AI consultant: how to choose one in Canada
An AI consultant helps a business decide where AI fits and builds it into the systems the business already runs. For a $5M to $50M manufacturer or distributor in Canada, a good one names the people who will do the work. It starts with a fixed-price stage on your own records. You end that stage owning the code and the accounts, and you know where your data is stored and processed. This guide covers the kinds of AI consultant, the questions to ask and the red flags. It also covers what you should own at the end, data residency and the public programs that pay part of the cost.
The kinds of AI consultant
Four kinds of firm sell AI consulting to a manufacturer or distributor your size. For what the work covers and how firms charge, see AI consulting.
| Kind | Fits when | Watch for |
|---|---|---|
| Big Four and large consultancies, such as Deloitte, EY, PwC and Accenture | AI is part of a company-wide ERP or platform rollout, or you need many people at once | Who works on your project, and whether your company is the size they build for |
| Boutique AI firms | You want one defined job built and tested on your own records | Named people, and who supports the system after launch |
| Freelancers | The job is small and has a clear end | Who maintains it if that person is unavailable, and whose name the accounts are in |
| Software vendors and their partners, such as your ERP reseller | You have already chosen the vendor’s platform | What it would take to change the model or the provider later |
Large firms say who they build for. Accenture launched Accenture Edge on June 23, 2026, for mid-market companies with annual revenues between US$300 million and US$3 billion. ThriveAI vs the Big Four compares what those firms publish.
Model makers and their investors also back services firms. On May 4, 2026, Anthropic, Blackstone, Hellman & Friedman and Goldman Sachs announced a services company to bring Claude into the operations of mid-sized companies. With any firm tied to one model maker, ask how you would change models later.
If you are weighing a permanent hire instead of a firm, see hiring an AI engineer in Canada.
Questions to ask an AI consultant
Ask every firm on your list the same questions, ThriveAI included, and get the answers in writing. Useful answers name your systems and your numbers.
| Question | A good answer | A weak answer |
|---|---|---|
| Which job would you start with, and why? | One job tied to hours or money in your business, chosen after looking at your records | A list of use cases, or a strategy phase before anyone sees your data |
| Who will do the work, and for how many days? | Names and days in the proposal, including who works on site | “Our team,” with nobody named |
| What will run on our records at the end of the first stage? | A working tool, the test it must pass and a date | A report or a slide deck |
| How will we know it worked? | Today’s number as a baseline, and a set of past jobs with known answers to test against | A satisfaction survey |
| Where is our data stored, and where is it processed? | A named location for each, the model, and the provider’s retention terms | “Hosted in Canada,” with no answer on processing |
| What do we own at the end? | The code, prompts, tests, data and accounts, in the company’s name | A licence that ends with the contract |
| What will it cost to run after you leave? | Each running cost named: seats, usage, hosting and support | “It depends,” with no breakdown |
| Who approves what the AI does? | A named person approves every message, order and ERP change | Fully automatic from the first day |
For how to build the set of past jobs behind the fourth question, see AI evals. Ask for a reference too: a manufacturer or distributor that has run the firm’s work for six months.
Put these questions to ThriveAI too
Tell Derik which job you want outside help with and which systems it touches. He will answer these questions for that job.
Start a conversationRed flags
- A long paid study before anything runs. Months of interviews and a roadmap come before anyone touches your records.
- A platform you have to license from the firm. Your system should run on accounts you hold, so you can change firms without starting over.
- “Hosted in Canada,” and nothing about processing. Storage and processing can sit in different countries, as the section on data residency below shows.
- Accounts in the consultant’s name. If the code, the model accounts or the keys that connect them sit with the firm, you cannot run or move the system without it.
- Savings promised before anyone has seen your data. Ask how the firm will measure the result on your records instead.
- No approval step. The AI sends email, places orders or changes ERP records with nobody approving it.
- A new ERP first. AI can work with the ERP you run, including an older system with no API, so ask why a replacement has to come first. AI for ERP covers the ways to connect.
- A promise that a grant will pay. Only the program decides who qualifies, and some programs count costs only from the day you apply.
What you should own at the end
When the work ends, your company should hold everything it needs to run and change the system without the consultant. Check each item before the final payment.
| Item | What it covers | How to check |
|---|---|---|
| Source code | Every script, connector and screen | It sits in a repository your company owns, with its full history |
| Prompts and settings | The instructions and configuration the system runs on | They are stored with the code, not in the consultant’s workspace |
| Test set | Past jobs with known answers, used to check every change | Someone on your team can run it and read the results |
| Data and database | The copy of your records the system reads | It sits on a server or account your company controls, and exports in a documented format |
| Accounts and keys | Model, cloud and software accounts, and the keys that connect them | Each one is billed to your company and administered by your staff |
| Documentation | How the system works, how to restart it and how to change it | Someone on your team has followed it once |
Code written by an outside firm or a freelancer does not become yours without a written assignment. Under section 13 of the Copyright Act, the author is the first owner of copyright. An assignment is valid only in writing, signed by the owner. Put the assignment in the contract, along with a licence to keep using and changing anything the firm reuses across clients. AI consulting covers the legal side in more detail.
