The chargeback you wrote off was winnable

Twenty minutes into a consultation call about fraud-model thresholds, the CFO asked the small question with the large invoice attached. "While we have you. Chargebacks. We mostly just... eat those, right? Everyone does?"

The numbers behind the question: disputes were running at six tenths of a percent of revenue. Each one also carried a network fee whether they fought it or not. Their win rate on the ones they did contest was 12 percent, achieved with a template letter that an ops person filled in when there was time, which there rarely was, against a deadline nobody tracked. The template had been losing politely for three years. Writing it all off felt like discipline. It was actually a standing donation.

A lost dispute is a refund issued to someone who kept the product.

Card networks do not adjudicate chargebacks on vibes. Every dispute arrives with a reason code, and every reason code has a published list of evidence that rebuts it. Fraud codes are rebutted with proof the cardholder was present: AVS and CVV match results, device fingerprint seen on prior orders, login from a known IP, a history of undisputed purchases on the same card. Product-not-received codes are rebutted with carrier confirmation and signature. For digital products, usage logs after the purchase date are close to a confession: the account that "never received access" streamed eleven hours of it on Tuesday.

Here is the part that makes this a data problem rather than a paperwork problem: every item on those lists already exists in the merchant's own systems. Order database, auth logs, device telemetry, logistics API, support transcripts. Nothing needs to be discovered. It needs to be joined, formatted to the network's requirements, and filed inside a window that is typically 20 to 45 days and absolutely unforgiving.

Losing at 12 percent was never a verdict on the merchant's case. It was a verdict on an ops queue being asked to do a data pipeline's job by hand

The method has two halves. The first is automated evidence assembly: map each reason code to its evidence checklist, pull each item from the systems that hold it, and generate the representment packet the moment the dispute lands, not the week the deadline threatens. Win rates on well-assembled packets do not sit at 12 percent. Across the industry they land in the 40s and above for merchants who fight with complete evidence, and the delta is almost entirely completeness and timing rather than eloquence.

The second half is knowing when not to fight. Each dispute gets an expected-value call: estimated win probability for this reason code and evidence profile, times the amount, against the fee and the effort. A calibrated model does this scoring; low-value low-odds disputes get conceded on purpose, which is a decision rather than a default. And the whole operation runs next to a ratio watch, because the networks put merchants above roughly one percent dispute rate into monitoring programs with fines attached. Fighting harder is not a substitute for disputing less; the same evidence graph that wins representments also feeds the fraud rules that prevent the next one.

We assembled this exact loop for a payments-adjacent platform: reason-code routing, automated evidence pulls, packet generation, an EV threshold tuned to their fee schedule. Recovered disputes became a line item that had simply not existed before, and the finance team's favorite property of that line is that it is nearly pure margin. The revenue was already earned once.

The win rate was never about writing a better letter. It was about showing up with the database instead of a paragraph.

The diagnostic: export last month's disputes with reason codes, and for the top code, check whether the rebutting evidence exists in your systems. It almost always does. Then check how many packets went out with all of it attached. That gap, times your average dispute, is the annual figure you have been calling a cost of doing business.

TensorLabs builds the pipelines that close that gap, evidence graphs, packet automation, the EV scoring that decides which fights are worth having. If your dispute line has quietly become one of your larger marketing budgets, write to us with your monthly volume and win rate. Two numbers are enough to size what is recoverable.