AI proposal writing at engineering firms, with the tools you have

AI proposal writing at an engineering firm means using AI to read the RFP, find past projects, adapt resumes and check the draft, while principals write the approach and approve the submission. Proposal teams at engineering firms answer RFPs and RFQs against tight deadlines, often reusing material from past submissions. This article looks at AI proposal writing at an engineering firm with the tools you already have, Copilot or Gemini on top of your past proposals, resumes and project sheets, and at where the NSPE Board of Ethical Review draws the line between marketing documents and sealed technical work.

Where proposal hours go

Most of the time on a proposal goes to five things: reading the RFP and listing what it asks for, finding similar past projects, updating resumes and project descriptions, writing the approach section, and checking the draft against the requirements before submission. Only the approach section is new thinking every time. The rest is finding, adapting and checking.

A useful first exercise is to take one recent proposal and write down, from memory and calendars, how many hours went into each of those five things and who did them. The split is rarely what the team expects. It also shows where the firm’s own material is hard to find, which is the first problem AI runs into: a model can only reuse past proposals and resumes that it can locate and that are current.

AI proposal writing at an engineering firm: five steps

AI can take over part of five proposal steps: the requirements matrix, the search for past projects, tailored resumes, first drafts of the approach and the final check against the RFP.

Reading the RFP and building a requirements matrix

AI can read the RFP and produce a table of every requirement: section, page limit, required forms, evaluation criteria and submission format. The proposal manager checks it against the document; a missed requirement is the most expensive error in a proposal.

Read the attached RFP. Build a table with one row per requirement: RFP section, requirement in the RFP’s own words, page or format limit, and where in our response it will be answered. Quote the RFP exactly and mark anything ambiguous.

Finding relevant past projects in SharePoint or Drive

If past proposals and project sheets sit in one well-organized library, AI can search it for projects that match the client type, size and services in the RFP and return a short list with links. A principal picks the projects.

Tailoring resumes and project descriptions

Starting from the master resume and project sheet, AI can draft a version that emphasizes what the RFP asks for, within the page limit. Each person confirms their own resume; nothing is added that is not in the master version.

First drafts of the approach section

AI can produce a first structure for the approach section from the RFP and from approaches used in past winning proposals. This is the step where AI helps least and the project lead’s judgment matters most, so treat the draft as a list of headings to rewrite.

Checking the draft against the requirements

Before submission, AI can compare the full draft with the requirements matrix and list anything missing, over length or answered in the wrong section. A person makes the corrections.

Compare our draft response with the requirements table. List every requirement that is not answered, answered only partly, or answered outside the section the RFP asks for. Do not edit the draft.

What to set up before using AI on proposals

Set up three things first: one organized library of past proposals, permissions that match the proposal team, and approved boilerplate. The first is a single library of past proposals, project sheets and master resumes, with clear names and current versions. The second is permissions: Microsoft says Copilot “only surfaces organizational data to which individual users have at least view permissions” (Microsoft Learn), so the proposal team needs access to the library and should not reach unrelated folders through it. The third is approved boilerplate for firm history, quality and safety sections.

On client data, Google states that “Your content is not human reviewed or otherwise used for Generative AI model training outside your domain without permission” (Workspace privacy hub), and Microsoft says prompts, responses and Graph data are not used to train its foundation models. Read the current terms for your plan before putting client RFP material into either.

Organizing the proposal library

A proposal library works for AI when past pursuits, project sheets, master resumes and approved boilerplate each sit in one place with clear names and current versions. Most of the value in the steps above depends on that library. A library that works well for people also works well for Copilot or Gemini, and it can be built in a few weeks by the marketing team.

• One folder per past pursuit, named with year, client and project, holding the final submitted version and the RFP.
• One project sheet per completed project, in a single template, with client type, size, services, team and outcome.
• One master resume per person, updated on a fixed date each year, with tailored versions saved next to the proposal that used them.
• One file of approved boilerplate for firm history, quality, safety and insurance, with the date it was last approved.
• An archive folder for drafts and superseded versions, outside the main library, so search results return final material.

The archive folder is the step most teams skip. Without it, a search for similar projects returns three drafts of the same proposal and the AI may quote from the wrong one.

Measuring the result

Pick two measures before you start: hours spent per proposal by the proposal team, and requirements missed or answered out of place at the final check. Track them for the next few submissions with AI and compare with the last few without it. A drop in missed requirements is often the first visible change, since the requirements matrix and the final check are the steps where AI does the most reading.

Debriefs and lessons learned

When a client offers a debrief after an award or a loss, the notes usually end up in someone’s inbox. Saving them in the pursuit folder, with a short summary of what the evaluators said about each section, gives the library one more useful layer. The next time the firm answers a similar RFP, the search can surface both the past proposal and what the client thought of it. AI can draft that summary from the debrief notes; the proposal manager checks it against what was actually said.

Proposals are not sealed documents

A proposal is a marketing document. Plans, specifications and reports that an engineer signs and seals are a different category, and the ethics guidance draws a clear line between them.

What NSPE’s Board of Ethical Review said in Case 24-2

In Case 24-2, Use of Artificial Intelligence in Engineering Practice (July 18, 2024), the NSPE Board of Ethical Review considered the use of AI in an engineer’s technical work. The board did not find the use of AI-assisted drafting tools unethical in itself. It said “AI-generated technical work requires at least the same level of scrutiny as human-created work.”

Responsible charge and review of AI output

The board concluded that “Engineer A did not maintain responsible charge in violation of licensure law which violates Code section III.8.a.” The case cites Code section II.2.b: “Engineers shall not affix their signatures to any plans or documents not prepared under their direction and control.” For sealed work, the engineer of record has to direct and review the work, whatever tool produced the first draft.

Disclosure to the client

On disclosure the board wrote: “There are currently no universal guidelines mandating AI disclosure in engineering work,” and “Ethical principles favor transparency when AI plays a substantial role in generating work products.” For proposals, our reading is that a firm should be ready to say how AI was used if a client asks, and should follow any disclosure requirement in the RFP itself.

What stays with the principal

The decisions that win or lose a pursuit, and that carry professional responsibility, stay with the principal and the project lead. The steps above save time on finding, adapting and checking; these do not change:

• the go or no-go decision on the pursuit
• the choice of team and of reference projects
• the technical approach and its commitments
• fees and schedule
• final review and signature of the submission

Preparing a proposal is one of the processes we list in Choosing the first process to review, and the review report marks which steps AI drafts and who checks them. As an example of work on recurring office activities, our work page describes a mapping of 41 activities at a manufacturing company.

Frequently asked questions

Sources

1. Microsoft Learn, Data, Privacy, and Security for Microsoft Copilot
2. Google Workspace Admin Help, Generative AI in Google Workspace Privacy Hub
3. NSPE Board of Ethical Review, Case 24-2, Use of Artificial Intelligence in Engineering Practice, July 18, 2024

This article is general information. It is not legal or tax advice. For your own situation, ask your counsel, your tax adviser or the professional body that governs your practice.

Back to blog