The compatibility file was wrong, and it was costing more than it earned

Over six years the brand had treated its compatibility data as a marketing surface. Each time a part was “probably fine” on another trim, the row went in; nobody ever took a row out. The file had grown to 1.1 million entries and was wrong often enough that a quarter of all orders came back.

The accounting made this look survivable and it was not. Gross revenue was flattering, but a returned mechanical part is usually unsellable — it has been handled, sometimes half-installed — and each one carried a return shipping cost, a processing cost and, frequently, a one-star review from someone who had spent a Saturday discovering the problem. The inflated data was buying traffic and paying for it twice.

We shrank the file until it was true.

• Row-level audit against engineering specification, prioritizing the vehicle-part combinations that generated the most “does not fit” returns.
• Unverifiable rows deleted rather than flagged — 480,000 of them, roughly 44% of the file.
• A change-control process so a new row requires a verified reference, closing the tap that had caused the problem.
• The lost visibility recovered honestly through listing SEO on the vehicles the parts genuinely fit, where the brand had never competed properly.

Returns fell from 26% to 9%, and net revenue — after returns and their costs — rose 44% even though the catalog now appears for far fewer vehicles.

Ratings on the top twenty ASINs climbed from 3.7 to 4.3, since the dominant source of angry reviews was people who had bought a part that could never have fitted. Deleting 480,000 rows of data felt like deleting demand; it was deleting a cost center.