Almost half of Canadian business leaders are running AI experiments that have not produced a return, and the licences are not the reason.
- BDO Canada found 46% of Canadian business leaders experimenting with AI without meaningful ROI, against 18% embedding it into workflows.
- The Bank of Canada puts significant AI use in core operations at 8% of firms.
- Return shows up when a process is redesigned, not when a tool is added to an unchanged process.
AI ROI goes missing in the same place almost every time. Somebody buys 40 licences, usage looks healthy, and at the end of two quarters no line in the budget has moved. BDO Canada put a number on it in June 2026: 46% of Canadian business leaders are experimenting with AI without meaningful return, while 18% are actively embedding it into workflows and operations. The licences were never the problem. The work around them stayed exactly as it was.
TL;DR
- BDO Canada surveyed 520 business leaders from 31 March to 3 April 2026 and found 46% experimenting without meaningful ROI.
- 18% are embedding AI into workflows and operations, and 27% expect minimal impact over the next four years.
- Bank of Canada research puts significant AI use in core operations at 8% of firms, with another 50% at low or moderate use.
- Pilots fail to produce return when the process, the handoffs and the headcount plan stay unchanged.
- Measure a baseline before the tool arrives or you will have no way to prove anything afterwards.
What does the Canadian AI ROI data actually say
The BDO Canada AI Vision Report, published 25 June 2026, surveyed 520 Canadian business leaders through the Angus Reid Forum between 31 March and 3 April 2026, with a margin of error of 4.3 points. The headline split is 46% experimenting without meaningful ROI against 18% embedding AI into workflows. Another 27% think AI will have minimal impact on their organisation over the next four years.
Bill Syrros, BDO Canada national AI leader, framed the next divide as the one between organisations redesigning work around AI and those funding disconnected pilots.
The Bank of Canada found the same shape from a different angle in August 2026. 8% of firms use AI to a significant degree in core operations. Half use it to a low or moderate degree. Read those two studies together and the picture is consistent: plenty of use, very little redesign.
Why AI ROI disappears inside a normal company
Picture the usual sequence. Marketing gets a copilot. Drafting time drops by a third. The team keeps the same deadlines, same approval chain, same number of people, and fills the saved hours with more drafts nobody asked for.
Nothing was measured before, so nothing can be compared after. The saving is real and it is invisible.
Return needs one of three things to happen to the saved time. It gets sold, so the same team handles more volume. It gets removed, so a role or a contract ends. Or it gets moved, so people do work that was previously deferred and that has a name and a value. Time that is merely freed has no financial existence. This is the same pattern we covered in the hidden time tax, viewed from the finance side.
What separates the 18% from the 46%
| Funding a pilot | Redesigning the work |
|---|---|
| Tool chosen first, use case found later | One process chosen first, tool chosen to fit it |
| Success measured in logins and sentiment | Success measured against a baseline recorded before rollout |
| Owner is whoever championed the tool | Owner is whoever carries the budget line it should move |
| Ends when the trial ends | Ends when the old process is formally retired |
The fourth row is where most Canadian mid-market firms stop. Running the old process and the AI process side by side forever is not a pilot. It is two processes.
What should leaders do next
- Pick one process with a number attached to it. Quotes issued per week, invoices coded per day, tickets closed per agent. If it has no number, pick a different process.
- Record the baseline for four weeks before the tool arrives. Volume, cycle time, error rate, cost per unit. This is the single step the 46% skip.
- Name the budget line the pilot is supposed to move, and name the person who owns that line. Not the person who likes the tool.
- Decide in advance what the saved hours get used for, and write it down. Sold, removed or moved. Pick one.
- Set a kill date. If the number has not moved in 90 days, stop paying and say why in writing so the next attempt starts smarter.
- Retire the old process on a date. Parallel running is where return goes to die.
The case for staying in the 46%
An experienced CFO could read all of this and keep funding experiments anyway, with reasons.
Capability takes time to build, and the first year of any technology looks like waste from a spreadsheet. Teams that learned to use AI badly in 2025 are the teams using it well in 2027, and none of that shows up as ROI in quarter two. There is also a defensive argument. If your competitor learns this a year earlier, the cost of catching up is not measured in licences.
That case works right up to the moment nobody is measuring. Learning without a baseline is not learning, it is spending.
What to watch over the next 90 days
- Statistics Canada business conditions data through the fall, where a rising share of firms reporting AI in core operations would show redesign rather than trial.
- Your own renewal dates. Most first-wave AI licences bought in 2025 come up for renewal this fiscal year, which is the natural moment to demand a number.
- Vendor pricing moves toward outcome based contracts, which shift measurement risk to the supplier.
FAQ
How long should an AI pilot run before it proves ROI?
Ninety days against a recorded baseline is enough for a process with weekly volume. Anything longer without a moving number is a subscription, not a pilot.
What is a realistic first AI ROI target for a mid-market Canadian firm?
One process, one measurable improvement in cycle time or cost per unit, and a written decision about the freed hours. Chasing a company-wide productivity figure first is how the 46% got there.
Does AI ROI have to mean cutting staff?
No. Selling the capacity counts, and so does moving people to deferred work with a known value. It only has to stop being invisible.
We never took a baseline. What now?
Take one on the current AI-assisted process, then rebuild the pre-AI comparison from whatever records exist, invoices, timesheets, ticket logs. An imperfect comparison beats none.
Closing analysis
Pick one process this week, write down its four numbers, and put a date on the wall for when the old version of it stops. That single act moves a company out of the 46% faster than any tool on the market.
Sources
- BDO Canada, AI Vision Report, past the pilot to the agentic future of work, 25 June 2026. bdo.ca
- Bank of Canada, Canadian businesses use of AI, what the evidence shows, August 2026. bankofcanada.ca
- Statistics Canada, Canadian Survey on Business Conditions, third quarter 2026, 31 August 2026. statcan.gc.ca
Related reading
- AI Was Supposed to Save Us Time. Why Are We Working Longer?
- AI Implementation in Canada Needs Repeatable Workflows
- AI Tool Fatigue and the Business AI Switchboard
Disclosure
The author has no relevant financial, advisory, or board relationships with any party named in this column.
Zak Hussein writes on AI for Canadian business owners and operators. He is the founder of AI Magazine Canada and CEO of ORKA AI.
This piece sits inside our reference on AI for business in Canada, which we keep current.
