Credit scores for people banks can't see yet.
Most adults in Pakistan don't have a bank account, let alone a credit file — but they pay electricity bills, top up phones and move money through wallets every month. CredAI turns that history into an explainable score and a lending decision a human can stand behind.
- Held-out AUC
- 0.78
- Default rate cut
- 25% → 7.3%
- To plug in real data
- 1 env var
- Paid 10 of 12 utility bills on or before the due date.+50
- Same mobile number for 8 years 7 months.+18
- Average wallet balance of PKR 98,163 against PKR 44,258 monthly wallet spend.+17
- Meets every policy check. Eligible for instant approval.
From raw payments to a decision in four steps
No black box. Every step is code you can read, and every score comes with the reasons behind it.
- 01Connect
The customer consents to share utility bills, mobile top-ups, wallet, bank and repayment history — whatever they have. Any source can be switched off later.
- 02Engineer
17 behavioural features computed from raw records: on-time ratios, income regularity, overdrawn days, savings buffer. One feature engine for training and production.
- 03Score
A logistic scorecard on the familiar 300–850 scale. Each feature contributes points, and the points add up exactly — so every score explains itself.
- 04Decide
Affordability, coverage and score checks produce a recommendation. Small loans can clear instantly; larger ones go to an analyst. Everything is audited.
Try the model
Shape a customer's behaviour and choose what they share. The score, the reasons and the lending recommendation update instantly — the same code the lender console runs.
Each change simulates 18 months of raw activity — 338 records right now — then runs the production feature engine and model in your browser.
- +50Paid 20 of 24 utility bills on or before the due date.
- −13Mobile account was topped up in 9 of the last 12 months.
- −113 of 5 data sources connected.
- +8Average wallet balance of PKR 72,708 against PKR 46,829 monthly wallet spend.
One engine, three workspaces
See your score, what moves it, and the exact data used. Switch sources off and watch it recalculate. Apply with a live affordability check.
A review queue with scores, reason codes, cash-flow charts and policy checks. Approve, refer or decline with a recorded reason. Tune the policy live.
Model card with ROC, calibration and a fairness audit; a data-source health check; and a complete audit trail of every decision.
Bring your own data
The demo runs on a simulated population of customers with realistic transaction histories. Swapping in a real source means implementing one JSON contract — then setting an environment variable.
- — Stateless
POST /api/v1/scorefor any raw record - — Schema-validated: bad records are rejected with a path to the problem
- — OpenAPI 3.1 spec, health checks and a mock partner API included
curl -X POST https://<your-deployment>/api/v1/score \
-H 'content-type: application/json' \
-d @examples/customer-record.json
{
"report": {
"score": 742, "grade": "A", "gradeLabel": "Excellent",
"repaymentProbability": 0.91,
"helping": [{ "label": "Utility bills paid on time",
"points": 49.7,
"reason": "Paid 10 of 12 utility bills on or before the due date." }, …],
"missingSources": ["bank","credit"]
},
"decision": { "recommendation": "approve", "tier": "instant", … }
}Apply as a customer, decide as a lender, audit as the platform team.