Payments & Transaction Banking · Module 3 of 4
The Chargeback Line
The risk engine has sorted 921,913 card payments into twenty bands by how likely each is to be fraudulent, and every single band contains both real customers and fraud. There is no clean cut, which means the only question available is which error to make. Fraud operations wants the risky bands stopped; merchant operations points out that every decline is a customer who tried to buy something and was told no. They have argued for a month about where the line goes, and both of them are wrong in the same way: they have assumed there is one line. You set two gates, approve below the first, review between them, decline above the second, while four meters run live and nothing is flagged or colour-coded, because the two numbers this business actually reports both point the wrong way. The decision turns on a price nobody has written down: what a falsely declined customer really costs, the sale plus the business that never happens, most of it landing after the transaction is over, in a report nobody connects to the decline. Put that number in and every boundary becomes an exchange rate. Does the band everyone wants to block clear it? Declining everything feels responsible the whole way down; compare it with approving everything. What does a review queue add that a single line cannot? And the scheme's chargeback limit, the one externally-imposed number everyone manages to, has no term for a refused customer at all. Then the fraud ring learns where your screen is and moves: which of your gates should follow?