Key takeaways
- Return comes from four places: recovered jobs, faster urgent response, staff time saved and after-hours coverage.
- Use gross margin, not revenue, for recovered jobs, and use loaded labor cost for staff time; benefits were 30% of private employer compensation costs in June 2026[1].
- Count callbacks honestly: only 19% of US adults generally answer calls from unknown numbers[2].
- Compare against a measured baseline and the same season, or the numbers will flatter or punish the tool unfairly.
What return means for call handling
Return on AI call handling is simple to define and easy to get wrong. The definition: the extra gross profit and saved costs the tool produces, minus everything it costs you, over a set period. The mistakes come from counting revenue instead of margin, counting jobs that would have come anyway, and forgetting the time your team spends setting up and reviewing the system.
Write the calculation down before the tool goes live, using your own baseline, and run it monthly. A number you calculated in advance, with rules you agreed, is far more useful than a vendor's dashboard or a gut feeling after a busy week.
This guide uses illustrative inputs in its worked example. Replace every one with your own figures from your field-service software, payroll and phone records.
The four sources of return
- Recovered jobs: calls that previously went to voicemail, rang out or were abandoned on hold, and now end with a booking or an urgent dispatch.
- Faster urgent response: emergencies reached in minutes rather than after a voicemail check, which wins jobs and reduces damage disputes.
- Staff time: fewer callbacks, fewer reschedule calls handled by hand, less message-taking at lunch and during busy spells.
- Coverage without overtime: nights, weekends and holidays covered without paying someone to sit by the phone.
Most shops find the first source is the largest, the second the hardest to measure, and the third smaller than expected but real. The fourth depends entirely on how you covered phones before.
The formulas
| Measure | Formula | Where the inputs come from |
|---|---|---|
| Recovered calls | Calls answered by AI that would previously have been missed | Phone system before and after |
| Recovered jobs | Recovered calls ร share that became booked jobs | AI records matched to jobs |
| Recovered gross profit | Recovered jobs ร average ticket ร gross margin | Field-service software |
| Staff time saved | Minutes saved per day ร days ร loaded hourly cost | Time sample, payroll |
| Overtime avoided | After-hours hours no longer paid ร overtime rate | Payroll |
| Total cost | Tool fee + setup hours + review hours, at loaded cost | Invoice, time log |
| Net return | Gross profit + time saved + overtime avoided โ total cost | All of the above |
Getting the inputs right
Average ticket and gross margin should come from your own software for the job types the AI actually books, not your overall average. Drain cleanings and water heater replacements produce very different tickets, and after-hours calls often differ from daytime ones. As a sense check on scale, plumbing, heating and AC contractors generated about $260,000 in receipts per employee in 2022[3]. If your recovered-job figures imply that the phones alone add more than a full employee's worth of revenue in a month, recheck the matching of calls to jobs before celebrating.
Staff time needs a loaded cost, not just the wage. In June 2026, private-industry employers paid an average $46.89 per hour worked in total compensation, with benefits making up $14.07 an hour, or 30.0%, and wages 70.0%[1]. For office roles, start from the wage you actually pay; for reference, the median US customer service representative earned $21.53 an hour in 2025[4]. Add benefits, payroll taxes and workers' comp to get closer to the true cost of an office hour.
- Wages and salaries 70%
- Benefits 30%
Source: U.S. Bureau of Labor Statistics, 2026[1]
Show the data
| Item | Value |
|---|---|
| Wages and salaries | 70% |
| Benefits | 30% |
For recovered calls, use your phone system's records from a baseline period: missed calls, voicemails and abandoned calls in the hours the AI now covers. Remove repeat calls from the same number on the same day, or you will count one frustrated homeowner as four lost jobs.
Callbacks are where the hidden loss sits
Many owners assume voicemails are not really lost because someone calls back. Test that assumption. Only 19% of US adults generally pick up calls from unknown numbers[2], so your callback may go unanswered even when the caller still needs help. For a fair baseline, track for two weeks how many after-hours voicemails led to a booked job, how many callbacks connected at all, and how long each took. That conversion rate, not the voicemail count, is what the AI is competing with.
Worked example
Example inputs, all fictional: a shop's baseline month shows 120 after-hours calls going to voicemail or ringing out. Its records show 25 of those became booked jobs after callbacks. With AI answering, 112 after-hours calls are answered in the same month a year later, and 52 become booked jobs or urgent dispatches.
