Our own test · November 2026
How accurate is QuoteDesk? We tested it.
89 request lines we wrote, run through QuoteDesk's list reader and price matcher against four sample price lists. It priced the lines it should, left the unclear ones for a person, and priced one line it should have left alone.
Read this first. This is a test on made-up data, not a customer result. We wrote the request lines and the expected answers ourselves, for our sample price lists. It tests the part of QuoteDesk that finds the product and sets the price. It does not test the AI model that reads free-text emails, so every line here was written as a list ("40 x ..."). We made two fixes after the first run, so these figures are not a blind test.
What we tested
Four sample price lists, all made up: fasteners and fixings (62 products), IT reseller (29), bearings and hydraulics (29) and a coffee brand's Shopify export (10). For each we wrote request lines of the kinds a buyer sends: a product code, a clear description, a description missing a size or grade, a size the supplier does not stock, and an item that is not on the list at all. For each line we wrote down what a careful person would do: price it from a named product, or leave it blank because it is unclear.
The rule being tested is the one QuoteDesk works to: a line is priced only on a product code or a clear match where every size and grade in the request is in the product. Anything else is left blank for a person.
The results
| Price list | Lines | Priced correctly | Left blank, correctly | Priced wrongly |
|---|---|---|---|---|
| Fasteners and fixings | 30 | 20 | 10 | 0 |
| IT reseller | 22 | 16 | 6 | 0 |
| Bearings and hydraulics | 21 | 16 | 4 | 1 |
| Wholesale (Shopify export) | 16 | 10 | 6 | 0 |
| All | 89 | 62 | 26 | 1 |
The one miss
The line "20 x 1/2in hydraulic adaptor" was priced as a 1/2in male to male adaptor. The price list also has a 1/2in 90 degree elbow adaptor whose description does not use the word "hydraulic", so the matcher saw the first as the clearer match. A person would have asked which type was wanted. This is the kind of mistake that comes from how a price list is worded, and it is why QuoteDesk shows every match and every draft is checked by a person before it is sent.
What the first run found
The first run priced two lines wrongly. "10 x tea bags" was priced as "Cold Brew Bags" because the matcher gave too little weight to the word "tea", which is in no product. And a clear-match check was judged partly on sizes the request never mentioned. We fixed both, added a test for the first, and re-ran. The result above is after those fixes.
How fast is it?
We timed the live demo on 2 Nov 2026 from a server in the cloud, with a sample price list and the sample request for that list, from sending the request to receiving the priced draft: a fasteners request (4 lines) took 10.8 seconds and an IT reseller request (11 lines) took 12.6 seconds. That includes the AI model reading the request. It is two runs, not an average, and a larger price list or a longer request will take longer. We will add more runs as we time them.
What this does not show
- How accurate QuoteDesk is on your price list and your buyers' wording. Try it on your own list with the free demo.
- How well the AI model reads requests written as free text. We have not published a measure of that.
- Any effect on win rate or revenue.
The test is in our code (web/scripts/accuracy-test.mjs with the lines in web/test/data/accuracy-lines.json) and is part of our automated tests, so we notice if a change makes the result worse.