Fraud decisioning for card and lending programs. The labeling layer, the decision logic, and the controls around them.
Detection systems that hold up against adversaries, scale, and scrutiny.
Fraud breaks the product playbook. Ship a change, hold back a control group, read the lift — that works when labels arrive quickly, populations are stable, and nobody is adapting to your test.
Fraud gives you none of that. So programs end up measuring the wrong thing slowly, then tightening by adding rules until few people understand what the system does or can defend why it does it.
WHO THIS IS FORYour loss rate moved and it is unclear whether it is credit or fraud, and if fraud, what kind.
You have addressed this before, it worked for a while, and the losses came back somewhere else.
You are launching a product, channel, or partner program and need a view of where it can be attacked.
You have several fraud and authentication vendors and no cohesive strategy layer connecting them.
You have a build-versus-buy decision being framed by the vendors themselves.
You need controls that hold up under second-line challenge and independent review.
This is strategy and analytics work. It is not day-to-day fraud operations and it is not interim staffing – if that is what you need, I will point you somewhere better.
THE MEASUREMENT PROBLEMWhy the usual playbook fails
Chargeback labels land months after the decision. Declines never produce outcomes, so you only learn from data your own rules selected. Rings hit many accounts at once and contaminate control groups. And the fraudster adapts to whatever you just deployed.
Each of those breaks a different assumption behind controlled testing, which is why adding more discipline to the testing does not fix it. The measurement approach has to change.
STARTING POINTFraud Discovery
A two-week diagnostic. It runs on conversations with your team plus whatever data you can put in front of me quickly – no integration, no long onboarding.
You get a memo covering:
The main loss vectors across your attack surface, sized to orders of magnitude
What your current controls and vendors actually cover
Where the primary gaps are
A prioritized sequence of what to do about them
The shape of the deliverable is subject to interview access and data availability, and we establish what is available before it starts.
$7,500
FIXED FEE · TWO WEEKSMy only revenue is client fees. If the answer is that you should buy something rather than build it, I have no stake in which one you buy.
Everything after the diagnostic is scoped against what it finds.
METHODHow the work goes
-
The points in the product where value can be extracted, and how each one is reachable. Most programs have never written this down, and the gaps are usually visible the moment it exists.
-
Chargebacks and valid claims become confirmation rather than the primary signal. A labeling layer that combines heuristics, anomaly detection, and structured adjudication of evidence the business already holds — customer service transcripts among it — produces usable signal within days.
-
The hardest fraud questions are the ones with no loss history yet — a new product, a new channel, a new partner. There is nothing to train on, but there are people who know how this gets attacked. An explicit causal model turns that knowledge into a defensible estimate now, and updates it as data arrives.
Where a number genuinely is not known, the methodology carries a range instead of a false decimal — so you find out which conclusions are robust to what you do not know, and which are not.
-
Shadow scoring tells you what a change would have done before it does it to your customers. I have built the engine that does this, not just recommended one — at a fintech card program, a simulation engine and ensemble scoring framework raised fraud detection by 30% while collapsing an overgrown rule set into a small core with targeted guardrails.
-
Instead of an accumulated rule stack that no one can reason about. Simpler systems detect more, not less — that has been a recurring result.
WHO I AMJose G. Cintron, CFA
Two decades in risk analytics – the last eight years in fraud, first building and deploying the models, then running strategy for card and lending portfolios. Synchrony, where I led a twelve-person acquisition fraud organization, and fraud strategy at fintech card and lending programs.
I came into fraud from quantitative finance and machine learning. I built stress-testing models and the pipelines behind them, then built and deployed real-time fraud models and defended their methodology to Model Validation, Legal, and Compliance. I have integrated, tuned, and lived with most of the vendors in the identity and fraud stack – which is why I can tell you when one is a fit and when it is not.
That is why the work looks the way it does: explicit models, stated uncertainty, and systems I have built, not just specified.
GETTING STARTEDIf you run a card or lending program and any of this sounds familiar, book a short call. I will mostly be asking questions. Within two business days, you will have a written note telling you whether Fraud Discovery is the right starting point — or whether you need something else, in which case I will say so.