A tender package goes out to five subcontractors. By the return deadline, five replies have arrived. They are all different.
One subcontractor has sent a polished PDF schedule — a numbered table with line items, quantities, units, and rates, formatted to two decimal places. One has sent a Word document where the prices are embedded in prose paragraphs. One has photographed a typed quote on their smartphone and emailed the image. One has replied on headed notepaper with a formal letter and a hand-written table. One has sent an email with the prices in the body text and a PDF attachment that is a scanned copy of their internal quote sheet.
All five must be mapped back to the master Bill of Quantities so that a meaningful, apples-to-apples comparison can be made. The line items must be aligned. Missing prices must be identified. Provisional sums must be handled separately. The cheapest compliant price per item must be established before a recommendation can be put to the client.
The tender deadline is tomorrow.
This is the subcontractor comparison problem — and it is one of the most time-consuming, error-prone tasks in routine QS practice.
The Scale of the Manual Task
The traditional manual approach involves reading each quote in full, manually transcribing every priced item into a comparison spreadsheet, cross-checking for alignment errors, resolving unit inconsistencies, calculating variance, and preparing a summary recommendation. For a single trade package with 20 to 40 line items, this process typically takes three to four hours.
Under tender deadline pressure — which is the norm, not the exception — it is rushed. Errors creep in. The QS who is working under pressure to get five comparisons done before close of business does not have the cognitive bandwidth for the kind of meticulous cross-checking that the task demands.
The errors that result are predictable and consequential. A line item priced by one subcontractor in pounds per linear metre when the BoQ is in pounds per square metre — missed in the transcription, applied to the wrong unit, and creating a 10x cost error in the comparison. A provisional sum item included in one subcontractor's total but excluded from another's — creating an apparent price gap that disappears when the provisional sums are stripped out. An arithmetic error in a hand-written quote — transcribed faithfully but incorrectly.
These are not exotic failures. They are the routine casualties of manual transcription under time pressure.
Why Format Diversity Is the Core Problem
The subcontractor comparison problem is not primarily a measurement problem — it is a format problem. The price information is all there. It is present in every document. The challenge is extraction and normalisation.
A structured PDF table is relatively straightforward to parse — the prices are in cells, the line items have identifiers, the arithmetic can be read. A prose paragraph email requires the QS to extract prices from continuous text, identify which line item each price relates to, and decide how to map it to the BoQ description. A handwritten photograph requires OCR — and at 150 DPI on a phone camera, the OCR is unreliable for prices that are written small.
The format diversity exists because subcontractors send quotes in whatever format is convenient for them. It is not malicious — it is just how small businesses operate. But it creates work for the QS that has nothing to do with professional expertise and everything to do with data entry.
What the AI Quote Scraper Workflow Does
The AI Quote Scraper workflow uses ChatGPT Advanced Data Analysis to perform the extraction and structuring step — converting all five formats into a common structured format — and Microsoft Excel to produce the comparison matrix, calculate variance, and identify the cheapest compliant price per line item.
The five steps:
- Step 1 — Scan and save all quotes as PDFs. Born-digital documents are converted via Print-to-PDF. Physical documents and photographs are scanned at 300 DPI minimum. Files are renamed using a consistent position-numbered convention.
- Step 2 — Upload all PDFs to ChatGPT Advanced Data Analysis. ChatGPT with Advanced Data Analysis — not standard ChatGPT — must be used, as the feature is required for file uploads and code execution. All five PDFs are uploaded simultaneously to give the model full context.
- Step 3 — Run the structured extraction prompt. The prompt instructs ChatGPT to act as a senior quantity surveyor, extract all priced items from each document, map them to the BoQ line item descriptions, flag missing prices, and output a structured CSV comparison table.
- Step 4 — Build the comparison matrix in Excel. The CSV is imported into Excel. Conditional formatting is applied: green for the cheapest price per row, amber for prices more than 20% above the average, variance percentages are calculated, and the cheapest compliant subbie per item is identified.
- Step 5 — Verify against original documents. The AI draft is reviewed against the original PDFs. A minimum 20% random sample of rows is checked. Discrepancies are corrected. The 8-point QS Risk Checklist is completed. The verified matrix is saved with a date suffix.
The 20% Verification Sample: Why It Is Non-Negotiable
The most important step in the workflow is Step 5 — and it is the step most likely to be skipped under deadline pressure.
ChatGPT Advanced Data Analysis makes fewer errors than manual transcription. It also makes different errors. It can misread a digit in a photograph (reading 8 as 0), misalign a line item when the source document's structure is ambiguous, or — on rare occasions — generate a plausible but fabricated price where the source document was illegible.
The 20% random verification sample catches these errors before they propagate into the final recommendation. It also provides the documented evidence trail that demonstrates due diligence if the comparison is ever challenged.
The verification log — a simple table recording BoQ number, subcontractor, AI-extracted price, source document price, and a pass/fail — is the QS's professional insurance policy. It demonstrates that AI output was reviewed, not relied upon blindly.
What "Compliant" Means — and Why It Matters
Not all quotes are directly comparable. A subcontractor who has failed to price a material proportion of the BoQ items, or who has applied a materially different scope, is not a compliant submission. Including them in the unit-rate comparison without flagging the scope difference produces a misleading result.
The AI Quote Scraper workflow requires the QS to assess compliance before including each subcontractor in the comparison. The basis of exclusion — where a subcontractor is excluded from the cheapest-price calculation because their submission is non-compliant — must be documented in the project file.
The final recommendation to the client must articulate not just which subcontractor is cheapest, but why. Lowest price is not always the right recommendation. Programme track record, quality references, and capacity to deliver may all be relevant factors that the QS must weigh and record.
The RICS Professional Standards Dimension
When a QS recommends a subcontractor to a client or project manager, they are providing a professional opinion. That opinion carries obligations that AI cannot discharge.
All compliant quotes must be evaluated on the same basis. Non-conforming submissions must be excluded transparently. The basis of the recommendation — lowest price, best value, programme advantage, or quality — must be recorded in the project file.
RICS has published guidance on AI in surveying practice. The established principle is clear: AI assists the professional; the professional remains accountable for every output. The verification step is not a quality check on the AI — it is the professional act that makes the QS the author of the final recommendation.
Related Resources
The complete five-step AI Quote Scraper workflow — including the structured extraction prompt, the Excel comparison matrix template, the conditional formatting setup, and the 8-point QS Risk Checklist — is documented in The AI Quote Scraper. That guide includes a worked example using a scaffolding trade package with five different document formats, the file naming convention, and the professional verification framework.
If you are working upstream from the tender analysis stage and need to produce the BoQ from which the comparison will be made, The 2D PDF Takeoff Accelerator covers the Kreo + Excel workflow for AI-assisted Bill of Quantities production — the document that anchors every subcontractor comparison.
Stop Guessing, Start Systemising
The subcontractor comparison problem is not a measurement challenge — it is a data extraction and normalisation challenge. AI handles the extraction; the QS handles the professional judgement about compliance, scope, and recommendation.
The AI Quote Scraper has the complete five-step workflow, the extraction prompt, the Excel comparison matrix template, and the 8-point QS Risk Checklist. Three hours forty-five minutes of manual work reduced to approximately 80 minutes — with full professional accountability at every step.
Grab it here.
Practical AI for Professionals