AI at Workfrom Acuity AI Advisory

Case studies · Construction contractor, Ireland

Claude tender drafting connected to Microsoft 365

By Ger Perdisatt · Reviewed

A construction contractor in Ireland had two people spending most of their week writing tender and pre-qualification answers, searching years of past bids across SharePoint and shared drives. We set up Claude Team on the firm's own licences, connected to its Microsoft 365, with an indexed library of staff CVs, project references and past answers. For each scored question, Claude drafts an answer within the word limit and names the document behind every claim, so the bid team starts from a sourced draft rather than a blank page. It is in use on the firm's own Claude Team workspace. The time saving is an estimate.

In use In use on the firm's own Claude Team workspace.

Client identity withheld to respect confidentiality.

The starting point

Public tenders and pre-qualification questionnaires ask scored questions with strict word limits. The evidence for each answer, such as who did what on which project, sat in past bids, CVs and project sheets spread across SharePoint and shared drives.

Writing an answer meant finding that material first, then cutting it to length. The firm had already lost a pre-qualification over a word count.

What we implemented

The firm bought its own Claude Team licences, so the workspace and its data belong to the firm, not to us.

Claude reads files from the firm's existing Microsoft 365 through Anthropic's Microsoft 365 connector, rather than copying them into a separate system. The firm's IT provider approved the connector for the firm's tenant.

We built the library in order: staff CVs first, then project references, then past answers. Each set was checked before Claude drafted from it, because a draft is only as good as the evidence behind it.

For each scored question, Claude drafts within the word limit, cites the document behind every claim, and leaves a marked note where the library holds no evidence, instead of inventing it.

Using it, step by step

  1. 1What the person supplies

    The tender question, its word limit and how it will be scored. The library of CVs, project references and past answers is already in place.

  2. 2What happens next

    Claude searches the library, drafts an answer within the limit, lists the sources it used and marks anything it could not find.

  3. 3What they check before using it

    The bid team checks every claim against its source, fills the marked gaps, confirms the word count the way the tender counts it, and rewrites for marks. Nothing is submitted without that review.

Example output

A draft answer to a scored health and safety question. Each claim carries a numbered reference to a document in the bid library, the word count is shown against the limit, and a note marks the one fact the library does not hold. For a full worked version on made-up material, see the tender response demonstration.

Claude's draft answer to a fictional tender question on health and safety management, limit 400 words, showing 361 of 400 words. Numbered references in the text point to a sources panel listing a staff CV, a project sheet for a school extension and a safety statement, each with its bid library folder. A highlighted note asks the bid team to add an accident frequency rate that is not in the library.
Illustrative example using fictional data. A draft tender answer beside its sources and word count, with a marked gap for the bid team.

Human checks

  • People check every claim, citation and word count before anything is submitted.
  • Where the library holds no evidence, Claude leaves a marked gap for the bid team instead of filling it.
  • Claude can reach what each person can open in Microsoft 365. Naming a folder in a request does not limit that, so SharePoint permissions need to be right before the connector is switched on.
  • We supplied a starter AI use policy for management to adopt.

Results

In use In use on the firm's own Claude Team workspace.

  • Estimate

    Hours, not days for a first draft of a scored answer. Not yet timed against the old process.

  • Observed

    Years of material past bids, CVs and project references, searchable from one place.

  • Observed

    Every claim sourced Each statement in a draft is linked to the document it came from.

Measured: counted or timed. Estimate: worked out from the old routine, not timed. Target: what the client is aiming for. Observed: seen in use, not a number.

Limitations

  • Drafts are only as good as the library. In early trials, an incomplete set of documents made a capable firm look underqualified, and two sources gave different values for the same project. Keeping the library checked is ongoing work.
  • Connecting Claude to Microsoft 365 needed the IT provider's approval and a change to the firm's access policies, which took several weeks.
  • Claude drafts; it does not decide what wins marks. Bid strategy and final wording stay with the bid team.
  • The time saving is an estimate, not a measured result.

Which Claude product did what

Claude Team
The firm's own workspace, connected to its Microsoft 365. The bid team drafts here. Anthropic's business plan for Claude. Staff use Claude in a browser or the desktop app, share projects, and connect it to systems such as Microsoft 365. The workspace and its data belong to the business.
Claude Code
Not used in this work. Anthropic's coding and automation tool. It runs on a computer, can read and write files, write software and carry out multi-step jobs that a person has set up and approved.

How Claude Team, Claude Code and software built with Claude differ

Other case studies

Interested in tender drafting for your business?

Tell us how the work is done now and who does it. We'll suggest a time for a short call. See also ways to start.

Talk to us about tender draftinghello@acuityai.co