One engine. Every decision you have to defend.
Fraud, credit, underwriting, disputes and AML all ask you to take a position, act on it consistently and defend it later. The data differs. The machinery does not.
These are the same decision, wearing different clothes.
Canonical entities and point-in-time-correct features, defined once and reused by every pack.
Features, model, policy, verdict. One synchronous call, whether the question is fraud, credit or eligibility.
Policies authored in plain language, simulated against real history, versioned and approved the same way.
An examiner reads a credit decision the way they read a fraud one: score, features, rule hits, model version.
This is why a second decision domain is a configuration exercise rather than a second procurement. See the architecture →
What each pack decides.
Every pack answers one question, on your data, in your deployment. The status says how much is already built, so you can tell what a first pilot would actually involve.
Fraud management
- Decides
- Approve, review or decline a transaction or account action, inside the authorization path.
- Run by
- Fraud strategy, fraud operations, financial crime.
Agentic Commerce
- Decides
- Can a customer's agent find and understand what you sell (ready to pilot), and when it comes back to transact, was it allowed to do that (in design)?
- Run by
- Digital commerce and risk teams at merchants, marketplaces and platforms.
Credit risk
- Decides
- Extend, limit, price or decline exposure, for a new applicant or an account you already carry.
- Signals in
- Bureau and alternative data, cash flow, repayment history and account behavior, on point-in-time-correct features.
- Decision out
- Decision, score, the policy that bound it, and reasons a customer can be given.
- Run by
- Credit policy, risk analytics, portfolio management.
Features are defined once and materialized for both training and serving, so a backtest and a live decision see the same thing. Policy is authored in plain language and run against your own history before it binds anything.
Underwriting
- Decides
- Accept, refer or decline an application against written policy, with the reasoning attached rather than reconstructed later.
- Signals in
- Application data, documents, verification checks and internal history.
- Decision out
- Outcome plus the policy clauses and evidence behind it, ready for review.
- Run by
- Underwriting, policy and quality assurance.
Most of underwriting's cost is reconstructing why a decision was made. Clauses are authored, versioned and approved in one surface, and the documents and checks that supported an outcome stay bound to it rather than being reassembled during a review.
Disputes and chargebacks
- Decides
- Accept, represent or write off a claim inside network deadlines, where assembling the evidence is most of the work.
- Run by
- Disputes operations, chargeback and back-office teams.
PRIMITIVES EXIST · PACK SCOPED WITH THE FIRST PARTNER
AML, KYC and transaction monitoring
- Decides
- Onboard, clear, escalate or file, with a trail an examiner can follow without asking anyone to explain it.
- Run by
- Compliance, financial crime, the MLRO's function.
PRIMITIVES EXIST · PACK SCOPED WITH THE FIRST PARTNER
Bring us the decision you actually need to make.
If a decision has data behind it, a policy around it and an auditor after it, it maps onto the same primitives.
Candidates, not shipped packs. We will tell you which parts exist, which we would build with you, and how long a first pilot would take.