How UponSale measures a price

UponSale is a price-intelligence service, not a shop. We do not sell anything, we do not take payments, and we have no stock. What we do is watch what independent retailers actually charge for the same product, record it over time, and tell you plainly whether today’s price is good.

Right now we track 14,777 products across 10,979 live offers from 10 retailers, and hold 17,444 recorded price observations.

1. Where the data comes from

Most online stores publish their catalogue in a machine-readable form so their own product pages can display it. We read exactly what any visitor’s browser receives, at a deliberately unhurried rate.

• We check and follow each retailer’s stated access rules before reading anything, and re-check them regularly.
• One page at a time from any one store, with a pause between requests and a firm hourly ceiling.
• Where a store can tell us nothing has changed, we take its word for it and read nothing further.
• If a store asks us to slow down we slow down, and if it keeps failing we stop and return later.
• We never work around an access control, and we never buy collected data of uncertain provenance.

We identify ourselves on every request, with a link explaining who we are. If you run a store and want us to read less often or not at all, tell us and we will.

2. Matching the same product across stores

A comparison is only useful if both sides are genuinely the same item. We match in a strict order of confidence:

1. Exact GTIN / UPC / EAN, with checksum validation.
2. Brand plus exact manufacturer part number.
3. Brand, model and an identical variant fingerprint.
4. Normalised title similarity, but only when attributes agree.

Some attributes can never disagree: capacity, size, condition, region, voltage, carrier lock, pack count and bundle status. If two listings conflict on any of those, they stay separate products no matter how similar their titles are. Anything in the uncertain band goes to a human review queue instead of being merged automatically.

3. What “verified” means

An offer can only become the best verified price if all of the following hold:

• It was checked recently enough to be inside the refresh window for its product tier.
• It is matched to the exact variant, not a near neighbour.
• The retailer reports it as available.
• It has no unresolved price anomaly.

We never present a stale price as current. When our latest check is older than the freshness window, the page says so rather than quietly showing yesterday’s number.

4. Anomaly detection

Retailer feeds occasionally report a zero price, a category price instead of an item price, a currency change, or a listing that has quietly been replaced by a different product. Before a price change is published we check for:

• a drop of more than 70% without corroboration from another store;
• a price of zero, or a currency that changed since the last check;
• variant attributes or a product title that changed materially;
• a lone offer far below every other store tracking the same item.

A flagged price is still recorded and still visible — but it cannot win best verified price until it is corroborated or reviewed.

5. Price history and market statistics

We write a history row when a price changes, when stock state changes, or as a daily checkpoint — never once an hour regardless of whether anything happened. Daily market low, median, interquartile range and high are computed from the last observation per store per day, so a retailer we happened to poll more often cannot skew the median.

6. The Deal Score

A deterministic 100-point model. The same inputs always produce the same number, and every product page shows the full breakdown:

7. Advertised discount versus real discount

A retailer’s crossed-out price is a claim, not a measurement. We store both: what the store says the discount is, and what it is against the price the product genuinely sells for across the market. Where those two numbers diverge sharply, we say so.

8. What we deliberately do not do

• We do not claim to have the lowest price on the internet — only the lowest among the fresh offers we track.
• We do not publish reviews or ratings, because we have not tested these products.
• We do not copy retailer product descriptions, and we do not re-host their images.
• We do not predict future prices. Buy-or-wait guidance describes observed history, nothing more.
• We do not let any commercial arrangement change which offer ranks first.

9. When we get it wrong

Prices change constantly and retailer data is sometimes inconsistent. Always confirm the final price on the retailer’s own page before buying. If you spot a mismatch, a wrongly merged product or a stale price, report it — corrections go to the front of our queue.

Common questions