Do it yourself or hire an AI firm: what it really costs

Do it yourself if the project is about adopting a tool: giving your teams an assistant, governing its use, building a few simple flows. Hire a firm if the project is about transforming a process: connecting an agent to your ERP, reading invoices or drawings, hosting a model on your premises, handling personal information. In between, the right setup is often mixed: the firm builds and hands over, your teams run it.

An honest comparison does not set “free” against “an invoice.” It sets internal hours, available skills and the risk of failure on one side against a firm’s price minus funding on the other.

An in-house project almost always draws on the same people: the controller who knows the books, the project manager who knows the quotes, the person who is “good with computers.” They are rarely idle. Every hour spent on the project is an hour taken from month-end close, from clients or from production.

To put a number on that cost, list the tasks the project requires and estimate the hours with the people involved:

An AI project that touches business systems needs four skills rarely found together in an SME: knowledge of the business process, software integration (ERP APIs, error handling), applied AI (model choice, instructions, evaluating answer quality) and compliance (Law 25, security, logging).

The first one you have: nobody knows your processes better than your teams. The other three can be learned, but the learning happens on your project, with its trial and error. If you are aiming for sovereign AI hosted on your premises, add administering a server with a graphics processor.

For an adoption project, licences are the main visible cost, and they are published. As of October 23, 2026: ChatGPT Business at CA$27 per standard seat per month, the Microsoft 365 Copilot Business add-on at CA$28.50 per user per month on an annual commitment, Claude Team at US$20 per seat per month billed annually. For flows, Power Automate Premium is CA$20.40 per user per month and Zapier Pro starts at US$19.99 per month billed annually.

These amounts are modest. That is what makes “homemade” attractive for adopting an assistant, and it is a good choice in that case. Our comparisons ChatGPT, Copilot or Claude and Power Automate, Zapier or an AI agent go through the offers.

In-house, the main risk is running out of steam: the project moves on Friday afternoons, then stalls at the first busy period. Next come architecture mistakes that are hard to undo (a tool chosen before the need was measured, sensitive data entered into a foreign cloud without an assessment) and dependence on one person who leaves with the know-how.

With a firm, the risk is the opposite: a solution your teams do not understand, code nobody else can maintain, a contract that ties you down. It shrinks with three requirements: a measured deliverable from the diagnostic on, ownership of what is built for you, and knowledge transfer written into the engagement. Our list of 12 questions to ask an AI consulting firm is built for that.

According to Statistics Canada, concerns about cybersecurity or privacy (13.4% of businesses) and cost (10.6%) are among the barriers Canadian businesses name. Both are better managed with a written framework, whether in-house or with a firm.

This is the point SMEs most often forget: several programs reduce the net cost of a project, and they should be compared before choosing between in-house and a firm. The rates below are those presented on our Funding page. Eligibility is confirmed case by case.

In the projects we see, the split of roles that works best is simple. The business keeps what it does better than anyone: defining the need, supplying real cases, approving results, adopting the tool. The firm takes what requires scarce skills: architecture, ERP integration, hosting, compliance. At the end, the knowledge moves in-house.

That is the path Les Industries Romy in Granby followed: Cogio built the agents on the company’s processes, and a person at Romy approves the steps that commit the business. According to management’s estimate, they now free up 152 hours a week. Details are in the Romy case study.

Bring together the two or three people who would carry the project. Fill in the task list above with their hour estimates, multiply by their real hourly cost and add what those hours usually produce. Set a firm’s price minus eligible funding against it. If the gap is narrow and the skills are missing, risk tips the balance. Our article on the cost and timeline of an AI project and our Funding page round out the exercise. For a view on your case, write to us.