How we helped an engineering firm raise operating margin from 15.9% to 23.8%.

The fastest-growing work was carrying the weakest margin.

Their firm designed building systems under fixed fees, so an owner changing equipment, layouts, or design requirements could create more work without increasing the agreed fee. Unless a manager recognized the change and reached a pricing decision, the firm absorbed the extra hours.

Leadership initially explained the difference between projects as project-manager discipline because some managers priced changes and others didn't. But the requests, drawing comments, and accounting records sat in separate systems, while managers saw budget updates two to four weeks after the hours had posted.

That meant a manager could discover the financial effect only after the team had done the work. The firm had already invested in AI and automation, but those tools hadn't connected an arriving request to the commercial decision needed to protect the project's margin.

We followed each change to find where the pricing decision was lost.

To see what managers were missing, we studied one live program and worked with finance to trace 214 owner-directed changes across twelve closed fixed-fee jobs. We followed each request from its first appearance through recognition, decision, and billing, so we could distinguish work the firm had delivered from work it had agreed to charge for.

Only 43% reached a documented price decision within ten business days. Across the closed jobs, the trace identified ~$2.1M of additional scope, but only ~$0.9M had been billed. The trace showed where requests had failed to become recorded commercial decisions.

The trace also linked earlier recognition with more decisions, which changed what we recommended. Managers needed the request and its commercial context while they could still act on it, so we focused the build on bringing those records together and preparing the next decision.

We connected the request to a decision before more work was absorbed.

The firm already had the records needed to recognize a change, but they weren't connected. We linked project accounting, correspondence, meeting notes, and drawing comments for the first program so a candidate change could carry the evidence a manager needed to review it.

We then built AI-assisted change capture and notice drafting around those records. Each flag included the source passages and a drafted additional-services notice, so the manager could inspect the request and choose to price it, absorb it with a recorded reason, or escalate it to a principal.

The manager decided what to send to the client. Agreed changes went into the project accounting record, and the firm made that record a requirement before an amendment could be billed. The change could now move from correspondence through a commercial decision into finance's billing process.

Managers also needed to see the cost of work already underway, so we started weekly budget reviews using their existing accounting data in the second week. The connected view later cut the reporting lag to three days, giving managers a chance to respond before the overrun became a closeout result.

Repeated drawing comments created another demand on review time, so we added a check against the firm's quality checklist and past comments before pilot drawings reached a licensed engineer. The check brought known issues forward for the team to address, while the engineer kept responsibility for review and sealing.

Their own examples set the standard before we expanded.

An earlier flag would only help if managers could trust it enough to review it. We therefore worked with two project managers and a principal to define owner changes, clarifications, and the firm's own coordination errors, then turn agreed examples into a reference set.

That standard gave us a way to test both missed changes and unnecessary flags before rollout. The detector had to find at least 85% of real changes while keeping false flags to no more than 20% of its flags. It reached 88% recall with 17% false flags on held-out examples, clearing the agreed bar for expansion.

We assumed every manager worked in the same channel.

But passing the test didn't mean every manager was seeing the flags. We carried the pilot team's shared messaging channel into rollout without checking each adopter's working habits, and one manager had muted that channel during a deadline. He reviewed only two of fourteen flags.

We traced the low review count to the muted channel. Their team moved the flags to his inbox with our engineer observing, and he reviewed eleven the next week. That result changed our rollout process: we now asked each adopter where the decision needed to arrive.

There was also a reason managers might postpone a notice even after seeing it: they were measured on utilization, while time spent preparing change notices counted as overhead. We proposed counting that work as project time and adding recovered fees to the scorecard. Leadership adopted both changes, so the firm's measures now recognized the work needed to recover the fee.

We equipped their team to add the next workflow themselves.

The firm needed to keep adding programs and handling new sources of change after we left, so handing over a working build wasn't enough. We gave its team operating instructions, a way to add channels, and tests that had to pass before an update reached production.

We then taught the innovation team to run, change, and maintain the build. Before handover, the team independently added a channel for requests for information and passed those tests. That demonstrated it could extend what we'd delivered while checking the changes against the same reference examples.

Decisions replaced missed fees, and the later margin result exceeded the goal.

With the requests reaching managers and decisions entering the project record, the first notices led to accepted prices, negotiated agreements, and recorded waivers. Finance could follow each priced change back to its source, which gave the CFO a way to identify the additional fees produced by the process.

In the first quarter after rollout, additional services billed across three programs rose from a prior run rate of ~$225K to $410K. The CFO attributed the $185K increase to notices produced by the new process.

Six weeks after handover, 86% of owner-directed changes across those programs reached a documented decision within ten business days, up from 43%. A waiver counted because the manager had made and recorded a commercial choice. For those changes, the firm had a recorded decision instead of simply absorbing the work.

Those operating results showed that managers were making more timely decisions and that additional work was reaching billing. The longer-term test was whether the firm could lift operating margin from its 15.9% starting point.

In the recent follow-up, the CFO reported 23.8% operating margin, well above the 20.5% target for 2026, and credited a large part of that improvement to our work. The firm had beaten its financial goal and gained a process its own team could run, maintain, and extend.

See how we measured decisions and checked the implementation.

The technical companion explains the change trace, evaluation, human decisions, financial measures, and handover.

The firm could see and decide on work it used to absorb.

The firm needed higher operating margin, but its managers were learning about changes after the extra work had already been absorbed. We connected the records, prepared the notices, and built a path from a manager's decision into billing, so the firm could charge for work it had previously missed.

Earlier budget reviews helped managers act on work still underway, and the team we trained could keep extending the system. The result was a process the firm could run itself, with more timely decisions, additional billed fees, and a later operating-margin result above its original goal.