A small clinic owner in Levis called me in April. His opening line: "Okay, but how much really? Not the marketing number. The actual bill." Good question. Too good for most ROI calculators online, which throw out "800% return" using $65,000 US receptionist salaries and $400 missed calls — numbers harvested from studies done in San Francisco or New York.
That doesn't apply to a small business in Trois-Rivières, Sherbrooke or Saint-Hyacinthe. Here is the honest math, in Canadian dollars, with Quebec payroll taxes, real platform costs I see billed to my clients every month, and one factor nobody includes: Law 25.
The trap of American ROI calculators
Visit Krispcall, Retell, Plura or Nextiva — their calculators all give you the same fantasy result: 300% to 1,500% first-year ROI, with payback in 30 to 45 days. Retell AI publishes these numbers themselves. The problem? The base assumptions.
These calculators start from:
- Receptionist salary: $50,000 to $65,000 USD fully loaded (≈ $70,000 to $91,000 CAD)
- Missed call value: $250 to $400 USD
- Platform cost: $0.05 to $0.07 USD per minute
- Zero line for compliance, security or internal training
A typical Quebec small business pays a receptionist $36,000 to $45,000 CAD base salary. Add Quebec payroll burden (QPP, EI, QPIP, HSF, CNESST) and you reach $42,000 to $52,000 per year. That is 30 to 40% lower than the American number. So the ROI calculated by these tools is inflated by the same amount. Not slightly: significantly.
The real bill of an AI voice agent for a Quebec small business
Here is what we actually bill our clients in 2026 — restaurants, auto shops, salons, clinics, law offices, professional services. Not a list price. The monthly average from my own books.
Implementation cost (one-time)
A serious deployment for a Quebec small business runs $1,500 to $4,500 CAD depending on complexity: integration with the booking system, CRM connection (Zoho, HubSpot, Salesforce), knowledge base creation, bilingual FR/EN scripts, simulation testing. This includes everything the knowledge base needs in preparation. Self-serve $99/month solutions cover none of this — you do it yourself, badly, and it costs more in time.
Recurring cost (monthly)
For typical small business volume (20 to 80 calls per business day, or ~500 to 2,000 minutes monthly):
- Platform + LLM (ElevenLabs or equivalent + Claude/GPT): $0.12 to $0.18 CAD per minute all-in
- SIP telephony (Twilio CA or local equivalent): $0.015 CAD per inbound minute
- Maintenance + supervision: $80 to $200 monthly for prompt tuning, quality monitoring, knowledge base updates
Typical monthly total: $200 to $600 CAD. Annual: $2,400 to $7,200 CAD. Not $99/month like the ad promised, but not $65,000 either.
The line nobody writes: Law 25
If your AI voice agent vendor hosts data in the US (most do), you must: complete a Privacy Impact Assessment (PIA), obtain specific consent for cross-border transfer, and document the data flow chain. Outsourced to a law firm: $2,000 to $8,000 CAD. With an integrator who handles it in the engagement (like TECHMA), it's included. The Commission d'accès à l'information du Québec has indicated routine compliance audits will become common in 2026, with penalties up to $25M for repeat offenses.
The Quebec formula (genuinely honest)
Here is the formula I use with my clients, adjusted for Quebec:
Annual ROI = (Salary Savings + Recovered Revenue − (Setup + Recurring + Compliance)) ÷ Total Investment × 100
Where:
- Salary savings: not the full salary — only the automatable portion (booking, FAQ, qualification, routing). Realistic: 30 to 60% of receptionist time, or $12,600 to $31,200 CAD/year
- Recovered revenue: missed calls × transaction value × recovery conversion rate. In Quebec: ~$80 per missed call for a salon, ~$300 for a dental clinic, ~$600 for a garage
- Total investment: setup + 12 months recurring + PIA
Three concrete Quebec scenarios
Scenario 1 — Hair salon (Sherbrooke, 4 chairs)
Volume: ~25 calls/day, 6 missed. Average value: $80.
- Year 1 cost: $2,200 (setup) + $3,600 (recurring) = $5,800
- Salary savings: $0 (no receptionist; the lead stylist answers)
- Recovered revenue: 6 calls × $80 × 50% conversion × 250 days = $60,000
- ROI = ($60,000 − $5,800) ÷ $5,800 × 100 = 935%
Verdict: it pays. Consistent with the Quebec salon analysis.
Scenario 2 — Accounting firm (Levis, 6 employees)
Volume: ~15 calls/day, mostly existing clients. Part-time receptionist: $24,000/year.
