Stop the fraud. Keep the customer.
Ward scores a payment in the authorization path, reviews an account's history when a transaction is disputed, and tells you which merchants to stay away from. It approves the genuine first purchase.
From the authorization to the dispute.
The same model follows a payment through its life, so the fraud team and the chargeback team read the same evidence.
Card payments
Device and IP fan-out, velocity, entry mode, 3-D Secure, CVV and AVS, read against the cardholder's own history and profile sent in the request.
CL_CARD_TESTING · CL_NOT_AUTHENTICATED · CL_CARD_BURST type: paymentAccount-to-account
New payees, scam-payment patterns and unusual amounts, with the account's recent movements as context.
review floors · declines never loosened type: atoAccount takeover review
Reads the logs before a disputed transaction: new device, SIM swap, contact change, payee added. Third-party fraud or first-party misuse.
likelihood · assessment cookbookChargeback triage
Upload a declines or disputes export. Ward reads the ISO 8583 data elements and sorts cases by what the evidence supports.
any layout · card network exports type: merchantMerchant risk and WMID
What a merchant's website sells against its declared MCC, and Ward Merchant Intelligence Data: merchants that collude to allow fraud.
MR_WEBSITE_MCC_MISMATCH · MR_YOUNG_DOMAIN agentsPayments started by AI agents
Ward works out whether a person, a merchant or an agent initiated the payment and routes agents through the agent engine.
initiator · fail-closed on conflictThe cost of a false decline is a customer.
New customers have no history, which is where false declines pile up. On our cold-start card benchmark Ward approved all but one genuine customer.
Cold-start card benchmark, October 2026: 229 genuine customers, 65 frauds. Full results →
Fits where your decision is made.
- In the authorization path. 40 ms p50 server time on card decisions. A component that cannot run returns a review, never a pass.
- Your thresholds. Set the step-up and decline cuts and your base rate per client. Each decision records what it ran with.
- Your outcomes. Send chargebacks and confirmed fraud back. A per-client model trains in shadow and decides only when you approve it.
- Explanations on demand.
GET /decide/{id}/explainshows which signals moved the answer and by how much.
See Ward on your own cases.
Bring a statement, a declines export or the case your team is stuck on.