A Customer's AI Report Said 13%. The Math Said 4%.

One customer accounted for roughly half of this company's revenue. The company, in turn, handled about 90% of that customer's work. That kind of concentration is either a partnership or a pressure point, depending on the day.

Pricing between the two gets reviewed every six to twelve months. This round, the customer arrived with something new: a competitive pricing analysis built using AI tools. Fifteen pages. The conclusion was that the company's pricing sat 13% above its competitors', and the customer wanted that difference back.

In the first seven months of the year, this single account had produced $423,000 in profit. A 13% concession would have taken it to break-even.

John Hoffman — FGP's CFO Lead, and the advisor working this account — did not argue with the report. He rebuilt it.

He ran the comparison by hand first and landed at 4.4%. Then he ran the same data through an AI model as an independent check and got 4%. Two methods, one answer, and neither of them was 13.

Next he called the customer's president directly and asked for the exact competitor information the analysis had been built on. He got it — with one limitation he named out loud rather than glossing over: he never asked for, and never saw, the competitor's actual invoices. Working from the customer's own inputs, the number came back the same. Roughly 4%.

The gap traced back to a $13 price difference between two locations. That $13 represented 4%. Somewhere in how the questions were asked, a dollar figure appears to have come out the other side as a percentage.

Manual analysis first, AI second. The order matters. Running the numbers by hand before touching a model means the model confirms your thinking instead of setting it.

Then, go to the source. Rather than debating the report, John asked the customer's president for the underlying data. That single move changed the conversation from "your number is wrong" to "let's look at the same inputs together."

From there he worked both sides of the margin rather than only the customer's side — taking the question upstream to the company's suppliers as well, so that any adjustment the customer received would not come out of the client's margin alone.

Worth naming what he was working against. The client does not use AI and does not want to. That is a defensible position for an owner to hold. It also meant that when fifteen confident pages landed on the table, there was no one inside the business who could open them up and find the error. The report showed conclusions, not calculations. There was nothing to put a finger on.

The discrepancy was traced to a single line item and documented in a form the customer could follow — which is what turned a demand into a conversation. The negotiation is ongoing.

The relationship held. And the owner walks into the next meeting with a number he can defend line by line. That is a different posture than conceding to a report he has no way to check.