The Missed-Call Math: An Honest AI Receptionist ROI Worksheet for Trades
Here's the honest version of AI receptionist ROI: it pays for itself when the monthly value of the calls it captures exceeds its monthly cost. That's it. No magic, no "10x," no hockey-stick chart — just six numbers you can measure, one multiplication chain, and a clear-eyed look at when the math doesn't work. The worksheet below, with a fully worked example using clearly labeled illustrative assumptions, is the same one we walk through with plumbing, electrical, HVAC, and roofing companies before anyone signs anything.
A note on how we work, since it shapes this post: we build AI receptionists — human-assisted AI, not AI-assisted humans. The system answers, books, and follows up; a person stays in the loop where judgment matters. Our pricing is transparent and monthly from $1,000 CAD, with no long-term contracts. That price is the cost side of the worksheet below, and we publish it because an ROI calculation with a hidden cost is a sales trick, not math.
How many calls are you actually missing?
Before the worksheet, the measurement problem. Most trades owners dramatically underestimate missed calls, because the calls they remember are the ones they answered. The missed ones left no trace in memory — but they left traces elsewhere:
- Carrier or phone-system logs. Your business line's call log is the ground truth. Count inbound calls and missed/abandoned calls over the last 30 days. This takes ten minutes and ends all debate.
- Voicemail count versus callback count. If the voicemail box fills and half never get called back, those are missed calls with extra steps.
- After-hours volume. Pull the log filtered to evenings and weekends. For trades, this is often where the miss rate is worst — and where emergency jobs (the highest-value calls) arrive.
- The one-week tally. If logs are hard to get, have whoever answers the phone tally every inbound call and every missed call for one normal week. Imperfect, but directionally true — and direction is all the worksheet needs.
What you typically find: a busy trades company misses 15–35% of inbound calls during working hours (on a job, on a ladder, under a sink) and most of them after hours. Write down your number. Everything below runs on it.
The worksheet: six numbers
The entire ROI case for an AI receptionist reduces to six inputs. Gather them honestly — guesses are fine if you label them as guesses, because the worksheet still tells you which guesses matter.
| Input | What it is | How to get it |
|---|---|---|
| A. Inbound calls per month | Total calls to your business line | Carrier or phone-system log, last 30 days |
| B. Missed-call rate | Share of A you never answer | Same log; missed ÷ total |
| C. Quote-worthy rate | Share of missed calls that are real job inquiries (not spam, wrong numbers) | Your judgment from call-backs; typically 50–70% for trades |
| D. Close rate | Share of quotes that become paying jobs | Your books; quotes ÷ jobs won |
| E. Average job value | Mean revenue per completed job | Your books; be honest, use the median if outliers skew it |
| F. Monthly system cost | What the AI receptionist costs | Ours: from $1,000 CAD/month, no long-term contract |
The formula:
Monthly recovered revenue = A × B × C × D × E
Monthly net = (A × B × C × D × E) − F
That's the whole model. Missed calls, filtered to real inquiries, filtered to won jobs, times job value, minus cost. If monthly net is positive, the system pays for itself. If it's deeply positive, the only question is implementation. If it's negative, don't buy it — from us or anyone.
The assumption to interrogate: this model assumes a captured call converts at your normal close rate. That's fair if the AI receptionist books and qualifies properly — a booked appointment with job details is at least as good as a voicemail you return two hours later, and usually better, because speed-to-lead is real. What the model does not assume: that the AI closes jobs (it doesn't — your techs do), or that every missed call was winnable (that's what input C is for).
A worked example (illustrative numbers)
Let's run it for a hypothetical residential electrical company. Every number below is illustrative — a plausible mid-size trades operation, not a client, not a promise. Plug in your own numbers; the structure is what matters.
- A. Inbound calls: 80 per month (about 20/week — a steady residential shop)
- B. Missed-call rate: 25% → 20 missed calls/month (on jobsites all day, no office staff)
- C. Quote-worthy rate: 60% → 12 real inquiries/month (the rest: spam, suppliers, wrong numbers)
- D. Close rate: 50% → 6 won jobs/month
- E. Average job value: $450 (mix of service calls and small installs)
- F. Monthly cost: $1,000 CAD
Recovered revenue: 20 × 0.60 × 0.50 × $450 = $2,700/month.
Monthly net: $2,700 − $1,000 = +$1,700/month.
Break-even check: at $1,000/month cost and $450 average job value at a 50% close rate on 60% quote-worthy calls, each missed call is worth 0.60 × 0.50 × $450 = $135 in expected revenue. Break-even is $1,000 ÷ $135 ≈ 7.4 missed calls per month — fewer than two per week. If you miss more than two calls a week, the system pays for itself on this math. Most shops we talk to miss that many before lunch on Monday.
Now stress-test it, because honest math includes the downside. Halve the close rate (25% — maybe the AI books less-qualified appointments): recovered revenue drops to $1,350, net +$350/month. Still positive, but thin. Drop the average job to $250 (handyman work, not electrical): recovered revenue $1,500, net +$500. The model is most sensitive to B (miss rate) and E (job value) — which is exactly why the first step is measuring your real miss rate and using your real ticket average, not industry folklore.
