On a fixed-fee engineering job, the change that eats your margin can arrive as an ordinary email. By closeout, it's buried in a thread and the hours are already spent. An AI workflow can bring that request together with the agreed scope, flag what needs attention, and prepare the next action while the project manager can still do something about it.

Our 12-week engagement with a retail engineering firm helped add a reported $1.98M a year in retail margin: $1.2M in annual billed changes plus an estimated $784K annual benefit from less rework, calculated from the pilot reduction. Their existing design automation already saved time. We built change capture and drawing checks to stop later work consuming that gain.

Here's how to build the change-capture part: connect the right records, teach the system your firm's rules, and test whether it puts the right request in front of the right person.

Start with one project's records

Choose a live project with a clear scope and a project manager who'll own the decisions. Give the workflow access to:

  • The signed scope, exclusions, revision allowances, and approved amendments, with their versions and effective dates.
  • The project inbox, agreed meeting notes, and revision or permit-comment log where new requests arrive.
  • The current fee, hours used, forecast hours to finish, and delivery milestones.
  • The firm's change procedure: who can approve a change, what needs to be recorded, and who handles unresolved questions.

Keep those records tied to the same project ID. A request for one store mustn't be checked against another store's scope. Use the environment and access permissions your firm has approved for client work.

Start small enough that the project manager can inspect every result. Once the workflow handles one project well, you have a basis for expanding it.

Turn the firm's rules into clear routes

One simple way of doing this is to start with your current triage sheet. For each request, the AI needs to identify what changed, find the relevant scope passage, and suggest a route:

  • Included work: The request fits the agreed deliverable and allowance. Send it through ordinary delivery.
  • Potential client change: The request changes a deliverable, allowance, or milestone. Put it in the project manager's commercial-review queue.
  • Internal correction: The work addresses something the team missed in its own requirements. Route it to the quality or delivery owner.
  • Needs clarification: The scope is missing, the records disagree, or the cause isn't clear. Ask for the missing fact.

Write the rules with the project manager and engineering lead, using past requests and the decisions they should produce. Include routine requests as well as obvious changes. Define what happens when rules conflict: the system should show the conflicting passages and route the question to the named owner.

Keep the evidence beside the recommendation. “Potential client change” becomes useful when the manager can immediately open the request and the scope passage that triggered it.

Name a project manager, engineering lead, and integration owner before building. Check whether your current project software can hold the review queue and automate fixed rules. Use AI for interpreting requests and finding the relevant scope, and reserve the team's time for testing, review, and maintenance.

Where does work get stuck in your engineering firm? Our Free Expert Assessment helps you find the bottleneck in one workflow and gives you a written recommendation for how to solve it. We'll help you choose the workflow and give your team a simple kit to collect the evidence. You'll get a practical plan, an estimate of what your team would need to carry it out, and two consultations to guide you through it.

Get my free assessment

Connect the AI to the next action

Build the workflow around the records your team already uses. A project-system automation or a small integration can run these steps:

  1. Capture the request. When a new email, meeting note, or logged comment arrives, save its project ID, source link, date, and message ID. Use the message ID to avoid creating the same item again when a job retries.
  2. Retrieve the right context. Load the relevant scope and approved amendments for that project, plus the firm's routing rules. Include the current budget and milestone data as separate fields with an “as of” date.
  3. Ask AI for a structured review. Have it extract the requested change, propose a category, cite the supporting passages, identify missing information, and suggest the next action.
  4. Check the output. Require valid source references, an allowed category, and an owner from the project record. If the project match or scope version is unresolved, route it for clarification. Keep these checks in the workflow so an email can't change who has approval authority.
  5. Create the review item. Put the recommendation in the project manager's existing queue. For a potential client change, prepare a draft notice using the firm's template. The manager reviews the scope, effort, schedule, and agreement before approving what goes to the client.
  6. Record the decision. Write the approved action back to the project log. Keep the request, recommendation, reviewer, and outcome together so finance can follow it through pricing, billing, and collection.

A useful instruction for the AI is:

Compare this request with the supplied project scope, approved amendments, and firm rules. Return the project ID, requested change, category, supporting request and scope references, missing information, decision owner, and next action. If the records don't support a category, use “needs clarification” and state the missing fact. Leave effort estimates for the delivery owner unless a current approved estimate is supplied.

That gives the model a defined job and gives the manager an output they can check.

Work through a layout change

For example, a retail project's scope covers the agreed store layout and one revision round. That round is complete. The client then emails a request to move the service counter and revise the associated electrical and plumbing drawings.

Here's what the review item should contain:

  • Request: Move the service counter and update the affected drawings, with a link to the email.
  • Scope evidence: The agreed layout reference, the one-round allowance, and the record showing that round is complete.
  • Suggested route: Potential client change.
  • Missing information: The delivery lead's estimate of the extra work and its effect on the next issue date.
  • Owner and action: The project manager checks the change procedure, gets the estimate, and decides the commercial response.

Now change one fact. The drawing doesn't match the layout the client already approved. That should send the item to the internal-correction route. If the system can't find the approved layout, it should ask for that record.

These variations matter more than whether the AI can write a polished change notice. They show whether it's using the firm's rules to distinguish the work.

Test the decisions before connecting client communications

Build a reference set with project managers and senior engineers. Include real changes, included revisions, internal defects, unclear permit comments, and requests from the wrong project. Have the team record the correct route and supporting evidence for each one.

Use some examples to develop the instructions, then test on examples the system hasn't seen. Check:

  • Did it miss a genuine change that needed a commercial decision?
  • Did it flag included work or an internal defect as a client change?
  • Did every scope reference point to the right project and version?
  • Did it ask for clarification when the information was missing?
  • How much time did the manager spend checking and correcting the output?

Agree on acceptable performance with the people who own the work before the pilot. Run it alongside the team's normal process first, with outbound notices waiting for approval. Review misses and false flags, then update the rule or source that caused the error. Keep a fresh set of examples for the next check.

In our retail engagement, senior engineers supplied examples of correct decisions, and we tested the additions on unseen examples against the diagnosis baseline. Project managers kept the billing decision, and licensed engineers continued to review and seal the drawings.

Measure whether the workflow improves finished jobs

At the weekly project review, show open requests beside the current fee, hours used, forecast hours to finish, and next milestone. Start with requests that don't have an owner, decisions waiting on the client, and work underway without a recorded response.

Track request-to-decision time, missed changes, false flags, review effort, and the work that becomes priced and billed. Keep internal rework hours visible too. Compare the completed job's margin with its forecast so you can see whether earlier decisions changed the result.

If requests are now captured but managers still can't act, inspect what's holding up the decision. The next constraint could be missing estimates, slow approvals, or poor original pricing. That finding tells you where to improve the workflow next.

Want to see how a 12-week engagement helped an engineering firm's retail operation add a reported $1.98M in annual margin? The full case shows how we found the loss, built the solution, and handed it over.

Want a clear plan for a workflow that's slowing your engineering firm down? Our Free Expert Assessment helps you find its bottleneck and decide what to change. Choose a workflow with us in an intake consultation, then use our collection kit to gather the evidence. You'll get a free written report with the findings, our recommended solution, alternatives, a first action and a measure of success, plus the skills, effort and costs your team would need. We'll send it before your second consultation so we can discuss the findings and next steps. The report and recommendation are yours to keep and put to work.

Get my free assessment