Winning work
Using AI to Write Proposals: What It Does Well, and What Your Engineers Check
AI proposal writing means using an AI tool to read the RFP and assemble first drafts of the repeating parts from the firm’s own library, while the engineers write the approach and the fee and check every claim.
On this page
Terms Used
- Library: the firm’s past proposals, project sheets, résumés and standard text, kept current in one place.
- First draft: the text an AI tool assembles from the library for a new RFP, before anyone at the firm edits it.
- Data terms: what the AI provider may do with what you enter, including whether it trains its models on it.
Why Proposals Are a Good First Use
Proposal hours are almost always non-billable, they recur every week, and much of the content repeats from one proposal to the next. That makes proposals one of the clearest places for AI to give hours back to billable work. It also makes them a place where a confident mistake costs a pursuit.
What the Industry Reports
| Architecture and engineering firms, 2026 | Figure | Source |
|---|---|---|
| Firms using AI | 70%, up from 53% a year earlier | Deltek, 12 May 2026 |
| Reporting a measurable positive business impact from AI | 38% | Deltek, 12 May 2026 |
| Saying it is too early to see benefits, or still piloting | 39% | Deltek, 12 May 2026 |
| Naming where and how to apply it as a challenge | 62% | Deltek, 12 May 2026 |
In Deltek’s study of 896 architecture and engineering firms, seven in ten are using AI and fewer than four in ten can show what it did for them. Proposals are a good place to close that gap, because the hours before and after are easy to count.
What AI Does Well
- Assembling the parts that repeat. Firm descriptions, project sheets, résumés and standard text, chosen and ordered for a new RFP.
- Reading the RFP. Listing every criterion, weight, mandatory requirement, page limit and deadline, as a checklist.
- A first outline. Headings in the RFP’s own order and words, with the evidence each section needs.
- Tailoring. Rewriting a project description to bring forward the features this client cares about.
- Editing. Shortening, plain-language rewrites, consistency of names and terms.
What Stays with Your People
- The approach. How the firm will do this project, and why, comes from the engineers who understand it.
- The fee. Built from the firm’s own costs and history; see setting hourly rates.
- Every claim. Each project, figure, name, date and role in the draft, checked against your records.
- The engineering. Any technical statement about the design is the engineer’s, and the engineer checks it.
It Is Only as Good as Your Library
A general AI tool knows nothing about your projects. Given nothing, it writes plausible text about a firm that does not exist, which is worse than a blank page. Given your past proposals, current project sheets and résumés, it writes drafts in your firm’s words about your firm’s work. So the first step is the library:
| In the library | Kept current by | Used for |
|---|---|---|
| Project sheets: client, scope, role, outcome, reference | The project manager, at closeout | Experience sections |
| Résumés, one page each, by project type | Each person, once a year | Team sections |
| Past proposals, won and lost, with debrief notes | The proposal lead | Structure, approach ideas, what scored |
| Standard text: firm description, quality, safety, insurance | One named person | Mandatory sections |
Our article on the statement of qualifications covers keeping project sheets and résumés current.
A Drafting Workflow, Step by Step
- Give the tool the RFP and ask for the checklist: criteria, weights, mandatory items, limits, dates.
- Ask for an outline in the RFP’s order and words, with the evidence each section needs.
- Have the proposal lead choose the team and the projects, from the library.
- Ask the tool for a first draft of the repeating parts: firm description, project sheets, résumés tailored to this RFP.
- Have the proposal lead and the project lead write the understanding, the approach and the fee.
- Check every claim in the draft against the firm’s records, and every technical line with the engineer.
- Score the response against the grid before it goes, as our article on RFP responses describes.
Checking a Draft
AI drafts fail in predictable ways, so the check can be a list:
- Names: people, clients and projects that are real, current and spelled right.
- Figures: every number traced to a record: lengths, budgets, dates, durations.
- Roles: what the firm actually did on each project, not what a similar firm might have done.
- Promises: nothing committed in the text that the approach and the fee don’t cover.
- The client: the right client’s name throughout, with nothing left from the last proposal.
Set the Rules First
- Confidentiality. Check the RFP’s terms and the tool’s data terms. Use an account whose terms say your entries are not used to train the model.
- Client information. Decide what may be entered: a public RFP is different from a client’s confidential drawings.
- Review. Name who checks each draft before it leaves the firm, and what they check.
- Security. Deltek’s study found 93% of firms reporting a cyber attack attempt in the past three years, and data and cybersecurity named as a challenge with AI by 52% (Deltek, 12 May 2026). Use the firm’s own accounts with sign-in controls, not personal ones.
Measure It
Record the hours your proposals take for a month before you start, and for a month after. That record is what turns “we use AI” into a figure the partners can act on.
For example, if a firm writes four proposals a month at 70 hours each and the drafting brings that to 50, it frees 80 hours a month (4 × 20). At $62 an hour of salary cost, those hours cost $4,960 a month (80 × 62), and the firm can put them on billable work or more selective pursuits. Your own before-and-after figures replace these.
Common Mistakes
- Starting without a library. The tool invents a firm to write about.
- Letting it write the approach. The section that wins needs the engineers’ thinking.
- Skipping the claim check. One wrong project or figure can cost the pursuit and the client’s trust.
- Personal accounts. The firm loses control of what was entered and who can see it.
- No before-and-after hours. Without them, nobody knows whether it paid.
How We Approach This
Proposal writing usually isn’t covered by a client fee. Our software and AI work builds first drafts assembled from your past work and project data, assistants that work from your own documents, templates and standards, and your engineers check and sign.
Questions
Can AI Write Engineering Proposals?
It can draft the parts that repeat, read the RFP into a checklist and outline the response, from the firm’s own past work. The approach, the fee and the check of every claim stay with the firm’s people.
What Should a Firm Give an AI Tool for Proposals?
Its past proposals, current project sheets and résumés, and standard text, kept current in one place, under data terms that keep entries out of model training.
Is It Safe to Put an RFP into an AI Tool?
Check the RFP’s confidentiality terms and the tool’s data terms first, use the firm’s own accounts, and decide what client information may be entered.
How Much Time Can AI Save on Proposals?
Measure it. For example, a firm writing four proposals a month that cuts each from 70 to 50 hours frees 80 hours a month (4 × 20).
What Goes Wrong with AI-drafted Proposals?
Wrong names, invented figures, roles the firm didn’t have, and text left from another client. Checking a draft against a list of those is the fix.
Sources
Updates
- : Rewritten at full length: Deltek’s figures on AI use, the library table, a drafting workflow, checking a draft, security, measuring the hours and common mistakes; Deltek re-read.