Sixty-six percent of contractors using AI say it saves them at least three hours a week. Forty percent say five hours or more. Those numbers come from ServiceTitan’s 2026 State of AI in the Trades report, released September 29 and built on a survey of 1,017 US contractors across HVAC, plumbing, electrical, roofing and four other trades. Here is the number that should bother you: only 47% of those same AI users report better profitability, and only 43% report more revenue. Hours saved and dollars earned are two different line items. Most shops count the first and assume the second.
The gap is not a software problem. Three saved hours a week is worth exactly zero if they dissolve into a longer lunch, a slower afternoon or one more round of checking email. Time only turns into margin when someone decides, in writing, where it goes. Heading into heating season, with no-heat calls about to stack up, this is the month to make that decision. The rest of this post shows how to run a 30-day scorecard on one AI workflow so you know by Halloween whether it pays or whether you cancel it.
Why Saved Hours Leak Before They Hit the P&L
The ServiceTitan data shows contractors stopped worrying about price. High purchase cost fell 12 points to 17% and now ranks last among adoption barriers. The new barriers are softer and harder to fix. Forty-four percent say they do not understand how to use AI in their business. Forty-three percent cite a lack of clear use cases. Forty percent struggle to connect it to the software they already run. Trust is worse still: only 11% of contractors say they have high or complete trust in their AI tools, while 58% report little or no trust.
Put those together and you get a familiar pattern. An owner buys an AI receptionist or turns on call summaries. The CSR handles fewer calls at night or spends less time typing notes. Nobody tracks what happens to those freed-up hours. Nobody checks whether the AI booked the job or just logged a message. Because trust is low, someone listens to every recording anyway, which eats the savings. Six months later the tool still bills monthly and the owner cannot say whether it made a dime.
The fix is boring and it works. Pick one workflow. Measure three numbers before you start. Name the job the saved hours get reassigned to. Check the numbers every Friday for four weeks. Then keep it, change it or kill it. The whole system fits on one sheet of paper taped next to the dispatch board, and it takes about 20 minutes a week to maintain. That is a small price for knowing, with numbers instead of a feeling, whether a monthly subscription earns its keep.
The 30-Day AI Scorecard, Step by Step
Start with the workflow most shops already pay for or are about to: after-hours and overflow call answering. ServiceTitan’s survey found 34% of contractors experiment with AI specifically to cover off-hours, second only to hiring pressure at 37%. It is also the easiest workflow to score, because every call has a clear outcome. Either a job landed on the board or it did not.
- Pull your baseline this week. From your phone system or field service CRM, grab the last 30 days of three numbers: inbound calls after 5pm and on weekends, how many of those reached a person or got a callback within 15 minutes, and how many turned into a booked job. Write them on one sheet of paper. That sheet is your scorecard.
- Turn on one tool, not three. If you run Jobber, the built-in AI Receptionist books straight into your schedule. Housecall Pro offers its CSR AI on higher tiers. ServiceTitan shops can pair Titan Intelligence call summaries with their existing answering setup. Shops that want a standalone option now have Wakeman, which launched September 28 with plans from $149 a month for after-hours only, $349 for full front desk coverage and $829 for a dispatch tier, plus a 14-day free trial and a direct Jobber booking link.
- Write the reassignment rule before day one. Decide exactly where your CSR’s saved hours go. Good options: calling back every open estimate older than three days, renewing lapsing maintenance agreements, or booking fall tune-ups from last year’s customer list. Put the rule on the scorecard. “Saved time goes to estimate callbacks, 4pm to 5pm daily.”
- Audit five calls every Friday. Not all of them. Five. Check whether the AI captured the address, the problem and the urgency correctly, and whether the booked job matched what the caller asked for. Five calls is enough to catch a bad script without rebuilding the trust problem by listening to everything.
- Score it on day 30. Compare after-hours booked jobs against your baseline. Then count what the reassigned hours produced: estimates closed, agreements renewed, tune-ups booked. Multiply each by your average ticket.
Here is how the math looks for a hypothetical six-truck HVAC shop. Say it books 20 after-hours jobs a month before the tool and 32 after, at a $450 average service ticket. That is $5,400 in new monthly revenue against a $349 plan. Now say the CSR spends the five reclaimed hours a week calling old estimates and closes two extra $9,000 replacements a month. The second number dwarfs the first, and it only exists because someone wrote the reassignment rule down.
Score the Back Office Too, Starting With Truck Spend
The call desk is not the only place hours leak. On September 30, ServiceTitan announced a partnership with Ramp that embeds Ramp’s AI bill pay and expense cards directly in the platform. Card transactions sync into ServiceTitan, where a supply-house purchase can be matched to the job that used it, with the receipt captured digitally. It is Ramp’s first embedded deal inside a vertical software platform, and it targets a real mess: techs buying parts on a shared card, receipts crumpled in a door pocket, and an office manager reconciling it all on the 28th.
The same scorecard applies. Before you switch anything on, measure two numbers for one month: hours your office spends matching receipts to jobs, and the share of material spend you can tie to a specific job number. Most owners guess the second number is above 90%. Many find it closer to 70% once they look. Every untagged dollar is margin you cannot see, which makes your job costing and your price book quietly wrong.
If you are not on ServiceTitan, you do not need this exact integration to run the test. Jobber, Housecall Pro and most mid-market platforms can export jobs to QuickBooks, and standalone expense tools like Ramp, Brex and Expensify all offer receipt capture by text or photo. The tool matters less than the rule. “Every card swipe gets a job number within 24 hours, or the card gets locked” will do more for your margins than any software feature.
What to Keep, What to Kill
On day 30, sort every AI tool you pay for into one of three buckets. Keep it if it moved a dollar number you can name: more booked jobs, more closed estimates, fewer hours on reconciliation that went somewhere productive. Fix it if it saved time but the time went nowhere. That usually means the tool works and your reassignment rule failed. Kill it if your Friday audits keep catching wrong addresses, wrong urgency or missed bookings, because a tool you have to babysit costs more than the subscription.
Watch the integration question as you sort. Sixty-one percent of AI users in the ServiceTitan survey say they want AI built into their core platform rather than bolted on. That preference is practical, not ideological. Every standalone tool that cannot write to your schedule or your job record creates a copy-paste step, and copy-paste steps are where saved hours go to die. A standalone tool can still win, but it should earn its spot with a booking or sync connection, not a text message summary someone has to retype.
Be honest about trust, too. Low trust is rational when a tool has not proven itself on your calls and your customers. The scorecard is how you build it. Four weeks of Friday audits with clean results gives you a reason to stop listening to every recording. Four weeks of errors gives you a reason to cancel. Either answer beats paying for a tool on faith.
The Decision to Make Before November
Heating season will hand you the stress test whether you plan for it or not. The first cold snap floods phones with no-heat calls, and every AI tool you run will either book those jobs cleanly or create a mess you clean up at 10pm. Run the scorecard in October, while volume is still manageable and mistakes are cheap.
The decision for this month: pick one AI workflow, write down its baseline and its reassignment rule by October 9, and put the day-30 review on your calendar for November 6. If you cannot name a dollar number it moved by then, cancel it and redirect that budget to the workflow that did. The 47% of contractors who see AI in their profits did not buy better tools than the other 53%. They decided where the saved time goes.
Atlas Unchained helps trades operators set up these scorecards and wire AI into the platforms they already use. If you want the one-page scorecard template, reply to The AU Brief or reach out through our contact page, and subscribe for a field-tested play every week. We read every reply, and the best questions from trades owners shape the next Thursday post.
