Technology and AI

AI at an Engineering Firm: Where the Hours Come Back

Sam&Sain Co. · Published · Updated · 8 minute read

AI at an engineering firm pays back where it takes hours out of the repeated work around the engineering: proposals, reports, minutes, correspondence and searching past projects. Seven in ten architecture and engineering firms now use it, and fewer than four in ten can point to a measurable gain (Deltek, 2026).

On this page

Terms Used

  • Generative AI: tools that write text, summarize documents or produce drafts from a prompt, such as ChatGPT.
  • An assistant on your own documents: an AI tool given the firm’s past proposals, reports, templates and standards to work from.
  • Integration: connecting an AI tool to the systems the firm already runs, so its work goes straight into them.
  • Pilot: one task, measured before and after, before anything is rolled out.

Where Firms Are

Deltek’s 2026 study of 896 architecture and engineering firms in the US and Canada found AI use rising fast and results lagging behind it (Deltek, 12 May 2026):

  • 70% of firms use AI, up from 53% a year earlier; 78% use generative AI, up from 64% (Deltek).
  • 38% report a measurable positive business impact from it; 39% say it is too early to tell or are still piloting (Deltek).
  • 59% agree that successful AI will create a significant competitive advantage (Deltek).
  • Their top challenges: where and how to apply it (62%), integrating AI systems (56%), and data and cybersecurity (52%) (Deltek).

Hours Saved so Far

A second measure points the same way. The Real-Time Population Survey, a US survey published in the Federal Reserve Bank of St. Louis’s FRED database, estimates the share of work hours saved with generative AI in architecture and engineering occupations, counting non-users as zero.

QuarterShare of work hours saved
Third quarter, 20251.45%
Fourth quarter, 20252.53%
First quarter, 20263.27%
Second quarter, 20263.16%
Third quarter, 20261.99%

Source: FRED, series RPSGENAITSOCC4, updated 20 October 2026. Across the occupation, the hours given back so far are small, and they move from quarter to quarter. A firm that wants more than the average has to choose where AI is used rather than leave it to chance.

Why Use Runs Ahead of Results

The three challenges Deltek’s firms named line up with the three things a firm decides before AI gives anything back. Where and how to apply it is a choice of task: a licence for everyone and an invitation to try it does not make that choice. Integration is a question of systems: an assistant that cannot read the firm’s records, or whose output is retyped elsewhere, returns less than it seems. Data and cybersecurity is a question of rules: which accounts, what may be entered, and who checks. The sections below take each in turn. Starting from a task also gives the firm its measure, because a task has hours that can be counted before and after.

Where to Look for Hours

The places worth trying first are in the paperwork around the engineering, much of which is not billed:

WorkWhat AI can doWho checks
ProposalsChecklists from the RFP, outlines, first drafts of repeating partsThe proposal lead
ReportsAssemble standard sections and data into a first draftThe engineer responsible
MeetingsMinutes and action lists from notes or recordingsThe project manager
CorrespondenceDraft letters, responses and summaries of long documentsThe sender
Searching past workFind similar projects, details and past answers in the firm’s filesWhoever uses the result
Office tasksSort and file documents, prepare invoices and reports from dataAdministration

Design, calculations and the engineering judgment behind a sealed document stay with the engineer. AI changes how much time goes into the work around them. Proposals have their own article: AI for proposal writing.

A Practical Order

  1. Map the hours. Use a month of timesheets to find where non-billable time goes. See how to improve utilization.
  2. Pick one task. It should repeat every week, follow similar steps, and take billable people’s time. Proposals are a common first choice.
  3. Give it your material. AI works from what it is given. Past proposals, report templates and project records make the difference between a useful draft and a plausible one.
  4. Set the rules. Which tools, which accounts, what may never be entered, who checks. See ChatGPT at an engineering firm.
  5. Measure. Record the task’s hours for a month, run the pilot for a month, and record them again.
  6. Decide on the numbers. Keep it, change it or drop it.
  7. Put the freed hours to work. Move them onto the resource plan, or they leak away.

What a Pilot Looks Like

A pilot is one task, one month before and one month after, with the hours written down. The measure is simple: hours freed a month = times a month × (hours each time before − hours each time after). Eight reports a month that drop from 6 hours of assembly to 3.5 free 20 hours (8 × 2.5).

