How we measured 83% of was-prices were fake (methodology deep-dive)

#3 — How we measured 83% of was-prices were fake (methodology deep-dive)

Tags: data, datascience, webdev, tutorial

We published a finding last week: 83% of "was" prices in our catalog were never the price. Here's exactly how we measured it, so you can replicate or critique it.

The dataset: every product in Fazy's catalog showing a reference ("was") price at least 5% above its current price, with the current price read within the previous three days.

The test per product:

1. Pull the full price history: every recorded observation with timestamps.
2. Require at least 30 days of tracking and 4+ observations. Anything thinner is excluded.
3. Check: did the product ever sell within 2% of the claimed was-price? If no, it fails.

Results: 83% failed. Average watch time was eight weeks with 13 observations each, so this isn't a thin sample artifact.

The category split surprised us: wellness 97%, cosmetics 96%, home 91%, footwear 91%, electronics 88%, appliances 75%, apparel 68%. Apparel's relative honesty makes sense: real seasonal markdowns exist there.

The ugliest stat: 43% of products showed the same price at every single check. Not a sale with a fake anchor — just a fake anchor.

Replicate it: our price history is queryable, free, no key: https://fazy.com/developers and MCP at https://fazy.com/mcp. The get_price_history tool returns the dated facts.

I'm the founder of Fazy, so read skeptically. The methodology is the whole point.