Compliance resources

Tendermark is a tender comparison tool for building surveyors. The surveyor authors or uploads a Schedule of Works — the structured list of work items priced in a tender. Contractors price against that Schedule, either by completing an online pricing form or by returning a document (PDF, spreadsheet, scan, or photograph). Tendermark extracts the contractor's prices and produces a side-by-side comparison table.

What the AI does: Tendermark uses a large language model (Anthropic Claude) for four distinct tasks.

• Contractor return matching. For every uploaded return — whether Excel, CSV, PDF, Word document, scanned document, or photograph — Claude reads the content and maps each contractor price to the correct Schedule of Works line item. The file is preprocessed into a readable form first (cell values extracted from spreadsheets, text extracted from PDFs); Claude then handles all line-item matching. This is the primary use of AI in Tendermark and applies to every contractor return regardless of format.
• Schedule of Works parsing (fallback only). When a surveyor uploads a SoW in an unusual format — a scanned document or prose-style Word file — Claude identifies the line items. Standard formats such as Excel and CSV are parsed by code without AI involvement.
• Tender Analysis Report narrative drafting. Three sections of the report are drafted by Claude: Arithmetical Accuracy (discrepancies between each contractor's stated and implied totals), Anomaly Handling (observations on flagged pricing items), and Summary of Submissions (price-difference analysis across contractors). These drafts appear in the report editor marked as AI-generated, pending the surveyor's review and approval before export.
• Pricing Commons normalisation and classification (opt-in users only). For users who have opted in to the Tendermark Pricing Commons, two background AI steps run automatically after each parsed return: (1) each line-item description is rewritten by Claude to remove all identifying detail before any contribution is stored, and (2) the normalised description is classified into one of 18 trade theme tags to place it in the correct pool. Neither step is visible during the normal tender workflow. The surveyor's pricing comparison is not affected.

What the AI does not do: Tendermark makes no professional judgements, gives no recommendations, and produces no output presented as a professional opinion. It extracts, organises, and drafts — the surveyor interprets and decides. Every extracted figure includes a confidence score so the surveyor can see where the AI is uncertain. Anomalies are flagged for the surveyor's review, not acted upon automatically. The Recommendation section of the report — the surveyor's professional opinion on which tender to accept — is never AI-drafted and requires the surveyor's own input.

What the surveyor must do: Review all extracted output before relying on it. Verify anomaly flags with professional judgement. Read and edit the AI-drafted narrative sections — they are generated drafts, not finished text. The Tender Analysis Report shows the surveyor's name as the author; the surveyor remains professionally accountable for all of its content. No report should be sent to a client without the surveyor having reviewed and approved the output in full.

The cleaner audit trail — contractor invitations: When a contractor prices online via Tendermark's invitation link, their figures arrive already structured — no AI extraction is involved. Every action (viewed, saved, submitted) is logged with a timestamp and the contractor's IP address. This produces the most transparent possible audit chain and is the recommended approach where contractors are willing to engage digitally.

Under s4.4, your firm must be able to provide written information on request about the type of AI system used, its ways of working and limitations, your due diligence, your risk management, and your reliability decisions. This section covers the first two; the sections below cover the rest.

s3.1 permits uploading private and confidential data to an AI system only where there is express written consent in advance from affected stakeholders, and your firm has taken reasonable steps to satisfy itself the upload does not pose an unacceptable risk. In practice: confirm your client engagement terms or data processing agreement permit AI-assisted processing of their pricing information before uploading, and record this page (and the supplier reference card below) as the reasonable steps taken. Tender pricing documents do not require end-client personal data — upload the contractor return, not the client instruction.