When the math doesn't work — the honest no-buy cases
We lose deals to this section and we're fine with that. An AI receptionist is the wrong purchase when:
- Call volume is low. If you get 15 inbound calls a month and answer 14 of them, there is nothing to recover. The worksheet will tell you this in thirty seconds. Marketing, not a receptionist, is your problem.
- You can't take the work. If you're booked out six weeks and turning jobs away, capturing more calls just creates a longer waitlist and worse reviews. Fix capacity first. (Our when AI isn't the answer post covers this family of problems.)
- Your close rate is the bottleneck. If you answer every call and still don't close, the problem is quoting, pricing, or reputation — an AI receptionist faithfully books appointments you then lose. Fix the close rate; the worksheet's input D will tell you.
- Your tickets are tiny and your miss rate is low. $80 average jobs with a 10% miss rate on 40 calls/month: 40 × 0.10 × 0.60 × 0.50 × $80 = $96/month recovered against $1,000 cost. Don't buy it. We'd rather tell you that now than onboard you into disappointment.
- Nobody will own it. The system needs one human owner — someone who checks the bookings, handles the escalations, and tunes the responses. "Set and forget" is how AI projects die quietly. This is the human-assisted part of human-assisted AI, and it's non-negotiable.
What the AI receptionist actually replaces — and what it doesn't
Let's be precise about the job description, because vague expectations are where AI projects go to die. An AI receptionist for a trades company does four things: answers every call (including the 7pm emergency and the Saturday morning), qualifies the inquiry (what's the problem, where, how urgent), books the appointment into your calendar with the details attached, and follows up on quotes and no-shows. What it doesn't do: diagnose over the phone, negotiate price, or replace the judgment of the person who shows up at the door.
Compared against the alternatives: a voicemail box returns calls hours later, after the customer has called three competitors — our receptionist vs booking system comparison walks through why booking beats messaging for trades. A human receptionist is excellent and costs multiples of the AI system — the AI vs hiring comparison runs that math honestly, including the cases where the human wins. An answering service takes messages; it doesn't book, qualify, or follow up.
The frame we use with every client: the AI handles the first five minutes of every call — the part that's pure process — so your people spend their time on the parts that need judgment. Human-assisted AI, not AI-assisted humans. The receptionist earns its keep on volume and consistency; your team earns theirs on craft. That's the division of labor, stated plainly.
Run your own numbers
Pull your last 30 days of call logs. Fill in A through F. If monthly net is positive by a comfortable margin, the next step is a conversation about implementation — what the system says, how it books, who owns the escalations. If it's marginal or negative, you've just saved yourself $1,000 a month and a disappointing quarter, and the worksheet did its job either way.
One more honest note: this worksheet values only recovered calls. It doesn't count the follow-up revenue (quotes that would have gone cold get chased automatically), the review lift (every job completed is a review opportunity the system can request), or the after-hours emergency jobs that never hit voicemail because a human-sounding system picked up. Those are real, they're just harder to model — so we leave them out of the worksheet and let them be upside. We'd rather underpromise here. (For the broader cost picture, see how much AI costs a small business and is AI worth it.)
Frequently asked questions
How is AI receptionist ROI actually calculated?
Monthly recovered revenue = inbound calls × missed-call rate × quote-worthy rate × close rate × average job value, minus the monthly system cost. Six inputs, one multiplication chain. Measure the inputs from your own call logs and books — the model is only as honest as its inputs.
How many missed calls does it take to break even?
It depends on your ticket size and close rate. In our illustrative example ($450 average job, 50% close rate, 60% of missed calls quote-worthy), each missed call is worth about $135 in expected revenue — so break-even against a $1,000/month system is roughly 7–8 missed calls per month, fewer than two per week. Run your own numbers.
What does an AI receptionist cost?
Our pricing is transparent and monthly from $1,000 CAD, with no long-term contracts. The broader market ranges widely — beware pricing that hides per-minute or per-call overages, which is where "cheap" systems get expensive.
Will customers know they're talking to an AI?
They shouldn't be deceived — and in our deployments they aren't. The system identifies itself appropriately, handles the first five minutes (qualifying, booking), and escalates to a human whenever judgment is needed. Human-assisted AI, not a robot pretending to be your office manager.
When is an AI receptionist NOT worth it?
Low call volume, no capacity to take new work, a close-rate problem (you answer but don't win), tiny tickets with a low miss rate, or no human owner for the system. The worksheet in this post will tell you in thirty seconds — we'd rather you run it than buy on hope.
Does it replace a human receptionist?
It replaces the first five minutes of every call — answering, qualifying, booking, follow-up. It doesn't replace judgment, complex problem-solving, or the relationship your office manager has with repeat customers. Many clients run both: the AI catches everything, the human handles what matters. See our honest AI vs hiring comparison.
About this post
Creatrixe is a Canadian AI consultancy. We build AI receptionists, lead-capture, and follow-up systems for local businesses in Canada and the GCC — human-assisted AI, with a person in the loop where it matters. Transparent monthly pricing from $1,000 CAD, no long-term contracts. The math in this post uses clearly labeled illustrative assumptions; your numbers will differ, and we'll help you run them honestly before you spend anything.
Want us to run the worksheet on your numbers?
Bring your last 30 days of call logs to a 20-minute call. We'll fill in the six inputs together and tell you straight whether the math works — including when it doesn't.