Article
From usage to margin: AI success criteria your CFO will believe
Tie construction AI results to margin with a four-rung value ladder and success criteria your CFO will accept before a pilot starts.
Nov 19, 2026 · By Trunk Tools team · 6 min read
Most construction AI is judged on usage ("our teams asked thousands of questions") or on promises. Neither tells a chief financial officer (CFO) what the tool did for margin, and margin decides whether a job was a success. The link is real, and it runs through steps nobody writes down. As an example, across Trunk Tools projects, the median submittal cycle fell from 54.8 to 14.0 days (platform data), and the hard part is turning a result like that into a number finance accepts.
This post covers why AI value feels hard to prove, the four-rung value ladder, the paths that carry AI results to margin, what saved hours are worth, who keeps the savings under each contract type, how to label each number, and how to write AI success criteria before a pilot starts.
The short answer
Measure every rung from usage to margin, agree the dollar assumptions with finance before the pilot, and label each number as measured, modeled or reported.
01 — The problem
Why does AI value feel hard to prove?
Ask a project team what AI did for them and you hear real things. Submittals move faster. Fewer pointless requests for information (RFIs) go out. Someone caught a change the architect didn't cloud.
Ask the CFO what those are worth and the conversation stalls. The steps between "cycle time dropped" and "margin went up" were never written down, so AI gets judged on weak evidence:
- Usage alone. Logins show interest. They don't show results.
- Hours saved times a labor rate. Finance will ask whether payroll went down, and if it didn't, where the hours went.
- Stories. "It caught a big one on the hospital job" is a good story and a weak business case.
The fix is to measure every step and connect them.
02 — Value ladder
What is the AI value ladder?
Four rungs. Each one is believable only when the rung below it is measured.
| Rung | The question | Who cares most | What to measure |
|---|---|---|---|
| 1. Usage | Are teams using it? | Technology and operations leaders | Share of submittals reviewed by AI, active users per project |
| 2. Workflow results | Is the work faster or better? | Project teams | Submittal cycle time, first-pass approval rate, RFIs avoided, drawing changes caught |
| 3. Project outcomes | Is the project better off? | Project executives, owners | Critical-path days protected, rework avoided, change orders captured |
| 4. Business outcomes | Is the company better off? | Chief executive, CFO, chief operating officer | Gross margin, capacity for more work, repeat business |
Most AI conversations sit on rung 1 or jump straight to rung 4. The value lives in the links between rungs 2, 3 and 4.
03 — Paths to margin
How do AI results turn into margin?
Faster submittals protect the schedule
A submittal stuck with the design team holds up procurement. A long-lead item ordered late becomes a schedule problem months later, and every extra day on site costs general conditions: supervision, trailers, equipment, insurance.
The chain: shorter submittal cycles, earlier release of long-lead items, fewer critical-path delays, less spent on extended general conditions.
To measure it, track cycle time on critical-path submittals, and agree on a daily general conditions rate with finance before the job starts. Each critical-path day protected is worth that rate.
Caught drawing changes avoid rework
A change that slips past the team shows up in the field as rework: work built to a superseded sheet, torn out and built again. The changes most likely to do this are the ones the architect never clouded, because nobody knew to look for them.
To measure it, count the changes caught before the affected work was installed. Value each one from your own rework history, and say plainly that the value is modeled.
Captured changes get billed
A missed change can also mean a change order nobody wrote. When a bulletin or an RFI answer adds scope and nobody prices it, the general contractor (GC) absorbs the cost, usually out of contingency.
To measure it, compare the change events identified with the change orders submitted, and track the value of scope that would otherwise have been absorbed. Finance recognizes this one fastest, because it shows up as revenue.
Fewer unnecessary RFIs cut cost and wait time
Every RFI takes time to write, track and wait on. Across Trunk Tools projects, 19.5% of potential RFIs were already answered in the project documents or earlier RFIs (platform data).
To measure it, count RFIs avoided and agree to a cost per RFI with finance. Much of the processing cost sits with the design team, so for a GC the bigger gain is time, and a design team free to answer the questions that drive cost.
04 — Hours saved
What are the hours saved worth?
Saved hours are the easiest AI benefit to count and the easiest to over-claim. A project engineer who saves an afternoon on submittal reviews does not turn into a payroll cut, and in an industry that can't hire fast enough, nobody wants one.
The margin value of saved hours is capacity: the same team runs more work, or the same work with less overtime. When revenue grows faster than overhead, margin improves.
To measure it, track contract value per project manager and project engineer over time. If AI is working, that ratio rises, or it holds steady while the team stops working late.
Also, we're in the people business, and if technology improves the morale of your people and helps get them home to their families, that's a win.
05 — Contract type
Who keeps the savings?
Contract type decides, and many AI business cases fall apart here.
- Lump sum and guaranteed maximum price (GMP): avoided rework, a protected schedule and recovered change orders flow largely to the GC's margin (on GMP, subject to the shared-savings split).
- Cost-plus: much of the rework and delay cost passes to the owner. The value shows up as a better owner experience, which leads to repeat and negotiated work, usually at healthier margins than hard-bid jobs.
A case that treats every avoided dollar as GC profit loses credibility the first time finance looks at the contract mix. Account for it up front.
06 — Label each number
How do you label a number a CFO will trust?
Label each number by how you know it:
- Measured: straight from the system. Cycle times, RFIs avoided, changes detected.
- Modeled: measured data plus assumptions agreed in advance. Daily general conditions rate, rework cost per missed change, loaded labor rate.
- Reported: from people. Surveys and field stories.
All three belong in a business case. Mixing them without saying so is how AI value gets called squishy.
07 — Success criteria
How do you write AI success criteria before a pilot?
For each rung, write down four things before the pilot starts: the metric, today's baseline, the target and the owner.
| Rung | Metric | Baseline | Type |
|---|---|---|---|
| Usage | Share of incoming submittals reviewed by AI before a person opens them | Today's share | Measured |
| Workflow | Median cycle time on critical-path submittals | Today's median | Measured |
| Workflow | First-pass approval rate from the design team | Today's rate | Measured |
| Project | Critical-path days protected, times the agreed daily general conditions rate | Starts at zero | Modeled |
| Business | Contribution to project margin from schedule, rework and recovered change orders | Plan margin | Modeled |
Set each target with the project team and finance together. Then:
- Get the baseline first. Without a before, every after is an opinion.
- Agree the assumptions up front. Settle the daily rate and the rework value with finance before the pilot, well ahead of the renewal meeting.
- Expect the rungs on different clocks. Usage and workflow results show in weeks, project outcomes in months, margin at closeout. Say so at the start.
- Watch usage as an early warning. If rung 1 is low, rungs 2 to 4 never show up.
- Report every rung, every month, on one page.
Published November 19, 2026. Written by Chris Boyd, Chief Product Officer at Trunk Tools.
08 — FAQ
Frequently asked questions
What is a good first metric for an AI pilot in construction?
One workflow result your team already tracks, such as submittal cycle time. Write the baseline down before the pilot and measure it on critical-path items.
Can hours saved count toward margin?
Yes, as capacity. Track contract value per project manager and project engineer over time.
When does margin show up?
At closeout. Usage and workflow results show in weeks, and project outcomes take months.
Next step
Bring one live project and its current submittal numbers to a 30-minute session, and we'll set your baselines together.