Risk had the story wrong. $5M didn’t have 72 hours to wait.
A payment processor moved to terminate a domestic account carrying $5M in annual revenue. Their risk team had misread how the client actually generated sales, and the fraud model did the rest. I got the client's leadership on a call with the processor's head of risk, built the record that showed what was actually true, and reversed the decision in 72 hours.
The situation
The notice came in the way these notices always come in: final, unemotional, and short on time. A payment processor had decided to terminate the account. $5M in annual revenue ran through that relationship, and the client had no real runway to argue the point through normal channels. Support tickets do not reverse a termination decision. Neither does a polite email to an account manager who does not own the call.
Losing that processor would not have been a paperwork problem. It would have meant an immediate stop on collecting payment, a scramble to onboard a new processor under a compressed timeline, and every new provider asking the same question that got this account flagged in the first place, only now with a termination on the record working against the answer. A second processor sees a termination from a prior provider and treats it as evidence, not as noise. This had to get fixed with the current processor, not routed around.
This was a domestic processor, not a correspondent bank three time zones away with a compliance queue measured in weeks. That mattered later. It meant the people who could actually fix this were reachable, on the same working day, in the same legal system. The problem was not distance. The problem was that risk had the wrong picture of the business, and nobody had corrected it yet.
How risk misread the model
Risk teams do not see a business. They see a pattern, matched against a list of patterns that have caused losses before. On paper, how this client generated sales looked close enough to that list to get flagged: the sequence between first contact and payment, the way a sale actually closed, the shape of it when you looked at it from a fraud model's distance instead of up close.
None of it was fraud. All of it looked, from that distance, like fraud sometimes does. That is the trap. A model built to catch bad actors cannot tell the difference between a business that operates unconventionally and a business that operates dishonestly, not without someone explaining the difference in terms the model's owners actually use. Nobody had done that yet. The account had been flagged, escalated, and scheduled for termination before anyone tried.
The room
I got the client's leadership on a call with the processor's head of risk. Not a junior account contact on either side. The people who actually understood the business, talking directly to the person who actually had the authority to undo the decision. That alone changes a conversation. A ticket gets read by whoever is queued to read it. A call with leadership on both sides gets someone's full attention.
Before that call happened, I worked with the client to pull together the documentation that actually mattered: transaction records, the real sales process laid out step by step, the paper trail connecting a customer's first contact to the charge that eventually showed up on their statement. Ample documentation, not a single reassuring summary. Risk teams do not get talked out of a decision. They get shown a record that leaves nothing for their model to still flag.
Then I wrote the piece that did the real work, a plain explanation of how the business actually operated, in the language a risk analyst uses to make decisions rather than the language a business uses to describe itself to customers. Not a defense. Not a pitch. A record, built to answer the exact question their model had flagged, so the person on the other end of the call did not have to take anyone's word for it.
Reversed in 72 hours
The call happened. The documentation did the rest. A head of risk does not need to be persuaded, they need a clean record that reconciles what the model flagged against what was actually happening. Once they had that record in front of them, in a format built for how they evaluate accounts, there was nothing left to argue about.
The termination was reversed in 72 hours. The $5M in annual revenue stayed on the account. The client did not lose a single day of processing, and the relationship with that processor is still intact.
What founders should watch
- Risk teams act on the file in front of them, not on the truth. If the file is thin or unclear, the model will draw the wrong conclusion every time, and it will not apologize for it.
- A termination notice comes with a clock attached. It will not wait for you to build the full picture after the fact. Have your documentation ready before you need it, not after.
- Get the right people on the same call. Support tickets and email threads do not reverse decisions like this. A direct conversation, backed by a record built in the reviewer's own language, does.
- Domestic accounts move faster. If you are working through correspondent banks or cross-border compliance layers, budget for that speed difference and start earlier than feels necessary.
- A termination from one processor follows you to the next one. Fix it with the processor you have. Switching to a new provider is rarely the faster path once a termination is already on the record.