The company sold cars online. The inventory sat on the lots of more than three hundred physical stores. The online business was a separate division with its own targets.
Conversion on the site was fine. Customers arrived, chose a car, committed.
Then six deals in ten never finalised. Fall-through was the KPI everyone in the business was watching, and it would not move.
Never for the same reason twice. A store would take four days to answer a question about a vehicle. Paperwork would go missing. Damage nobody had recorded would turn up on a car that had been photographed clean. Every failure had an explanation, and every explanation was plausible on its own.
The company read this as an operations problem. Slow processes, poor handoffs, systems that did not talk to each other. That is the natural reading, and it is the wrong one.
What the reviews were saying
The customer reviews had no middle. Twenty-three per cent were one star and almost every one of them described a defect. Sixty-three per cent were five star and almost every one of them named a person, someone who had chased something down or gone well past what the job required.
A healthy process produces a cluster of fours. This was a barbell, and a barbell means the outcome is not coming from the system. It is being delivered, or withheld, by individuals deciding case by case.
That is the point the question changed.
What was actually happening
Follow the money to the store manager.
Sell a car on the lot and the manager books the margin and the finance and insurance income that goes with it. Sell the same car online and the attribution goes to the online channel. The manager loses the margin, loses the F&I, and then has to go and source a replacement unit for the space on the forecourt.
Nobody was sabotaging anything. Three hundred managers were each responding rationally to the way they were paid, and the sum of those rational responses was a channel that could not get inventory out of its own network.
That is the tell, and it generalises. When failures are individually explainable and collectively relentless, stop reading it as a process problem and find out who loses money when the process works.
What I did
The fix was smaller than the diagnosis.
The store keeps the credit. Margin and F&I are attributed to the store that held the car, exactly as they would be on a forecourt sale. The online business still reports the deal as its own acquisition channel, because that is a different question and a different number.
Everyone had assumed the credit was zero-sum. One sale, one owner. It never was. Marketing attribution and P&L credit answer different questions, and nothing requires them to point at the same place.
Months of work had gone into mapping the process. The process was fine. What needed changing was who had a reason to help the deal through.
Outcome
Closed deals went from 40% to 72% by the time I finished on the project.
Same stores. Same cars. Same website. The only thing that changed was who got paid when the car sold.
Pattern
The diagnosis was cheap once the question changed. Weeks of work had gone into mapping processes that were not broken. The organisation could see every individual failure and none of the shape they made together, because nobody had asked what the failures had in common on the incentive side rather than the process side.
Incentive problems present as operations problems. They almost always do.