportional Hazards survival model (optional, gated by
RETAINIQ_SURVIVAL_MODEL_ENABLED) - — Stripe webhook ingestion (Phase 5)
- — PostHog usage-signal ingestion (Phase 5)
- — SOC 2 Type II: not yet audited
- — Intercom, Mixpanel, Amplitude: not started
- — No model training on your customer data, ever
Questions a skeptical head of CS asks
The executor is real. It lives at src/empires/retainiq/executor/ and runs the action registry pattern end to end — executor loop, action handlers, day-N scheduler. The first action handlers (log, create_task, send_email) are wired. We built it ground-up instead of bolting on Dittofeed because your customer data should not leave your workspace to fire a re-engagement email.
Both, depending on your setup. The default path is a heuristic model wrapping Methodology #209 — transparent, fast, no training data required. Under the RETAINIQ_SURVIVAL_MODEL_ENABLED feature flag, a Cox Proportional Hazards survival model (Methodology #217) takes over and returns time-to-churn in days plus the top factors per customer. The ML path is optional because honest heuristics beat black-box ML when you have no baseline to compare against.
Four dimensions, weighted: engagement (35%), usage (30%), support (20%), financial (15%). The weights live in Methodology #215 (Customer Health Score Aggregation) so they are versioned, testable, and changeable per org once you have enough data to recommend a re-weighting.
HubSpot is wired at the integrations layer. PostHog for product usage events and Stripe for billing signals are Phase 5 work — they are scheduled but not yet imported into the RetainIQ routes. Intercom, Mixpanel, Amplitude, and Segment are not started. We say this up front because a "retention platform" with no billing integration is not a retention platform.
Most CS tools ship a single health score number with no explanation, no per-dimension breakdown, no re-planning when the situation changes, and no automated intervention. RetainIQ gives you the score, the factor breakdown, the recommended playbook, and the executor that actually runs the playbook — with guardrails that refuse to spam a customer who just got one intervention last week.
Keep the customers you already earned.
Join the RetainIQ waitlist. Launch pricing locked for the first year.