- Year 1 cost: $3,500 (setup) + $4,800 (recurring) = $8,300
- Salary savings: 35% automatable time × $24,000 = $8,400
- Recovered revenue: low, since most calls are existing clients
- ROI = ($8,400 − $8,300) ÷ $8,300 × 100 = 1.2%
Verdict: year one barely breaks even. By year 2 (no setup cost), ROI climbs to 75%. Not a case I push hard — I tell clients honestly.
Scenario 3 — Dental clinic (Trois-Rivières, 3 dentists)
Volume: ~60 calls/day, 12 missed (lunchtime peak). First-appointment value: $300.
- Year 1 cost: $4,200 (setup) + $6,600 (recurring) = $10,800
- Salary savings: 1 receptionist reassigned 40% of her time to higher-value patient management. Avoided second hire: $22,000
- Recovered revenue: 12 × $300 × 35% conversion × 250 days = $315,000 gross
- At 25% net margin: $78,750 recovered profit
- ROI = ($78,750 + $22,000 − $10,800) ÷ $10,800 × 100 = 833%
Consistent with the dental clinic missed-call anatomy: gross numbers look enormous but, once adjusted, remain excellent.
When the ROI is NOT there
I turn down roughly one in four small businesses. Here are the patterns where the math doesn't hold:
- Low volume with loyal clients: under 8 calls/day, few missed, few prospects. ROI is marginal.
- Long, complex calls: if every call needs 12 minutes of expert human qualification, the AI agent will frustrate clients without saving real money.
- No CRM or booking system: most of the gain comes from booking automation. Without a digital system to plug the agent into, you lose 60% of the value.
- Extreme seasonal swings: 3 months at 200 calls/day and 9 months at 5 calls/day means the agent mostly pays for itself during peaks — fine but not spectacular.
The TECHMA method: ROI or no signature
For every small business that contacts me, I run this math before proposing an engagement. I present three lines:
- Total all-in cost in CAD over 3 years
- Recovered revenue and salary savings, computed with the company's actual numbers (not a sector average)
- Pessimistic-scenario payback in months
If payback exceeds 18 months, I say so. I would rather lose a sale than sell a bad fit. That is what separates a serious integrator from a self-serve platform: the right LLM, the right prompt, the right integration are worth nothing if the use case is wrong from day one.
Three questions to ask yourself before investing
- How many calls a day? Under 10, rarely worth it. From 15 to 40, the sweet spot. Over 60, almost always a slam dunk.
- What percentage of my calls go unanswered? If you don't know, that is itself a signal — the first week after deployment usually reveals a missed-call rate much worse than estimated.
- Are my systems (CRM, booking, calendar) connected? If not, add $1,500 to $3,000 to setup to integrate them.
What changes between year 1 and year 3
One number that ROI calculators always ignore: the math gets dramatically better in year 2. The setup cost ($1,500 to $4,500) is amortized once, then disappears. The agent's accuracy also improves — every call adds to the knowledge base, every misroute gets corrected, every awkward phrasing gets rewritten. By month 12, the agent typically handles 15 to 25% more call volume autonomously than at month 3.
For the dental clinic in scenario 3, year 2 economics look like this:
- Year 2 cost: $0 setup + $6,600 recurring + $400 minor adjustments = $7,000
- Year 2 recovered revenue (improved 15%): $90,560 net profit
- Year 2 salary leverage: $26,400 (the receptionist's reassigned time generates direct revenue now)
- Year 2 ROI = ($90,560 + $26,400 − $7,000) ÷ $7,000 × 100 = 1,571%
This is why I always tell clients to evaluate over 3 years, not 12 months. A bad year-1 ROI can be a great year-2 ROI. A great year-1 ROI almost always becomes spectacular in year 2 once the setup amortization disappears.
The hidden cost most integrators won't tell you
One last line item, rarely discussed: the cost of not deploying. Every month you wait, your competitors who already deployed are capturing the calls you miss. In tight Quebec markets — dental clinics in Lévis, accountants in Drummondville, garages in Saint-Jérôme — once 2 or 3 competitors deploy AI voice agents, the missed-call patient or customer rarely comes back. They've found a place that answers. The opportunity cost of waiting 6 months in a competitive segment can easily exceed the entire 3-year cost of deployment.
That isn't fear-marketing. It's just the math when you run it across the full local market, not just one practice's books.
The honest conclusion of this math: for 60% of Quebec small businesses with moderate-to-high call volume, an AI voice agent in 2026 has real ROI — usually between 200% and 900% in year one — once adjusted for Canadian numbers. But it isn't universal. If someone promises 1,500% without looking at your books, run. If someone says "it depends, let's run the math together," you are in the right place.