Introduction
Finic gives calibrated risk scores and answers for fraud and risk, calibrated to your own fraud rates. Two endpoints, score and ask.
Finic gives you calibrated scores and answers for fraud and risk: payments, sign-ups, logins, applications, alerts, disputes. Every answer is a probability you can act on: from a calibrated scoring model, a score of 250 means about a 25% chance the outcome is bad.
What makes Finic different is that its answers are calibrated to your company's actual fraud distribution, not a generic one. A scoring model is fitted to your own labeled records, and recalibrated as you send more. A question can carry your base rate, and its answer's probabilities shift to match it.
Finic runs on two endpoints:
| Endpoint | Use it to | Price |
|---|---|---|
POST /v1/score | Rate the risk of a record, from 0 to 1000, with a scoring model calibrated to your own past outcomes | Usually under $0.0001 |
POST /v1/ask | Ask your own questions about a case and get a calibrated probability for each answer. Works out of the box | About $0.0002 for a 5,000-token case |
Score a payment
Create a scoring model once, from a description of the decision, and calibrate it with 100+ past payments labeled is_bad (or create it from them in one step). Then score each new one:
{
"model_id": "scm_01M4XRC000RT2GZSKYPD8J86G9",
"score_id": "pay_88213",
"data": { "amount_usd": 1850.0, "merchant_category": "electronics", "account_age_days": 6, "new_device": true, "billing_country": "US", "ip_country": "RO" }
}{ "model_id": "scm_01M4XRC000RT2GZSKYPD8J86G9", "score_id": "pay_88213", "score": 874, "request_id": "req_01M4YF1068X8B4C458ENPD4JHT" }About an 87% chance of a chargeback: decline it. As you learn how more payments turned out, calibrate the scoring model again with them.
Ask about an alert
Send the case and your questions, with how often the answer is yes in your cases:
{
"model": "grayson-1",
"data": {
"events": [
{ "ts": "2026-09-20T02:14:09Z", "type": "login", "new_device": true },
{ "ts": "2026-09-20T02:16:40Z", "type": "email_change" },
{ "ts": "2026-09-20T02:21:03Z", "type": "ach_out", "amount_usd": 4900.0 }
]
},
"context": "Customer called 9/21: never changed their email, doesn't recognize the transfer.",
"questions": {
"ato": { "type": "noul", "instructions": "Is this account takeover?", "base_rate": 0.02 }
}
}"answers": { "ato": { "type": "noul", "noul": 0.91, "calibrated": true } }A 91% chance of account takeover, on your 2% base rate: hold the transfer.
Answers come from Grayson, Finic's decision model. Base URL: https://api.finic.ai.
Start here
Developer quickstart
Ask a question, then create a scoring model and use it, with curl or Python.
Agent quickstart
Add Finic to Claude, ChatGPT, Cursor or any MCP client, and let your agent set it up.
Pricing
$0.035 per million input tokens, for scores and answers. Creating and calibrating scoring models is free.
Calibration
How scores and answers match your own fraud rates.