AI was sold to Australian businesses as a way to spend less on marketing. Canva just cut its 2026 revenue growth forecast because the cost of serving AI ran too high, and if the maths does not work at that scale it is unlikely working for the small business bolting three AI tools on to save money. The real question is not whether AI is cheap. It is whether you can measure your cost per outcome well enough to know.
If a US$42 billion company relying on the best models in the world had to slow down to make the maths work, the maths is not quietly working for the cafe running three AI ad tools either.
The Take: AI was pitched to Australian businesses as a way to spend less. In marketing it is doing the opposite. The cost of running AI is a floor under your spend, not a discount on it. That floor is rising for every business that bolted a tool on to save money. The businesses that will actually know whether it pays are the ones who can already measure their cost per outcome. Most cannot, which means most are guessing.
Is AI actually cheaper to run?
Canva, valued at about US$42 billion, cut its 2026 revenue growth forecast from roughly 30% to 20% because the cost of serving each AI task ran higher than expected. That is the tell.
Canva downgraded its CY2026 revenue growth forecast by a third, from about 30% to about 20%, after the cost of rolling out its AI suite came in higher than planned (Startup Daily). Co-founder Melanie Perkins told shareholders that "the average cost of serving an AI task was too high" and that the company had been "relying too heavily on frontier models" (Startup Daily). Revenue still climbed 25% year on year to US$921.9 million in the June quarter, strong growth that nonetheless fell short of the company's own guidance (B&T).
Sit with what that means. The company that builds AI design tools, at global scale, with in-house models and the best engineers money can buy, had to slow down because the unit economics of AI were not working. If they could not make it cheap, the odds that your three bolted-on tools are cheap are slim.
The cost floor that never hit the invoice
Old software had a shape you could budget. You paid a seat, you paid a licence, the cost sat still while you used it more. AI does not behave like that. Every task has a marginal cost, because every generation, every image, every rewrite calls a model that charges by the token. The more your team uses it, the more it costs. Usage is the bill.
This is not a cost cut. It is a cost floor, and the floor rises with adoption. The saving people imagined was real for about a quarter, back when volumes were low and vendors were subsidising the compute to win you. Then volumes climbed. The subsidies thinned. The floor started lifting under all of them.
Canva cut its 2026 revenue growth forecast by a third after the cost of serving AI ran higher than planned
What does this mean for a business bolting AI onto its marketing?
Australian business adoption of AI accelerated through 2024 to 2025, according to the ABS (ABS). Across the Australian businesses we score, the pattern is consistent: most of the ones that have bolted AI tools onto their marketing still cannot tell you their cost per acquisition, let alone whether the tools moved it.
That is the real problem. It was never whether AI is cheap. It is whether you can see your cost per outcome clearly enough to know if AI is paying for itself. A cafe adds an AI ad tool, a subscription writer, an AI image generator, three logins that each charge by usage. The subscriptions show up on the card. The outputs do not show up anywhere you can measure. So the spend is visible and the return is invisible, which is the exact opposite of what a saving looks like.
The subscriptions are visible. The return is not. That is not a saving. It is a leak with a nice interface.
What I'd do about it
Start with the number, not the tool. Before you add or renew a single AI product, write down your cost per acquisition or cost per lead as it stands today. If you cannot, that is the first thing to fix, because you are about to spend more without a way to tell if it worked.
Then price the floor, not the sticker. Add up every AI subscription plus the usage-based charges that scale with your team, not just the headline monthly fee. That is your real AI marketing cost. Most owners have never totalled it.
Then run the honest test. For each tool, ask what outcome it is meant to move and whether you can see that outcome move. If the answer is no on both, you are not saving money, you are buying comfort. Cut it or instrument it.
How we score: we grade Australian businesses across six marketing dimensions, and one of them is how well a business measures its own performance. Where our sample in a given niche is thin we say so and keep the finding directional rather than dressing it up as precise. The full methodology is at how we score.
Where I land
My opinion, stated as opinion: AI will make good marketing teams faster and make unmeasured marketing more expensive, and right now far more Australian businesses sit in the second group than the first. The tool is not the decision. The measurement is. Canva slowed down to get the economics right before scaling. A business with a fraction of the resources should be at least that disciplined before it adds the next login.
If you want to see where your own cost per outcome actually sits before you buy another AI tool, that is what Lens is for.
Frequently asked questions
Does AI reduce marketing costs?
Not reliably. AI carries a usage-based cost that rises as your team uses it more, so it behaves like a floor under your spend rather than a discount. Even Canva cut its 2026 growth forecast over AI serving costs. Whether it saves you money depends entirely on whether you can measure your cost per outcome.
Why did Canva cut its revenue forecast?
Canva lowered its CY2026 revenue growth forecast from about 30% to about 20% because the average cost of serving an AI task ran higher than expected and it had been relying too heavily on expensive frontier models. It chose to slow its AI rollout to rebuild the economics.
How do I know if AI is worth it for my marketing?
Measure your cost per acquisition or cost per lead before and after you add the tool. Total your real AI cost, including usage charges, not just subscriptions. If the outcome does not move or you cannot see it move, the tool is not paying for itself.
Is spending on AI marketing tools a mistake?
Not on its own. The mistake is adding usage-based AI spend without a way to measure the return. Used against a clear cost-per-outcome baseline, AI can be worth it. Used blind, it quietly raises your costs while feeling like progress.