AI Quote Comparison for Estimators: Strengths and Limits

What estimators actually need from quote comparison

Comparing subcontractor and supplier quotes is mostly three jobs: reading documents that all look different, working out which lines correspond to each other, and doing careful arithmetic so the totals are on the same basis. Each job suits a different kind of tool.

Where AI helps

• Reading unusual layouts. Quotes arrive as tables, paragraphs, scanned letters and spreadsheets with merged cells. Language models are good at recognising that “Cartage” in one document and “Delivery – tipper loads” in another are both freight.
• Interpreting terms. Picking exclusions, validity periods and payment terms out of free text.
• Drafting. Turning a list of gaps into a polite, specific clarification email.

Where AI should not be trusted on its own

• Arithmetic. Language models can produce plausible but wrong numbers. Rates, extensions, GST and totals should be calculated with deterministic decimal arithmetic and reconciled against the document.
• Equivalence. Similar wording is not proof that two items are the same specification. 20mm drainage gravel is not 20mm crushed aggregate; a 5 t excavator is not a 13 t excavator.
• Missing information. A model asked for a total may quietly fill a gap. A TBA freight line must stay unknown.
• Market judgement. A model has no reliable knowledge of today’s local rates for your materials and trades.

A worked check you can do by hand

The worked example below uses three synthetic aggregate quotes, read by QuantDoc’s parser without AI. You can verify its key figures yourself:

Coastal: 40,000 kg × $0.0572/kg = $2,288.00 inc GST
         $2,288.00 ÷ 1.1 = $2,080.00 ex GST
         40,000 kg ÷ 1,000 = 40 t → $2,080.00 ÷ 40 t = $52.00/t ex GST
Ridgeline: 40 t × $48.50/t = $1,940.00 ex GST
Metro: bedding sand 8 m³ × $70.00 = $560.00, but the quote shows $590.00
       difference = $590.00 − $560.00 = $30.00 → flagged as a possible calculation issue

Note what the engine did not do: it did not assume a density for Metro’s cubic metres, did not treat Coastal’s TBA freight as zero, and did not treat “drainage gravel” as the same product as “crushed aggregate” without confirmation. The result is labelled INCOMPLETE, which is the honest answer.

How to test any AI quote comparison tool

1. Take a tender you have already levelled by hand.
2. Check five figures in the tool’s output against the source documents. Can you find each one?
3. Insert a deliberate arithmetic slip in a test quote. Is it flagged?
4. Leave a line as “TBA”. Does the total treat it as zero?
5. Ask the vendor which steps use AI, whether it can be turned off, and where your documents go.

How QuantDoc uses AI

QuantDoc uses Anthropic Claude models only to read documents its standard reader can't read well (paid plans and the trial), and flags every AI-read figure for review. Extraction, matching rules, unit conversion, GST handling and every money calculation run as deterministic code, and every figure carries a page/line or sheet/row reference. QuantDoc does not check market prices or consult external pricing sources, and does not train models on customer documents. If AI processing is enabled, extracted text may be sent to Anthropic’s API to interpret the document.