How accurate is QuantDoc? Here are the measured numbers.
What we optimise for
No invented figures
QuantDoc shows the quantities, rates and amounts it read from the document and never fills a gap with a guess. A figure can still be misread, which is why each one links to its source.Every figure traceable
Each extracted figure links to the page and line (or sheet and row) it came from, so you can check it in seconds.No false equivalence
Different specifications, units or scopes are kept apart. A cheaper total that leaves work out is not called cheaper.Honest uncertainty
If figures or scope can't be verified, QuantDoc says no supplier can be recommended yet and lists what to check.
fresh7: 16 international typed quotes
- Line items found
- 85%
- 93 of 109 labelled lines. Recall only, not overall accuracy.
- Line amounts correct
- 62%
- 39 of 63 labelled amounts; the rest unread or wrong.
- Wrong figures
- 11
- 8 unit rates, 3 amounts, in 3 quotes. All 3 were flagged for review.
- Wrong supplier named
- 0
- No false equivalence; all 4 comparisons flagged for review.
Unit rates were the weakest field: 24 of 66 labelled rates were read correctly. The cheapest compliant supplier was identified in 1 of the 4 comparisons; in the 3 comparisons with a clear cheapest quote, QuantDoc asked the reviewer to settle scope before naming one.
The same frozen version was also run with the AI-assisted reading, in a test environment: the same 85% of line items, 42 of 63 amounts correct and 7 wrong figures, again all in quotes flagged for review. On quantdoc.online, AI-assisted reading runs on paid plans and the 60-day trial for documents the standard reader can't read well; the free preview never uses it.
Earlier sealed sets
fresh6: 32 international quotations · 1 November 2026
32 real quotations from the UK, New Zealand, the US, Canada, Australia and Ireland (18 public sources), with 4 comparisons. Deterministic reader (no AI). Typed PDFs (30): 72% of line items found (236 of 329); 91% of the line amounts it read were correct (184 of 203). 12 figures in 3 quotes were wrong: quantity and rate swapped on one invoice, prices including tax read as excluding it on another, and one wrong rate on a price list. The first two causes have since been fixed, but that is not yet re-measured on a new sealed set. One 83-line price list was only partly read (30 lines). Scanned PDFs (2): 3 of 15 lines found. Comparisons: no false equivalence, no wrong supplier named with confidence, all 4 flagged for review, and no supplier recommended in the 2 comparisons that had a clear cheapest quote.
fresh5: 26 quotations, UK, US, Australia and Canada · 31 October 2026
26 real quotations (11 public sources), with 2 comparisons. Deterministic reader, no AI. Typed PDFs (21): 55% of line items found (46 of 84), and every figure read matched the document (0 wrong). One 24-line price list was not read at all. Scanned PDFs (5): 21% of line items found. Comparisons: no false equivalence, no wrong supplier named with confidence, both flagged for review.
fresh4: 24 Australian quotations · 31 October 2026
24 real Australian supplier and contractor quotations from five public sources, with 3 comparisons. Deterministic reader (no AI).
| Document type | Quotes | Line items found | Line amounts correct (of those found) | Wrong figures |
|---|---|---|---|---|
| Typed (digital) PDF | 15 | 67% (63 of 94) | 100% (46 of 46) | 0 |
| Scanned typed PDF | 7 | 77% (23 of 30) | 48% (11 of 23) | 0 |
| Handwritten | 2 | 2 of 3 | 2 of 2 | 0 |
Comparisons: none of the 3 treated different products as equivalent, none named a wrong supplier with confidence, and all 3 were flagged for human review. The cheapest compliant supplier was identified in 1 of 3.
fresh3: 51 Australian quotations · 28 October 2026
51 quotations. 54.8% of line items found, with 4.9% of figures wrong. Every one of those was flagged for review. The sets are small, so treat these as indicative rates, not guarantees.
Known limitations
- Some line items are missed, especially in long price lists and unusual layouts. When QuantDoc sees priced or numbered lines it didn't read, it flags the quote as possibly incomplete and lists them so you can add or correct them.
- Some figures are read wrongly, mostly unit rates in price lists and quotes with unusual layouts. In the latest sealed test every quote with a wrong figure was flagged for review, but check quantities, rates and tax treatment against the source before relying on them.
- No supplier is recommended until you have reviewed each quote and marked its review complete. Comparisons of different designs or options are explained, not ranked.
- Scanned documents are read by OCR and are less reliable than typed PDFs. Every figure read that way is marked for checking.
- Handwritten quotes are not reliably supported.
- QuantDoc compares the quotes you upload. It does not check prices against the market or judge whether a specification is fit for purpose.
- A comparison is a decision aid, not professional advice. Check important figures against the source before accepting a quote.
See how QuantDoc works for the matching rules, and security & privacy for how documents are handled.
Frequently asked questions
Does QuantDoc read every line of every quote?
No. On the two latest sealed sets of real typed PDF quotes it found 72% and 85% of the line items. That is line-item recall, not overall accuracy: fewer of the unit rates and amounts on those lines were read correctly, and some figures were read wrongly (11 in 3 of 16 quotes in the latest set). Each quote with a wrong figure was flagged for review, but check the source pages for anything that matters. Lines are never invented.
Does QuantDoc automatically pick the cheapest supplier?
Only when the figures are verified and the quotes cover the same scope with equivalent items. Otherwise it says no supplier can be recommended yet and lists what to check. In the latest sealed test it named no wrong supplier with confidence and flagged all four comparisons for review.
What happens when QuantDoc isn't sure about a figure?
The figure is marked for review with a link to the page and line it came from. Figures read from a scan, or by AI, or that don't add up to the quote's total, are never used for a confident recommendation until checked.
Are these results measured with AI?
Mostly not. The fresh3 to fresh6 results use QuantDoc's deterministic reader only. The latest set, fresh7, was measured twice on the same frozen version: once with the deterministic reader and once with the AI-assisted reading in a test environment. Both found the same line items; the AI-assisted run had 7 wrong figures instead of 11. On quantdoc.online, AI-assisted reading runs on paid plans and the trial for documents the standard reader can't read well; its figures are accepted only when they appear in the cited source line, and they are flagged for review.
Make the next quote comparison clearer.
Upload competing quotes, check the evidence and see what each option really costs.
Free preview, no card required. Paid plans: 60-day trial, card required.