We tracked 262,000 fashion products hourly for 6 months. 1 in 17 "sales" turned out to be fake

We launched TrendCompare today (a price tracker for Pakistan's fashion e-commerce), but this post isn't about the launch — it's about what the data taught us, because I think it applies to e-commerce everywhere.

We record the price of every product across 22 fashion brands, every hour. Six months of history, ~150K price-change records. Three things surprised us:

1. The "original price" is a creative writing exercise. 1 in every 17 sale listings advertises an original price more than TRIPLE the actual selling price. The pattern is always the same: quietly raise the anchor price 3–7 days before a sale event, then "discount" it back.

2. Real discounts are actually common — which is what makes fakes work. The average genuine discount is 41%. Shoppers see enough real deals to trust the banner, and that trust is exactly what the inflated anchors exploit.

3. A price snapshot is worthless; only history has value. Every fake discount is invisible at any single moment in time. It only becomes visible when you've been recording for weeks before the sale. This is why price-comparison sites that poll daily miss all of it.

We ended up publishing a CPI-style monthly index from the data (base March 2026 = 100 — clothing is actually 9% CHEAPER than March, against the inflation narrative) and made the whole dataset free under CC BY.

Question for this community: for those of you building data products — how do you handle the incumbents-hate-you problem? The brands whose prices we track are also the advertisers every publisher in our market depends on. Curious how others have navigated being commercially inconvenient.