Most AI dashboards report the wrong thing. They show conversations handled, messages sent, calls answered, hours saved. Those numbers go up almost automatically once you turn a tool on. None of them tells you whether your schedule got fuller.
For a contractor, the return on AI comes down to two numbers: booked jobs and the revenue attached to them. This guide shows how to measure both, how to set a baseline so the comparison is fair, and how to do the payback math without flattering yourself. It is the measurement companion to AI for contractors: where to start.
Why "conversations" is the wrong metric
A conversation is an activity. A booked job is an outcome. Vendors report activities because activities are easy to count and always look big.
Consider a texting tool that "handled 1,200 conversations" last month. Of those, how many were:
- Existing customers asking when the tech would arrive?
- Spam and wrong numbers?
- Automated replies to your own outbound campaign?
- New customers who ended up on the schedule?
Only the last group pays you. If the vendor cannot show you that count, the dashboard is a vanity page. The same logic applies to "time saved." Saved time is real only if someone used it for something that books work. A CSR who gets 40 minutes back and spends it on the same low-value tasks has saved you nothing.
For context, even industry surveys are mostly about sentiment, not return. ServiceTitan's 2026 State of the Trades report, a survey of 1,000 residential contractors, found 48% of AI adopters reporting increased productivity. That is self-reported and does not say what the gain was worth. Your business needs a number in dollars.
Step 1: Take a 30-day baseline before you change anything
You cannot show improvement against a past you never recorded. Before launching any AI project, spend 30 days measuring the current state. Thirty days is long enough to smooth out a bad week and short enough that you do not stall the project.
Record these for the leads your project will touch (phone, web forms, ad leads, old estimates):
| Metric | Where it comes from |
|---|---|
| New-job leads by source | Call tracking, web forms, field service software |
| Answered vs missed calls | Phone system report |
| Time to first contact | Timestamps in your CRM or field service software |
| Appointments or estimates booked | Field service software |
| Booked jobs | Field service software |
| Revenue from those booked jobs | Invoices, not quotes |
If your CSRs do not tag lead source consistently, fix that first. A booked job with no source cannot be credited to anything. In a multi-location business, keep every number split by branch. One branch often drags the average, and you want to see whether the project helps there.
Also write down the baseline conditions: season, ad spend, headcount in the call center, any promotions. You will need them in step 4.
Step 2: Choose four numbers and ignore the rest
Four numbers are enough to judge most projects aimed at leads and follow-up.
- Booked jobs from the targeted lead sources. The headline number.
- Revenue from those booked jobs. Use invoiced revenue, since quoted revenue is a forecast.
- Lead-to-booked rate. Booked jobs divided by new-job leads. This shows whether you are converting better or just receiving more leads.
- Time to first contact. An input metric. If it improves and booked jobs do not, you have a different leak than you thought.
Everything else, such as messages sent or minutes of call time, is a diagnostic you check only when one of the four moves unexpectedly.
If the project targets old customers or unsold estimates, swap in the equivalent: jobs closed from estimates older than 30 days, or jobs from customers with no visit in 24 months. See following up on unsold estimates for how to define that pool.
Step 3: Count only incremental jobs
The question is not "how many jobs did the AI touch?" It is "how many jobs would not have happened without it?"
Some of those jobs would have booked anyway. A caller who reached voicemail may have called back. A customer who got no follow-up text may have booked on their own. Crediting every touched job to the tool overstates the return and sets you up for a bad decision when the number collapses later.
Practical ways to get closer to incremental:
- Before and after, same branch. Compare the 30-day baseline to the 30 days after launch, adjusted for the conditions you recorded.
- Branch comparison. If you run several locations, launch at some and hold others back for the first month. The held-back branches are your control group.
- Year over year. Compare to the same month last year if your business is seasonal, which most are. A no-heat surge in January makes any December comparison meaningless.
- Hold-out slice. Some systems can leave a small share of leads on the old process. Even 10% gives you a comparison.
None of these is perfect. Each is better than crediting the tool with every job it touched.
Step 4: Do the payback math
Payback is how many months of added profit it takes to cover what you spent. Use profit, not revenue. A $450 service call does not put $450 in your pocket.
Monthly net gain = (added booked jobs per month x average ticket x gross margin) minus monthly running cost
Payback in months = one-time setup cost divided by monthly net gain
Worked example (illustrative numbers)
These numbers are made up to show the arithmetic. They are not benchmarks or results from any company.
A plumbing company records a 30-day baseline: 210 booked jobs from phone and web leads, averaging a $450 ticket, so $94,500 in revenue. It launches a missed-call recovery and follow-up project. Thirty days later, in a comparable month, it records 228 booked jobs from the same sources.
- Added booked jobs: 18
- Added revenue: 18 x $450 = $8,100
- At a 40% gross margin: $3,240 in added gross profit per month
- Assume monthly running cost of $1,500 and a one-time setup cost of $6,000 (also illustrative)
- Monthly net gain: $3,240 minus $1,500 = $1,740
- Payback: $6,000 divided by $1,740, about 3.4 months
Now stress it. Suppose half of those 18 jobs would have happened anyway. Added jobs fall to 9, added gross profit to $1,620, and monthly net gain to $120. Payback stretches past 4 years. That is why step 3 matters: a result that depends on every touched job being incremental is fragile.
Run your own version with both the optimistic and the skeptical case. If the project only works in the optimistic case, say so before you scale it.
Step 5: Read it weekly, decide at 90 days
Weekly numbers are noisy. Read them weekly anyway, because a sudden drop usually means something broke, such as a number that stopped forwarding or a sync that failed.
Make the decision on a longer window. A reasonable rule: look at the first 30 days to confirm the system works, and judge return over 90 days against the baseline. Set the success threshold before you launch, written down, so you are not choosing the goalposts after the fact. For example: "If added gross profit does not exceed running cost by month three, we change or stop."
Common ways ROI gets overstated
- Counting leads you would have gotten anyway. Ad spend went up that month and so did leads.
- Using quoted revenue. Quotes are not sales.
- Ignoring running costs. Platform fees, voice minutes, texting fees and someone's weekly time to review exceptions all count.
- Skipping gross margin. A tool that adds revenue at low-margin jobs may add little profit.
- Measuring too short. One good week looks like a trend.
- Mixing months. July air conditioning calls and February furnace calls are different businesses.
What a good report looks like
If you hire anyone to build or run this, ask for a weekly report that shows the baseline next to the current week, split by source and branch, with booked jobs and revenue at the top. That is the structure we use in the weekly booked-job report that is part of the Booked-Jobs Recovery System, our fixed-scope engagement that goes live in 30 days and is measured in booked jobs. If you want a second opinion on how to set up measurement before you buy anything, we cover that in AI strategy consulting.
For the first number to chase, start by sizing the leak. What a missed call really costs a contractor walks through that formula, and speed to lead in home services covers the response-time side.
Frequently asked questions
What should I measure to know if AI is working in my contracting business?
Booked jobs and the revenue from them, compared with a 30-day baseline taken before launch. Add lead-to-booked rate and time to first contact to see why the numbers moved.
How long should I run a baseline?
Thirty days is a reasonable minimum for most contractors. If your business is strongly seasonal, also compare to the same period last year.
How do I calculate payback on an AI project?
Divide the one-time setup cost by monthly net gain, where net gain is added gross profit from incremental booked jobs minus monthly running cost. Use gross profit, not revenue, and test a skeptical case where only part of the added jobs are truly incremental.
Is time saved a valid ROI measure?
Only if the saved time is redeployed into work that books jobs or cuts a real cost, such as overtime. Otherwise it is a nice-to-have, not a return.