Example recovered jobs: 52 minus 25 is 27 extra jobs. Using an example average after-hours ticket of $450 and an example gross margin of 50%, that is 27 ร 450 ร 0.5, or about $6,075 in recovered gross profit for the month.
Example staff time: the office previously spent about 45 minutes each morning returning overnight voicemails, now about 15 minutes reading summaries. Thirty minutes saved over 22 working days is 11 hours. At an example loaded cost of $30 an hour, that is $330.
Example costs: the tool's monthly fee, which you would take from your own quote, plus 4 hours of rule updates and 6 hours of review time at $30, or $300. Net return in this example is $6,075 plus $330, minus $300, minus the tool fee. Whether the result is worth it depends on that fee and on your own numbers, which is why every input here is labelled Example.
Costs people forget
- Setup time: writing the knowledge base, booking rules and the on-call path, often 10 to 20 hours in the first month.
- Review time: reading summaries and listening to samples, heavier in the first weeks.
- Errors: the cost of a wrong booking, an unnecessary truck roll or a missed urgent call. Count them honestly.
- Seasonal updates: rules need revising before summer and the freeze season.
- Phone system changes: call forwarding, number porting or new routing may carry their own cost.
Attribution traps
- Season: comparing a January pilot with an October baseline mixes freeze calls with normal demand. Compare like months, or adjust.
- Marketing changes: a new ad campaign during the pilot raises calls regardless of who answers.
- Callers who would have called back: some customers ring again in the morning. Count only jobs where the record shows the AI-handled call led to the booking.
- Double counting: an urgent call routed to the on-call tech is one job, not a recovered call plus a dispatch.
- Revenue as profit: always multiply recovered tickets by your gross margin.
Value the formula leaves out
Some benefits are real but hard to price, and it is better to list them separately than to stretch the formula to include them. Note them in the monthly review so they are part of the decision without distorting the numbers:
- Fewer late-night calls to the owner's personal cell, and fewer interrupted evenings for the on-call tech when calls are screened properly.
- Better records: every call has a summary and recording, which helps settle disputes and train new staff.
- Reputation: fewer reviews complaining that nobody answered.
- Resilience: phones keep working when a CSR is sick, on vacation or leaves.
- Data: a clearer picture of when calls arrive and what callers want, which helps with staffing and marketing decisions.
Equally, list the soft costs: callers who disliked the experience, staff who found the change stressful, and time spent explaining the new process. A fair review weighs both lists alongside the hard numbers.
A monthly return review
Pull the records
Export AI call records, phone system data and jobs booked in the period.
Match calls to jobs
Link each booked or dispatched job to the call that created it; mark unclear cases as unattributed rather than guessing.
Apply the formulas
Use the same formulas and the same margin and labor assumptions every month.
Subtract every cost
Tool fee, setup, review and error costs, including the time of everyone involved.
Compare and decide
Compare with the baseline and the same month last year, then decide whether to widen scope, tighten rules or change nothing.
Add the monthly result to your weekly scorecard so it is reviewed alongside your other numbers, not in isolation.
Frequently asked questions
How do I calculate ROI on an AI answering tool?
Add the gross profit from recovered jobs, the value of staff time saved and any overtime avoided, then subtract the tool's fee and the setup and review time. Compare with a measured baseline.
Should I use revenue or profit for recovered jobs?
Gross profit. Multiply recovered jobs by your average ticket for those job types and by your gross margin, or the result will be overstated.
How long before ROI is clear?
Usually one to three months, once setup time is behind you and you can compare a full month with a like-for-like baseline month.
What if I didn't measure a baseline?
Use phone system records from the same months last year, or run a short period without the tool. Without a baseline, any return figure is a guess.
References
U.S. Bureau of Labor Statistics (2026). Employer Costs for Employee Compensation - June 2026 (opens in a new tab) โ
Pew Research Center (2020). Most Americans don't answer cellphone calls from unknown numbers (opens in a new tab) โ
U.S. Census Bureau (2022). Statistics of U.S. Businesses 2022: U.S. and states, 6-digit NAICS (opens in a new tab) โ
U.S. Bureau of Labor Statistics (2025). Occupational Outlook Handbook: Customer Service Representatives (opens in a new tab) โ