Your team has three AI proposals. One speeds up proposals, another generates drawings, and a third catches client changes. Each promises to save time. Which deserves the people and budget?
Start with the work that limits the business result you want. In our 12-week engagement with a retail engineering firm, we 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. The loss came later, through unpriced changes and repeated work.
Here's a practical way to make that choice in your own firm: trace finished jobs, identify the step consuming the gain, compare ways to change it, and write a decision rule for a small trial.
Name the result and follow a completed job
For a fixed-fee engineering team, a useful starting question is: “Why are completed projects earning less margin than we planned?” That gives you a clearer investigation than “Where could we use AI?”
Ask a delivery leader and someone from finance to select recent jobs with enough records to reconstruct what happened. Include successful jobs and jobs that went over budget. Follow each from the agreed scope to final delivery and billing.
Put these records together:
- The original scope, fee, planned hours, and key delivery dates.
- Requests, revisions, review comments, and approved amendments.
- Hours spent before and after the first issue.
- Priced changes, billed changes, and the completed job's margin.
Keep the project IDs consistent. Finance's final figures and the delivery team's history need to describe the same work.
Find the step that consumes the gain
For each job, ask where the plan changed. Did the client add work? Did drawings return for correction? Did a complete package wait for a licensed reviewer? Identify the event, the owner, and the effect on the job.
Separate doing time from waiting time. A drawing package can spend an hour in review and three days waiting for that review. Those point to different interventions.
Then ask what would happen if the proposed AI worked perfectly. Faster drawing generation won't help a reviewer clear incomplete packages. More proposals won't solve the loss on projects already won. An early change flag helps only when a project manager can act on it.
Choose a candidate whose output can change the limiting decision. Give the team a source record they can inspect, rather than asking them to accept a label such as “inefficiency.”
If you’re an engineering firm leader and want to grow your firm without working harder on projects that earn less than they should, we’ll identify the biggest bottleneck standing in your way, build the plan to remove it, and implement it in eight weeks* or less - guaranteed or we continue to work for free and you don’t pay.
Find my bottleneck* Eight-week delivery, or twelve weeks when substantial data foundations are required, as agreed before work begins. If a covered stage fails its agreed benchmarks, we provide up to four additional weeks of remedy work at no charge, then waive or refund that stage’s fee if it still fails. Successfully delivered stages remain payable. Agreed client-caused delays pause the clock. Delivery guarantee terms.
Compare projects against the same evidence
For example, suppose finished-job reviews show that initial drawings arrive on time, but client changes reach the project manager after the team has already done the work. Some packages also need correction before senior review.
Use a comparison like this:
| Candidate | What it changes | Evidence needed before funding |
|---|---|---|
| AI proposal drafting | Time to prepare bids | Evidence that proposal capacity limits suitable work won and delivered. |
| More drawing automation | Time to produce the first issue | Evidence that first-issue preparation is still holding up profitable delivery. |
| AI change capture | How early a manager sees a scope decision | Missed or late requests, their source records, and a manager who can act sooner. |
| A package-readiness check | What reaches senior review | Repeated defects, their causes, and the effort they consume. |
In this example, change capture deserves a closer test. Readiness may deserve a separate one. The evidence hasn't established that either requires a new platform.
Check the simplest workable route for the chosen problem. An existing project system may already handle routing and reminders. AI may help interpret requests and retrieve scope passages. Our build-or-buy guide shows how to compare the delivery options.
Turn the candidate into a testable workflow
Define the proposed change from input to action. For scope capture, that could be:
- A client request arrives in the project inbox.
- The workflow matches it to the project and current scope.
- AI identifies the requested change and cites the relevant passages.
- Fixed rules check the project, source version, and decision owner.
- The project manager reviews the item, gets an effort estimate, and decides the response.
- Finance follows the approved work through pricing and billing.
The project manager owns the decision. An engineering lead supplies examples and checks quality. A technical owner handles integration and maintenance. If those people can't make time for a trial, include that capacity gap in the funding decision.
The full engineering implementation guide gives the routing rules, prompt, worked example, and evaluation steps.
Decide what would justify expansion
Build a test set from past requests, including included work, client changes, internal corrections, and unclear cases. Agree the expected decision with the people responsible for the work. Keep some examples aside until the workflow is ready to evaluate.
Run the trial alongside the normal process. Measure missed changes, false flags, review time, and how early managers can decide. Follow flagged items far enough to see which become approved and billed work.
Count the work the system creates too. If a manager spends hours dismissing weak flags, that effort belongs in the result. If the system finds changes but estimates still arrive late, investigate that next step before expanding capture.
Write the expansion rule before the trial: which errors must be absent, how much review effort is acceptable, and what improvement would justify the ongoing cost. Set those thresholds from the firm's economics and responsibilities.
Make the funding decision easy to inspect
Your final brief should fit on a page: the business result, the evidence locating the problem, the proposed workflow, alternatives, required people and access, test criteria, and owner after handover. Attach the source records and cost assumptions.
That gives leadership something concrete to approve. It also gives the team a reason to stop or change direction if the trial doesn't support the proposal.
Want to see how this approach helped a retail engineering operation add a reported $1.98M in annual margin through a 12-week engagement? The case study shows how finding the loss changed what we built.
If you’re an engineering firm leader and want to grow your firm without working harder on projects that earn less than they should, we’ll identify the biggest bottleneck standing in your way, build the plan to remove it, and implement it in eight weeks* or less - guaranteed or we continue to work for free and you don’t pay.
Find my bottleneck* Eight-week delivery, or twelve weeks when substantial data foundations are required, as agreed before work begins. If a covered stage fails its agreed benchmarks, we provide up to four additional weeks of remedy work at no charge, then waive or refund that stage’s fee if it still fails. Successfully delivered stages remain payable. Agreed client-caused delays pause the clock. Delivery guarantee terms.