Data residency: where your data is stored and processed
Data residency has two parts. Storage is where your records sit at rest. Processing is where the model reads a request and writes the answer. A consultant should answer both for every tool in the design.
The two often differ. Here are three examples from the providers’ own pages, checked on October 5, 2026:
- Anthropic stores data in the US and, by default, may route requests to the US, Europe, Asia and Australia, according to its server location article.
- OpenAI’s API has a Canada region that stores data at rest and does not process it there, according to its API data guide.
- Amazon Bedrock, used from AWS’s Canada (Central) region with cross-region inference, keeps data at rest in Canada. Inference may run in another region, according to an AWS post.
Your legal responsibility stays with you either way. Clause 4.1.3 of PIPEDA makes an organization responsible for personal information it transfers to a third party for processing. The organization must protect that information by contract or other means. Ask the consultant to name the storage location, the processing location, the model and its retention terms in the contract.
An open-weight model is one whose maker publishes the files, so you can run it yourself. On a server you control in Canada, it keeps both storage and processing in the country. To compare that option with hosted plans, see private AI for business. For which laws can reach your data, see data sovereignty.
Start with a fixed-price first stage
A fixed price for a defined first stage puts the risk of overruns on the firm, within the agreed scope. It also gives you a working result to judge before you commit to more. A good first stage has five parts, written into the statement of work, the document that lists the tasks, the deliverables and the tests:
- One job, such as entering customer orders at a distributor or pricing quote requests at a manufacturer, with today’s time or cost written down.
- Your own records, read-only, from the systems the job runs on.
- A working tool that your team uses on real work, with a person approving its output.
- A test of past jobs with known answers, and the score the tool has to reach.
- A decision point at the end, where you keep what was built and decide whether to go further.
If a proposal does not say what will run on your records when the first stage ends, ask the firm to add it.
Canadian programs that pay for outside help
Several public programs pay part of the cost of outside help with AI or digital projects. Each sets its own rules on who qualifies and which costs count, and some count costs only from the day you apply. Canada’s national AI strategy, AI for All, commits $500 million to expand the Regional Artificial Intelligence Initiative, according to Innovation, Science and Economic Development Canada. The status of each program below comes from its official page on October 5, 2026.
| Program | What it pays for | Status on October 5, 2026 |
|---|---|---|
| Regional Artificial Intelligence Initiative in Quebec, from Canada Economic Development for Quebec Regions | Interest-free repayable support of up to 50% of authorized costs, for SMEs ready to adopt AI. Costs count from the date you apply. | Accepting applications. The initiative runs until March 31, 2031. |
| Regional Artificial Intelligence Initiative in southern Ontario, from FedDev Ontario | AI adoption projects that go beyond buying off-the-shelf tools. Contributions to SMEs are normally repayable. | Not accepting applications. FedDev Ontario says it will share details of the next intake. |
| ESSOR, volet 1, from Investissement Québec for the Government of Quebec | A non-repayable contribution of up to 50% of eligible costs for a digital diagnostic and a digital plan, which can include AI. The plan’s implementation is covered too. External consultants’ fees count. The diagnostic and plan streams are for businesses with 250 employees or fewer and revenue of at least $2.5 million. | Open. The program ends on March 31, 2027. |
| Advanced Manufacturing and Innovation Competitiveness stream, from the Government of Ontario | Loans of up to 15% of eligible costs, partly forgivable, for advanced manufacturers investing at least $500,000 in a project. Applicants need at least three years of operations. Third-party engineering and software development count as eligible costs. | Open for applications from June 30 to November 5, 2026. |
| LIFT, from BDC | A required plan from BDC Advisory Services, then financing to carry it out, including integration and implementation costs. Rates are preferential when Canadian suppliers or integrators do the work. For businesses with at least $1 million in revenue. | Taking requests online. |
| Industrial Research Assistance Program, from the National Research Council | Advice and, in some cases, funding for research and development of innovative, technology-driven products and services, for firms of up to 500 employees. It does not fund day-to-day operating costs. | Ongoing. A senior executive of the business calls to start. |
| Canada Digital Adoption Program, Boost Your Business Technology stream | Grants for a digital adoption plan, with loans to carry it out | Closed to new applications since February 19, 2024, according to BDC. |
The research program’s AI Assist stream helps SMEs build generative AI into their core products and services. It suits a manufacturer building AI into a product. Training has its own programs, which AI training covers for Ontario and Quebec. Apply before you sign the work order, and treat any firm that promises approval with caution, because only the program decides.
How ThriveAI helps
ThriveAI is an AI engineering company in Ottawa. It builds private AI systems on a company’s own data, for businesses that make, move or sell physical goods. It works hands on with your team. Derik Lawlis, the founder, leads every project and stays close to the build. Working sessions run on site with your team, in French or English.
In stage one, ThriveAI builds a working tool for the one job that costs you the most, at a fixed price, in weeks. It runs on your own systems, and if you stop there, you keep it. If you stop working with ThriveAI, you keep your data, and everything the system recorded exports in full in a documented format. The platform is designed to keep each client’s data on its own server in Canada, and a hosted model may process requests outside Canada. ThriveAI holds no SOC 2 or ISO 27001 certification.