Risk, Fraud & Decision Systems
Production decision systems with model-validation and fair-lending discipline: a $150M risk portfolio and $1B annual transaction policy at Drip Capital; 16 million fraud decisions a month at Citibank.
- PRACTICE NO.
- 06
- PRINCIPALS
- SV · NB
- CASES ON FILE
- 02
- FIELD
- BANKING & RISK · FINANCIAL SERVICES · TRADE FINANCE & LENDING · CRYPTO & BLOCKCHAIN
PROBLEM SPACE
Every transaction is a decision, and most of them cannot wait for a human. The hard part is not the model. It is everything a regulator, an auditor, or a fraud ring will eventually test: stability under drift, fairness under scrutiny, a policy that holds at scale, and a paper trail for every call the system made.
We build decisioning the way banks have to run it: scored in production, validated on a schedule, and designed so the consequential edge cases route to a person while the routine majority clears in milliseconds.
WHAT WE DELIVER
- 2.1fraud & risk modelsreal-time and batch scoring
- 2.2underwriting / KYC / AML flowsonboarding through ongoing monitoring
- 2.3model validation (KS, PSI)stability and drift, on a schedule
- 2.4policy designthresholds, limits, exception paths
- 2.5decision audit trailsevery call reconstructable
HOW IT SHIPS
Systems in this practice run as a loop, not a launch. Inputs are scored, routine outcomes execute automatically, and anything consequential holds at a review gate where a named person clears it with context attached. Every decision (human or automatic) lands in the audit trail.
PROOF
SELECTED PROJECTS · CLIENTS ANONYMIZED
WHO BUILDS IT
ADJACENT
Asked about this practice
Questions we get.
The same answers we give on a first call about this practice.
FAQ-01What scale of decisioning have you actually run?
One of our principals ran fraud decisioning at Citibank making 16 million calls a month for five-plus years, and designed the risk policy governing $1B in annual transactions, alongside a $150M portfolio, at Drip Capital. This is production decisioning, not a scorecard demo.
FAQ-02How do you keep a model defensible to regulators?
With discipline that is mostly procedural: model validation using KS and PSI, fair-lending review built in from the start, a written policy layer, and decision audit trails. A model that silently drifts is a liability with a dashboard, so we monitor for it on a schedule.
FAQ-03Does the system decline customers on its own?
The model produces a score; a documented policy decides what happens at that score, and consequential calls route to a named human with context. We will not let AI make an irreversible decision about someone's money or eligibility alone.
Tell us the use case. One call is enough to scope whether there is a fit, and what it takes to ship.
Start a brief