Quick answer: The single most predictive question is "who specifically does the build, and how many years have they run operations in a business my size?" Operator-led agencies ship working systems. Account-led agencies ship plans.
If you run a Canadian SMB and you're evaluating AI agencies, this guide is the shortlist of questions Compass wishes more buyers asked us — and that the right agency will answer plainly.
Eight questions to ask any AI agency
1. Who specifically does the build?
The cleanest tell of agency quality. The right answer names a person, their years of operational experience, and the proportion of the build they personally write. "We have a senior team" is a non-answer.
What to listen for: Real names, real experience, real ratios. If the senior who pitches isn't the senior who builds, the operational context evaporates in the handoff.
2. Will you map my process before designing anything?
The difference between a $20,000 build that pays back and a $20,000 build that gathers dust is whether the agency understood your actual process before designing the AI for it.
What to listen for: Phrases like "current-state map," "value-stream," "Gemba walk," "process discovery." The right agency runs Lean tools whether or not they use the word "Lean."
3. Are you building or reselling?
Some Canadian "AI agencies" resell third-party SaaS with a thin services wrap. Nothing wrong with SaaS, but you should know which you're buying.
What to listen for: "We build the agents and integrations on your infrastructure" (custom build) versus "We implement [platform X]" (reseller). The pricing tells the truth too — SaaS resellers usually quote a monthly subscription plus implementation fee.
4. What metric will improve, and by how much?
A real engagement names the bottleneck, the metric, and the result up front. Vague promises of "transformation" are a flag.
What to listen for: Specific numbers. "We expect proposal turnaround to drop from 6 hours to 30 minutes" is right. "AI will transform your business" is sales.
5. How do you handle PIPEDA, and any vertical-specific Canadian regulations?
Canadian regulation shapes how data can move and where AI can sit in the workflow. US-flavoured shops trip on this. PIPEDA is the universal baseline; verticals like legal (Law Society) and real estate (RECO) add their own.
What to listen for: Specific naming. "PIPEDA-compliant data handling," "Law Society of Ontario standards-aware," "RECO-aware." If the agency hesitates, ask for a written summary of their data handling.
6. What does a small engagement cost, and what does a big one cost?
Toronto AI agencies range from $5,000 to $1M per engagement. The right one publishes price ranges, scope ladders, or offers a free diagnostic. Refusal to discuss pricing without a long sales process is a flag.
What to listen for: Real ranges. "Focused builds run $8,000–$25,000, full-stack platforms $50,000–$150,000" is a useful answer. "We need to scope before we can discuss pricing" without a free diagnostic is sales.
7. How long until I see something working?
Working AI in 4–8 weeks is achievable for most SMB engagements. Working AI in 6+ months is usually a sign of scope creep, technology mismatch, or an agency that staffed the project wrong.
What to listen for: Iterative delivery. Phrases like "weekly review," "pilot then scale," "Plan-Do-Check-Act." The right agency ships small slices and earns the next phase.
8. What happens when something breaks?
AI workflows fail. The question is how the agency handles it: monitored runtime with auto-escalation, or radio silence until a client complains?
What to listen for: Specific operational answers. "Every workflow has logged execution, error alerts to my mobile, and an audit trail per run" is right. "We're confident in our quality" is a non-answer.
Red flags
Three patterns that consistently predict a bad engagement:
Reflexive use of buzzwords. "Transform," "unlock," "leverage," "ecosystem," "synergy," "paradigm" — if the pitch can't survive the buzzword filter, the engagement won't survive contact with your actual business.
No published pricing or scope examples. Hidden pricing is sales theatre. The right agency publishes ranges and uses a low-cost diagnostic (Compass's is the $2,500–$3,500 Bearings Report) to scope precisely.
Pre-signing pressure on contract length. Annual contracts at month one are a flag. The right agency earns the renewal with the first build.
Where Compass fits
Compass is an operator-led Canadian AI agency. Bobby Atwal does the build personally, 30 years of inside-the-business experience. Process discovery on every engagement using Lean tools we actually use. PIPEDA-compliant, Law Society and RECO standards-aware. Pricing ranges on every service page. Free 30-minute Bearings call before any quote.
If that pattern fits, book a Bearings call. If you're earlier in your evaluation and want to compare options honestly, the Compass vs. alternatives matrix lists where Compass is the right call and where we aren't.
Frequently asked questions
Are AI agencies in Toronto expensive? Range: $5,000 (focused chatbot build) to $1M+ (enterprise platform). For Canadian SMBs, the realistic working range is $10,000–$50,000 for a first useful build. Anything quoted under $5,000 is usually SaaS resale; anything quoted over $150,000 for a first SMB engagement is usually scope inflation.
What is the difference between an AI agency and a development shop? Development shops write the code. AI agencies map your process, design the AI for it, then write the code. The right AI agency includes Lean discovery work that a dev shop typically skips.
Do I need to be technical to work with an AI agency? No. The right agency translates between your operation and the technical build. If you're sitting in pitch meetings and the agency is using language you don't understand, that's the agency's failure, not yours.
Should I sign with a US AI agency or a Canadian one? For Canadian SMBs handling Canadian customer data, a Canadian agency is usually the better fit — PIPEDA, Law Society, RECO, CRA all carry weight here. US agencies can do good work but often require you to absorb the regulatory gap.
What is the smallest useful AI agency engagement? For most Canadian SMBs, $8,000–$15,000 buys a focused single-workflow automation or a voice-only AI receptionist. That's a useful first build. Anything smaller is usually consulting or a pilot, not a working system.