Skip to main content

​
Churn Risk Prediction

Windback continuously scores every customer in your project to predict how likely they are to churn. Scores update daily so you can intervene before a cancellation happens.

​
How It Works

The scoring algorithm analyzes 13 negative signals and 5 positive signals drawn from billing data, product usage, and support activity. Each signal carries a configurable weight, and the final score is normalized to a 0-100 scale where higher means more risk.

​
Negative Signals

#SignalWhat It Measures
1No events in 14 daysCustomer has gone completely silent
2Declining usageWeek-over-week drop in event volume
3Support ticketsRising number of open or recent tickets
4Failed paymentsOne or more payment attempts have failed
5Short sessionsAverage session duration is falling
6Feature drop-offCustomer stopped using features they previously used
7Billing downgradesCustomer moved to a lower-tier plan
8Login gapsIncreasing time between logins
9API error spikesSudden increase in client-side API errors
10Competitor page visitsVisits to known competitor comparison pages
11Cancel page viewsCustomer viewed the cancellation page
12NPS detractorsNPS response of 0-6
13Payment method expiringCard on file expires within 30 days

​
Positive Signals

Positive signals reduce the overall risk score:
#SignalWhat It Measures
1Feature adoptionCustomer is actively adopting new features
2Billing upgradesCustomer upgraded to a higher-tier plan
3Team invitesCustomer invited new team members
4API usage growthSteady increase in API call volume
5High NPSNPS response of 9-10

​
Risk Levels

Scores map to four risk levels:
LevelScore RangeRecommended Action
Critical80 - 100Immediate personal outreach
High60 - 79Automated win-back email + internal alert
Medium40 - 59Monitor closely, consider a check-in
Low0 - 39No action needed
Focus your team’s energy on Critical and High customers. Medium-risk customers often recover on their own once a product issue is resolved.

​
Auto-Email Delivery Modes

When a customer crosses a risk threshold, Windback can notify them (or you) automatically. Choose one of two delivery modes.
Windback sends the email to the customer on your behalf using your configured sender domain. No code required.
  • Emails are generated by Windback AI and match the tone you select (friendly, professional, urgent).
  • You can preview and edit templates in the dashboard before they go out.
Direct mode requires a verified sender domain. Go to Settings > Email to configure one.

​
Configuration API

Manage churn-risk settings programmatically with the configuration endpoints.

​
Get Current Config

curl https://api.windbackai.com/api/v1/projects/:slug/churn-risk/config \
  -H "Authorization: Bearer <token>"
Response: 
{
  "enabled": true,
  "delivery_mode": "direct",
  "risk_threshold": 60,
  "webhook_url": null,
  "email_tone": "friendly",
  "check_interval_hours": 24
}

​
Update Config

curl -X PUT https://api.windbackai.com/api/v1/projects/:slug/churn-risk/config \
  -H "Authorization: Bearer <token>" \
  -H "Content-Type: application/json" \
  -d '{
    "enabled": true,
    "delivery_mode": "webhook",
    "risk_threshold": 70,
    "webhook_url": "https://example.com/hooks/churn",
    "email_tone": "professional",
    "check_interval_hours": 12
  }'
Changing delivery_mode takes effect immediately. Any in-flight emails for the previous mode will still be delivered.

​
Dashboard Features

The Churn Risk page in your project dashboard gives you a real-time view of your customer base.

Risk Distribution

A breakdown of how many customers fall into each risk level, updated daily.

Customer List

Filterable table of all customers with their current score, top signals, and last activity date.

Score History

A per-customer timeline showing how their risk score has changed over the past 90 days.

Alert Log

A record of every auto-email or webhook delivery, including delivery status and timestamps.
Churn risk scoring requires at least 7 days of event data before scores become meaningful. New projects will see scores populate after the initial data collection period.