If you run a Quebec SMB and you deployed a voice AI agent six months ago, I have one question for you. Not how much it costs. Not how many calls it handles. How much it earns you — to the dollar, per month, on a sheet you can show your accountant.
Drawing a blank? You're not alone. According to G2's 2026 report on AI voice assistant adoption, almost half of organizations say they're feeling the benefits of their voice agent — faster response, lower service cost, self-service that finally works — without having any tool to put a number on what they feel. The return is there. It's just invisible.
And invisible, in business, is dangerous. Because a leader who can't show ROI on a technology will eventually cut it at the next belt-tightening. Not because it doesn't work. Because no one bothered to measure it.
Why Quebec SMBs are (almost all) measuring the wrong things
When I sit down with a clinic owner, a law firm partner, or a garage operator in Laval or Sherbrooke and ask how they evaluate their voice agent, the answer is almost always the same: "Well, it takes calls, and I pay it $200 a month."
That's it. That's the KPI. Monthly cost divided by call count.
The problem is that this measurement misses just about everything that creates value: calls that would have been abandoned at 7 PM, appointments booked at 6 AM by an insomniac client, the staff turnover you no longer have to absorb, the time your receptionist now spends with on-site customers instead of on the phone. Those are the numbers that turn a voice AI agent from a "cute tool" into a real business asset.
And it's exactly six numbers — not one, not three — that you need. Here are the ones your platform should be giving you, and what to do with them.
Number 1: Call containment rate
Containment rate is the percentage of calls your agent handles end-to-end, without escalating to a human. It's the first metric you should demand from any vendor. According to Sobot's 2026 benchmarks for AI contact centers, a well-designed voice agent should hit 70–85% containment on routine requests — appointment booking, status checks, FAQ.
Below 60%? Your agent is a smart answering machine, not an agent. Above 85%? You're literally saving a part-time salary every month.
The trap: don't confuse answer rate with containment rate. An agent can answer 100% of calls and only contain 30%. The first number flatters your ego. The second fills your bank account.
Number 2: Cost per contained call, vs. human
Here's the calculation nobody does. Take your monthly platform bill (minutes, LLM, TTS — everything in the real cost of a voice AI agent for a Quebec SMB). Divide it by contained calls (not total — contained). Compare that to what a human would have cost for those same calls at Quebec's loaded labor rate ($28/hr on average for a receptionist with payroll burden, per Statistics Canada 2025).
Concrete example. A dental practice in Trois-Rivières pays $380/month for its voice agent. Of 1,100 inbound calls, 740 are contained. Cost per contained call: $0.51. A human would have spent about 4 minutes per call — 740 × 4 min × $28/hr ÷ 60 = $1,381. Net savings: $1,001/month, or $12,012/year. Without counting the owner's sleep.
You see? The number existed. You just had to calculate it.
Number 3: After-hours capture rate
This one's my favorite, because it's almost always where the hidden ROI lives. After-hours capture rate is the percentage of calls received outside your normal business hours (evenings, nights, weekends) that your agent actually handled instead of letting them fall into a voicemail nobody ever listens to.
For a Quebec City plumber, it's the Sunday-night emergency. For a lawyer, it's the mandate no human would have caught at 10 PM on a Tuesday. For an Eastern Townships hotel, it's the Boston traveler booking at 11:30 PM Eastern.
A typical Quebec SMB gets 35–50% of its calls outside normal hours. If your agent captures 80% of that volume and converts even 12% into customers, you've just turned a flat loss into gross margin. Do the math with your average customer value.
Number 4: Revenue per contained call
You don't need an MBA for this one. If your agent books 740 appointments a month, and your average value per appointment (dental, legal, vet, real estate — whatever) is $320, your agent contributed to a $236,800/month pipeline.
No, it didn't generate all of it (the customer was calling anyway). But it locked in the conversion. Before the agent, how many of those calls ended in "call me back tomorrow" followed by a customer going to the competition? That's the friction the agent removes, and that friction was probably worth 5–15% of your inbound revenue.
If you want to measure cleanly: compare your human-vs-agent conversion rate on the same call type, over 60 days. The difference, multiplied by your average value, is the net incremental revenue from the agent. No other metric gives you that.
Number 5: First contact resolution (FCR)
FCR — first contact resolution — measures the percentage of calls where the customer leaves with their answer without having to call back. It's a quality KPI, not a financial one, but its indirect financial impact is huge. A customer who has to call back three times means three times the operational cost, and it's also a 2-star Google review in the making.
Aim for 75% FCR minimum on routine requests. Below that, your agent is under-trained or its knowledge base is incomplete. Above 85%, you're in the top 10% of Quebec deployments — based on what we see at TECHMA.
And here's an often-forgotten point: an agent that escalates well (without frustrating the caller) is worth more than one that answers badly out of stubbornness. If you can spot frustration signals in real time, your FCR stops being a passive indicator and becomes an operational lever.
Number 6: "Time-to-live" — days until the first dollar saved
This one, few vendors will admit out loud, but it's probably the most telling. How many days between signature and the first call your agent handles solo, in production, with a real customer? At TECHMA, our Quebec SMB deployments run 14 to 21 days, CRM integrations included (we handle it — you don't touch an API).
According to a French study of 47 AI projects in 2025, the median payback timeline lands at 4.2 months for voice projects. In the U.S., platforms report 3 to 6 months depending on SMB size. Which means an agent activated at the end of May 2026 should, in theory, be profitable before fall — if, and only if, you're measuring the right numbers.
What the May 12 ElevenLabs update changes
Quick technical aside, because it's relevant: ElevenLabs released API version 2.47.0 on May 12, 2026, adding — among other things — workspace-level request analytics and RAG chunk listing (see the official changelog). In practical terms, that means as of now, the KPIs above are measurable without an external script. The platform exposes natively what you had to reconstruct by hand six months ago.
If your vendor doesn't give you access to that data, it's no longer a technical limit. It's a choice.
The overlooked link between compliance and ROI
One last thing, and it's Quebec-specific. Law 25 changed the game in 2024–2025. A non-compliant voice AI agent can look profitable… until the day the CAI sends you a letter. The maximum penalty for an SMB is $10 million or 2% of global revenue.
The ROI of a compliant agent is therefore gross minus the discounted regulatory risk. Most SMBs ignore this calculation. If you want to do it right, follow the 9 steps for a Law 25–compliant voice AI agent deployment. At TECHMA, we include this by default in every engagement — not out of kindness, but because without it, the ROI is an illusion.
How to start tomorrow morning
Three actions, in order. Ask your vendor for a monthly report containing the six numbers above. If they can't deliver them, it's probably time to look elsewhere — and our Vapi vs ElevenLabs vs Retell comparison is a good starting point. Next, set internal thresholds (minimum containment rate, FCR target, etc.) and revisit them quarterly. Finally, share these numbers with your team — a voice agent that's just a black box will never get internal adoption.
The ROI is there. It's waiting. You just have to stop staring at your invoice, and start staring at your operations. The dirty secret of voice AI in 2026 isn't that it doesn't pay back — it's that most SMBs are sitting on returns they have never bothered to count, and a competitor down the street already has.
Based in Quebec City, Montreal, Laval, or Sherbrooke and want to audit the real ROI of your current voice AI agent — or deploy one with the six numbers baked in from day one? The TECHMA team handles deployment and integrations end-to-end (CRM, calendar, Law 25 compliance). Let's talk.