Customer Intelligence
AI-powered payment failure predictions, churn risk, and health narratives
Customer Intelligence brings AI-powered predictions directly into RevKeen. Every customer is automatically analysed for payment failure risk, churn probability, and overall health -- with plain-language explanations and actionable recommendations.
What You Get
- AI health narratives -- A natural-language summary of each customer's payment behaviour, generated daily
- Payment failure predictions -- A probability score for each customer's next payment failing, based on a 20-feature analysis
- Churn predictions -- Monthly assessment of which customers are likely to leave, with contributing signals
- Unified Intelligence Panel -- All predictions consolidated on the customer detail page
- Dashboard summary card -- Aggregate risk metrics across your entire customer base
Setting Up
Customer Intelligence is a built-in RevKeen feature. No API keys, credentials, or external accounts are needed.
- Go to Apps > Customer Intelligence in your RevKeen dashboard
- Click Activate
- Predictions will begin on the next scheduled run (see schedule below)
Customer Intelligence requires at least one customer with invoice history. Predictions improve in accuracy as more payment data is collected.
How Predictions Work
Each prediction type follows the same pipeline:
- Feature extraction -- RevKeen collects relevant signals from the customer's payment history, subscription status, and health score
- AI analysis -- The feature vector is processed by a specialised AI model that returns a probability score, risk level, and explanatory factors
- Structured output -- Results are stored with the customer record and surfaced in the dashboard
Payment Failure Predictions
The payment failure model analyses 20 features per customer, including:
| Feature Category | Examples |
|---|---|
| Payment history | Recent failure rate, hard-decline count, average days to pay |
| Method health | Card expiration proximity, method age, number of methods on file |
| Billing patterns | Transaction amount volatility, frequency consistency, dunning depth |
| Account signals | Tenure, subscription tier, recent plan changes |
Each prediction produces:
- Failure probability (0--100%)
- Risk level (low, medium, or high)
- Top risk factors with relative importance
- Recommended action (e.g. "Update payment method", "Switch to manual invoicing")
Churn Predictions
The churn model runs monthly and evaluates:
| Signal Category | Examples |
|---|---|
| Health trend | Score velocity, direction (improving/stable/declining) |
| Subscription signals | Downgrades, cancellation requests, pause history |
| Payment behaviour | Failure frequency over 90 days, recovery rate |
| Engagement | Portal login frequency, support ticket volume |
Each prediction produces:
- Churn probability (0--100%)
- Risk level (low, medium, or high)
- Risk signals ordered by importance
- Recommended action (e.g. "Schedule retention call", "Offer discount on renewal")
AI Health Narratives
After health scores are recalculated, an AI model generates a plain-language narrative for each customer explaining:
- Current score and what it means
- Recent changes and their causes
- Specific risks or positive trends
- Suggested next steps
Narratives can be regenerated on demand from the customer detail page.
Prediction Schedule
| Prediction | Schedule | Scope |
|---|---|---|
| Health narratives | Daily at 03:30 UTC | All customers with health scores |
| Payment failure | Daily at 04:00 UTC | All customers with active payment methods |
| Churn | Monthly at 05:00 UTC (1st) | All customers with 30+ days of history |
You can also trigger any prediction on demand from the customer detail page.
Risk Levels
| Level | Probability Range | Meaning |
|---|---|---|
| Low | 0--30% | No action needed |
| Medium | 31--60% | Monitor and consider proactive outreach |
| High | 61--100% | Immediate attention recommended |
Outcome Tracking
RevKeen records actual outcomes (did the payment fail? did the customer churn?) alongside predictions. This data is used to continuously measure model accuracy and improve future predictions.
Where to View Predictions
| Location | What's Shown |
|---|---|
| Dashboard home | Intelligence Summary card with aggregate metrics |
| Customer list | Risk indicators alongside health scores |
| Customer detail | Full Intelligence Panel with narrative, payment risk, and churn risk |
| Apps > Customer Intelligence | Overall status, last run timestamps, and summary statistics |
Troubleshooting
| Issue | Solution |
|---|---|
| No predictions appearing | Predictions run on schedule -- check that at least one customer has invoice history |
| Narrative says "No data" | The customer needs at least one completed invoice for narrative generation |
| Risk level seems wrong | Predictions improve over time as more payment data accumulates |
| Predictions not updating | Check the prediction schedule -- churn runs monthly, payment failure and narratives run daily |
| Summary card shows zeros | New merchants will see data after the first scheduled prediction run |
Deactivating
To disable Customer Intelligence:
- Go to Apps > Customer Intelligence
- Click Deactivate
Existing prediction data will be retained but no new predictions will be generated.