ACQUIRING RISK INTELLIGENCE

THE LOSS ARRIVES LONG AFTER THE DECISION.

It starts here. A card is presented, an authorisation is taken, a merchant is paid — fifty-four thousand times across this book. An acquirer does not lose money the day a merchant commits fraud. It loses money weeks later, when the chargebacks land and the merchant is gone. Everything that mattered was knowable at onboarding.

SCROLL ↓  ·  or press PLAY to watch it run

MA HPS · POWERCARD 1 240.00 MAD CONTACT · MCC 5732 APPROVED 00
01 — THE CORRIDOR

TWO COUNTRIES, ONE BOOK.

Morocco and France are not two markets side by side — they are one payment corridor. Cards issued on one side are presented on the other every day, under two regulators, two currencies and one acquiring licence.

02 — THE BOOK

THE ACQUIRER CARRIES THE RISK.

Not the merchant. If they take payments and disappear, the chargebacks sit with the acquirer. The exposure has to be visible before the losses arrive — across the whole book, every day.

03 — THE APPLICATION

EVERY FIELD CHECKS OUT.

A name. An ICE or a SIRET. A bank account, a director, a terminal, an address. Nothing in the file is false, and the application is approved in minutes — as it should be. Read on its own, it is clean.

04 — THE NETWORK

THE FRAUD IS IN THE EDGES.

28 July. Atlas Électroménager is thrown off the book — its chargebacks crossed the line.

20 August. Sahara Distribution is onboarded under a new name and a new legal id. The file is clean.

It settles to the same account, with the same director. Neither file says so. The edge between them does.

05 — THE PATTERN

SOME LEAVE NO EDGE.

An operator careful enough to use a new account, a new director, a new terminal and a new address shares nothing at all. No detector fires, and the graph is not wrong — on the facts it holds, this merchant is a stranger.

What it could not change is how it trades. Same hours, same ticket sizes, same decline curve. That is a vector, not a relationship.

06 — THE GAP

0.90 AGAINST 0.34.

Comptoir Atlantique. New account, new director, new bank, new address. It shares no edge with anyone — and no detector fires.

On conduct alone it matches two merchants — and one of them was thrown off the book in July.

The next merchant down is nowhere near. That distance is the finding; a table cannot show it.

07 — THE CATALOGUE

SIX WAYS A BOOK LOSES MONEY.

Five are joined by something they share; the sixth by how it trades. One graph, one set of vectors, six short queries — and every one of them found in this book.

08 — THE RULES

FRANCE AND MOROCCO.

Every finding has to survive a regulator. The console checks the book against named instruments — not a generic compliance score — and says plainly where the evidence is missing rather than scoring it green.

09 — THE DECISION

BEFORE THE LOSS ARRIVES.

The graph finds the ones that reuse something. The vectors find the ones that are careful. A small language model running on this machine reads both and writes the case — no external API, no merchant data leaving the building.

OPEN THE LIVE CONSOLE →
Demonstration on synthetic data. The banks and their BICs, the cities, MCC codes, ICE and SIRET formats, IBAN formats, chargeback reason codes and the regulatory instruments are real. The merchants and the people behind them are invented — nothing here describes a real business.