Example, round numbersTimes a monthHours each, beforeHours each, afterHours freed a month
Report assembly863.520
Meeting minutes3010.2522.5

In the first example, 20 hours a month go back to the engineers who assembled the reports. Whether that is worth it depends on what those hours do next: billed on projects at $150 an hour, they would earn $3,000 a month (20 × 150); spent waiting for work, they earn nothing. The pilot answers the first question; the resource plan answers the second. In the second example the hours are spread across many people in small pieces, which makes them harder to turn into billable work, even though the total is larger.

Set the Rules Before the Tools

  • Accounts. The firm’s own business accounts, with sign-in controls, not personal ones.
  • Data terms. Tools whose terms keep the firm’s entries out of model training.
  • What may never be entered. Client data a contract keeps confidential, personal information, anything the client has not cleared.
  • Client terms. Some contracts and RFPs say how AI may be used on their work; read them.
  • Review. Every AI output that leaves the firm is checked by a named person, and anything technical by the engineer responsible.

In Ontario, Professional Engineers Ontario has adopted Engineers and Geoscientists BC’s practice advisory on AI “for use as a guideline in Ontario”. It says engineers “remain responsible for their work that incorporates or is based on AI output”, and that firms “should consider implementing a robust set of internal policies and procedures” (PEO, adopted advisory, version of 22 November 2024). Our article on ChatGPT at an engineering firm goes through what it asks.

Integration, the second challenge in Deltek’s list, matters as much as the first. An AI tool that cannot read your project records, or whose output has to be retyped into another system, gives back less than it seems. The larger gains come when it works inside the systems the firm already runs: the timesheet and project system, the accounting package, the document store. A draft invoice that appears in the accounting package for approval saves the retyping; a draft that appears in a chat window, to be copied across, saves only the writing. Our article on what to automate first lists where those connections pay, and many of them need no AI at all.

Small Firms

ACEC’s 2026 benchmarking study put the overall median at 26 out of 100 on technology readiness. Firms of 50 people or fewer had a median of 21, firms of 51 to 200 scored 27, and firms of more than 200 scored 43 (ACEC, May 2026). A small firm does not need the large firm’s programme. It needs one or two tasks done well, on its own data, with the hours counted. See digital transformation for a small firm.

Common AI Mistakes

  • Starting with the tool. A licence without a task produces trials, not hours.
  • No material. A tool that knows nothing about the firm writes about a firm that doesn’t exist.
  • No measure. Without hours before and after, nobody can say whether it worked.
  • Personal accounts. The firm loses sight of what was entered, and where.
  • Letting it near the engineering unchecked. A plausible technical sentence is still the engineer’s to check.

How We Approach This

We don’t start with the tool. We start with where the hours go, then build what removes them, linked to the systems you already run: assistants that work from your own documents, templates and standards. Your engineers check and sign.

Questions

How Are Engineering Firms Using AI?

Deltek’s 2026 study found 78% of A&E firms using generative AI. The uses worth trying first are the writing and searching around the engineering: proposal and report drafts, meeting minutes, summaries of long documents, and finding information in past projects.

Does AI Save Engineering Firms Money?

It can, where the hours it frees go to billable work. Only 38% of firms in Deltek’s 2026 study reported a measurable positive impact, so count one task’s hours before and after rather than assume a gain.

Which AI Tools Should an Engineering Firm Use?

The task matters more than the brand. General assistants suit language work; an assistant given the firm’s own documents suits proposals and reports; features inside the systems the firm already runs suit invoicing and filing. Choose on data terms and on whether the output lands where it is used.

Is It Safe to Put Client Documents into an AI Tool?

Not by default. Check the client’s contract, the tool’s data terms (including whether entries train the model), who can access what is uploaded, and how long it is kept. The practice advisory PEO adopted warns that uploaded data may be used to further train a model.

Will AI Replace Engineers?

The design, the checking and the seal remain the engineer’s. The practice advisory PEO has adopted says engineers remain responsible for work that incorporates or is based on AI output. AI changes how much time goes into the paperwork around that work.

Sources

Updates

  • : Rewritten at full length: the hours FRED’s survey shows saved, why use runs ahead of results, where to look for hours, a pilot with two worked examples, the rules to set first with the practice advisory PEO adopted, integration, ACEC’s readiness scores by size and common mistakes; Deltek, FRED and ACEC re-read.

Discuss Your Project

Topic

We read every application. If we can help, the first 30-minute video call is free. Or contact us.