Approve more.
Lose less to fraud.
Verdikt is the AI decisioning layer for e-commerce. We score every transaction in real time, stop fraud before it ships, and take the chargeback liability off your books. One platform for retail, marketplaces, travel, subscriptions and beyond.
An AI engine, not a rules engine.
Legacy fraud tools bolt machine learning onto decades-old rule engines. Verdikt is AI-native from the ground up, with five models working in concert on every transaction, whatever you sell and wherever you sell it.
A transformer trained on transactions
Most engines bolt rules onto gradient-boosted trees. Verdikt runs a transaction foundation model pretrained on billions of payment sequences, so a fraud pattern seen on one merchant is caught on yours at first sight.
See the whole fraud ring
Coordinated fraud hides across linked devices, recycled emails and mule accounts. Our graph neural network scores the network behind a transaction, not just the transaction, to expose rings before they scale.
Every verdict, in plain language
A reasoning layer turns hundreds of signals into a human-readable rationale for each decision. No black-box declines. Your risk team sees exactly why an order was approved, held or blocked.
Built for delayed labels
Chargebacks mature in 30-90 days while fraud shifts weekly. Verdikt fuses proxy signals and human-in-the-loop feedback to adapt in hours, not quarters: the hardest problem in fraud, solved by design.
Catches AI-powered fraud
Fraudsters now weaponise deepfake KYC, synthetic identities and bot farms. Verdikt is built for the era of AI attackers, modelling behavioural biometrics and device intelligence legacy tools never see.
Real-time at checkout scale
A single decisioning call returns approve, review or decline in under 50 milliseconds at p95, fast enough to sit inline with checkout without ever adding friction for good customers.
One call. Four models. A verdict in milliseconds.
Ingest
One API call streams the order, device, identity and behavioural signals into Verdikt.
Reason
The foundation model, graph engine and risk models score the transaction and its network in parallel.
Decide
A verdict (approve, review or decline) returns in <50ms with a confidence score and full rationale.
Learn
Outcomes and analyst feedback flow back continuously, sharpening the model against emerging fraud.
Live in an afternoon, not a quarter.
A single REST endpoint returns a complete risk decision. No model training, no data science team, no months-long integration.
- One decisioning endpointSend an order, get back a verdict, score and rationale.
- SDKs for every stackNode, Python, Ruby, Go, PHP and a typed REST API.
- Signed webhooksGet notified the moment a held order resolves or a chargeback arrives.
- Sandbox & replayTest against synthetic fraud and replay historical traffic before you go live.
import { Verdikt } from "@verdikt/node";
const verdikt = new Verdikt(process.env.VERDIKT_KEY);
// Score an order inline at checkout
const { verdict, score, reasons } =
await verdikt.decisions.create({
order_id: "ord_8842",
amount: 249.00,
currency: "USD",
product_type: "physical",
customer: { email, ip, device_id },
});
if (verdict === "approve") fulfill(order);
// → verdict: "approve" score: 0.04
// → reasons: ["established_device", ...]We don't just score the risk. We take it on.
On guaranteed plans, every order Verdikt approves is backed by a financial guarantee. If an approved order turns into fraud, we reimburse the chargeback, moving liability off your balance sheet and turning fraud from a variable cost into a fixed, predictable line item.
Ship fewer chargebacks by next quarter.
Book a 30-minute demo. We'll replay your historical traffic through Verdikt and show you, in dollars, exactly what it would have caught.