How to choose an AI consultant in Canada: A checklist
A good AI consultant leaves you with something running in your own tools, plus a clear answer to who looks after it next month. Everything on this checklist tests for that. We sell AI consulting ourselves, so read it with that in mind. We wrote it as the list we'd want a Canadian business to bring to any first call, including one with us.
Do this before you call anyone
Write down the 2 or 3 problems that cost you the most. Use plain words: "web leads wait until the next morning for a reply", "someone retypes every supplier invoice into QuickBooks", "we hear about a competitor's sale from our customers". A consultant can work with that. "We want to use AI" gives them nothing to go on, and an unclear brief is how projects drift.
Next to each problem, note 3 things:
- the tools involved, like Gmail, Outlook, HubSpot, Salesforce, Shopify or QuickBooks;
- who on your team deals with it today;
- how you'd know it got better, such as reply time to new leads or hours spent on data entry each week.
This page of notes does 2 jobs. It keeps the first call on your business instead of the consultant's favourite tool, and it gives you a way to compare proposals side by side.
Questions to ask an AI consultant
These are the questions to ask a firm that offers AI automation consulting, or a solo consultant, on the first call. Good answers are specific to your business. Weak answers sound the same for every client.
- What would you build first, and why that one? You want a single project tied to one of your problems, with a reason it comes first. A list of 10 ideas with no order is a sales deck.
- Which of our tools would it connect to? The answer should name your systems: the inbox, the CRM, the store, the accounting software. If they plan to move you onto a new tool, ask what that costs your team in training and migration.
- Where does our data go, and who can see it? Ask which AI model provider the work uses, whether that provider can use your data to train its models, and which of your accounts the consultant needs access to. Ask how they handle personal information under PIPEDA.
- Who approves what the AI sends? For anything that reaches a customer, like a reply, a quote or an ad, a person on your side should approve it until you trust the output. Ask how the approval works day to day.
- What happens after launch? Automations break when a form changes, a field gets renamed or a model update shifts the output. Ask who watches the runs, who fixes failures, how fast, and who tracks model costs.
- How is the price set? Ask whether the first project has a fixed price, what's included, and what would trigger extra charges. Ask the same for ongoing support.
- Who owns what you build? The automations, prompts, logs and accounts should end up in your name. If the relationship ends, you should be able to keep running what you paid for.
- Can you show me one running? A live demo of a real workflow tells you more than slides do. Ask to see the run log, the approval step and what happens when something fails.
Red flags
The pitch starts with a tool
If the first call is a demo of a platform, a chatbot builder or a specific model before anyone asks how your team works, the consultant is selling that tool. Tools matter, but they come after the problem. A consultant who knows the field should be able to say which model or stack fits your case and why, and change it if your needs change.
No plan for running it after launch
Without that plan, the project ends at the handover meeting. The demo works, the consultant leaves, and 3 weeks later a renamed field in HubSpot breaks the workflow and nobody notices. Ask to see the plan for monitoring, fixes and cost control in writing. If the answer is "your team can maintain it", check that your team has the time and the skills.
Vague pricing
"It depends" is fair on the first call. It isn't fair in a proposal. By the time you sign, you should know the price of the first project, what it includes, what costs extra and what ongoing support costs per month. Open-ended hourly billing on a fuzzy scope puts all the risk on you.
No clear answer on data handling
PIPEDA is Canada's federal privacy law for businesses that handle personal information in commercial activity. Alberta, British Columbia and Quebec have their own private-sector privacy laws. A consultant working with your customer data should explain, without looking it up, what data the automation touches, where it's processed, which providers see it and how access is removed when the project ends. Hesitation here is a reason to walk away.
Results promised before they've seen your data
Nobody can promise hours saved or revenue gained before they've looked at your tools, your data and your process. A fair consultant gives you a range of options and tells you how they'll measure the result once it's live.
AI consultant, agency, freelancer or DIY
We'd compare the 4 options like this.
| Option | Good fit when | Watch for |
|---|---|---|
| AI consultant | You need a strategy, a roadmap or advice on what to buy | Advice that stops before anything is built |
| AI agency or consulting firm | You want something built, launched and looked after over time | Firms that build but don't stay to run and improve it |
| Freelancer | You have one well-defined task and someone in-house to own it | One person is a single point of failure when something breaks |
| DIY | You have a technical person with time, and the task is low-risk | Side projects that stall, and personal AI accounts holding company data |
Most small and mid-sized Canadian businesses don't need a strategy deck first. They need one automation that works and someone who keeps it working. That points to a firm that builds and runs the work, or to DIY if you already have the in-house skills. A pure strategy engagement makes more sense for larger companies with their own engineering team ready to build the roadmap.
Freelancers can do excellent work. The risk is continuity. If your one freelancer takes a full-time job, ask who will fix the workflow that routes your leads.
DIY works best for internal, low-stakes tasks, like summarizing meeting notes or drafting first versions of internal documents. Be careful with anything that sends messages to customers or moves money, and keep company data in company accounts.
What a good first project looks like
A good first project is small, close to revenue and easy to measure. In practice that means:
- One problem, one team. For example, every web form lead gets a drafted reply within minutes, built from what's already in HubSpot, and a salesperson approves it before it goes out.
- Tools you already use. It runs in your Gmail or Outlook, your CRM and your store. Your team doesn't learn a new app.
- A person approves the output at first. Approval can loosen once the team trusts what the automation produces.
- A measure you track yourself. Pick it before launch: reply time to new leads, invoices entered by hand per week, abandoned carts recovered. Write down the current number so you can compare.
- Live in weeks. If the first project needs 6 months of data work before anything runs, it's the wrong first project. Pick a smaller one.
- A named owner for after launch. Someone on your team, the consultant, or both, with the split written down.
Projects that touch revenue tend to make the case for the next one more easily. Faster replies to leads, calls answered after hours and follow-ups to lapsed customers all show up in sales your team can count.
Get these in writing before you sign
- The first project, the tools it connects and the measure of success
- The fixed price, what's included and what costs extra
- Which AI model providers the work uses and what data each one sees
- The accounts the consultant gets access to, and how that access ends
- Who monitors, fixes and improves the automation after launch, and what that costs per month
- That the automations, prompts and logs belong to you
If a consultant can put all 6 in a proposal without pushback, you've likely found a good one.
