Pricing and token counting
A Ward decision costs its story tokens plus the checks that ran, at $0.04 per 1M tokens. Explanations (Ward Ask and Ward Chat) cost $0.40 per 1M tokens, input and output. Checks that pay an outside provider are billed at the provider's cost × 1.5. Sandbox keys (ward_test_…) are never billed.
You pay for the checks that ran on your call, never for one that did not apply. The counting rules are fixed and public, so you can work out the most a call can cost before you make it.
Story tokens
The story is what Ward reads for the decision: your input, plus the context it reads around it (the customer's profile from earlier decisions, your policies, coverage notes). 1 token = 4 bytes of UTF-8 text, rounded up.
| Type | What counts as input |
|---|---|
payment, intent, agent, screen, ato, merchant |
The request's input object as compact JSON (sorted keys, no spaces) |
statement |
The transactions read from the files: date, description, amount, balance and counterparty per row (about 17 tokens a row), plus the other fields. The raw file bytes do not count, so a PDF and a CSV of the same statement cost the same |
statement alert sheets, credit portfolio reports |
The cells read from the file, one row at a time |
document |
1,500 tokens for the attached image or PDF, plus the other fields as JSON |
kyc |
The applicant as JSON; an attached ID document adds 1,500 tokens, as in document |
Checks
Each check that runs adds a fixed number of tokens.
| Area | Check | Tokens |
|---|---|---|
| Transactions | Fraud | 6,000 |
| AML | 6,000 | |
| First-party misuse | 3,000 | |
| Merchant lists (WMID) | 2,000 | |
| Account takeover | 6,000 | |
| Typology | 6,000 | |
| Agents | Mandate gates | 6,000 |
| Behavioral overlay | 3,000 | |
| KYA Score | 4,000 | |
| KYA certificate | 1,000 | |
| Screening (per party) | Sanctions | 2,000 |
| PEP | 2,000 | |
| Enforcement | 1,000 | |
| Adverse media | 8,000 | |
| Documents and identity | MRZ | 2,000 |
| MRZ read from a photo (OCR) | 8,000 | |
| Alteration | 6,000 | |
| Data match | 1,000 | |
| Driver's licence barcode | 3,000 | |
| ePassport chip | 4,000 | |
| Proof of address | 6,000 | |
| Liveness | 15,000 | |
| Statements and credit | Each row scored | 200 |
| Each counterparty screened (up to 200 per statement) | 4,000 | |
| Affordability | 6,000 | |
| Merchants and businesses | Merchant risk | 6,000 |
| Website and domain | 1,000 | |
| Business registries | 3,000 | |
| Each owner screened (sanctions and PEP) | 4,000 | |
| KYC components | 2,000 | |
| Phone | 1,000 | |
| Contact history | 2,000 | |
| IP intelligence | 1,000 | |
| Wallet | 1,000 | |
| Address | 1,000 |
Bundles. Some sets of checks cost less together. A bundle applies when every check in it ran or did not apply to the call, and at least two ran. It never costs more than the checks that ran.
| Bundle | Checks | Tokens |
|---|---|---|
| Payment complete | Fraud, AML, WMID, first-party misuse | 14,000 (17,000 separately) |
| Screen complete | Sanctions, PEP, enforcement, adverse media, WMID | 12,500 |
| ID document complete | MRZ (OCR), alteration, data match, sanctions, PEP, enforcement, adverse media | 23,000 |
| KYC standard | Email, phone, contact history, IP intelligence, and the ID document checks | 28,000 |
| Statement AML and fraud | AML, fraud | 10,000 |
| Merchant onboarding | Merchant risk, website and domain, sanctions, enforcement, adverse media, WMID | 16,500 |
Examples
| Call | Tokens | Cost |
|---|---|---|
| A card payment: fraud and AML (input of 93 tokens) | 12,093 | $0.00048 |
| A card payment with first-party misuse and WMID (Payment complete) | 14,093 | $0.00056 |
| Screen a person: sanctions, PEP, enforcement, adverse media (Screen complete) | 12,525 | $0.00050 |
| Screen a name against sanctions and PEP only | 4,025 | $0.00016 |
| Account takeover check | 6,117 | $0.00024 |
| A passport photo with holder screening (ID document complete) | 24,540 | $0.00098 |
| A bank statement of 300 transactions, AML and fraud, counterparties not screened | 80,200 | $0.0032 |
| The same statement with 40 counterparties screened | 240,200 | $0.0096 |
| A Ward Ask question on a result (2,000 tokens) | 2,000 | $0.0008 |
Every decision carries its typologies, signals, scores and the rest of the answer at no extra cost.
Before and after a call
- Before: Ward holds the most the call can cost: its input, an allowance of 2,000 tokens for the context it reads, and every check the type can run at full price. If your prepaid balance does not cover the hold, nothing runs and you get
402 insufficient_balancewith the amount and your balance. - After: the call is priced on the checks that ran, and the rest of the hold comes back to your balance at once. If the decision fails, the whole hold comes back.
- On the bill: each call's tokens and cost are in usage and the ledger and on the console's Usage page.
Per-call limits
Real-time types (payment, intent, agent, screen) take up to 32,000 input tokens per call. File types take up to 1,000,000. See Sessions and billing for sessions, the context window and the other limits.