Three in four small and midsize businesses now use AI regularly, according to a piece published this week by CPA Practice Advisor. On its own, that sounds like good news, and mostly it is. But buried in the same piece is a line worth sitting with: adding a faster tool on top of a messy process just gets you a faster tool working on top of a messy process.

It's a plain description of something most business owners already sense but rarely say out loud. The tool usually isn't the problem. What's underneath it is.

The same piece cites Gartner data that makes the shape of the problem clearer. 85% of functional leaders plan to increase their AI spend in 2026. Nearly a quarter of them, 23%, don't know what return their current AI investment is generating.

Sit with both numbers together. Most leaders are about to spend more. Roughly one in four of them can't say what the last round of spending actually did.

That isn't a caution against AI. It's a caution against buying it the way most businesses buy most software: because a competitor has it, because a salesperson made a confident case, or because doing something felt safer than doing nothing.

The question that changes the outcome

In almost every Discovery Session I run, someone arrives with a shortlist of tools before we've talked about what's actually broken. That's not a criticism. It's the natural order most people reach for: see the shiny capability, imagine the business with it, go and get it.

The better order runs the other way. Before any tool gets bought, there's one question worth answering properly: what, specifically, should get faster, cheaper or easier once this is in place, and how will you know if it worked?

If the honest answer is vague, "efficiency," "staying competitive," "everyone else is doing it", that's not yet a reason to buy. It's a reason to look harder at the process the tool would sit on top of first.

What a messy process actually costs

This matters more with AI than with most software, because AI tools tend to amplify whatever they're layered onto. A well-defined process gets genuinely faster. A vague or inconsistent one gets automated at speed, producing more of the same inconsistency, just quicker and with more confidence attached to it.

That's the trap in the CPA Practice Advisor piece: businesses buying AI fast enough that governance, data readiness and process clarity haven't caught up. Not because anyone was careless, but because the tool is the visible, exciting part, and the process underneath it is invisible until something goes wrong.

A better order of operations

None of this is an argument for waiting. It's an argument for a different sequence. Know what you're trying to make faster, cheaper or easier before you shop for what does it. Ask what "working" would actually look like in three months, specifically enough that you could tell if it hadn't. Then buy.

For a small business, this is a cheap discipline. It costs a couple of hours of honest thinking, not a consultant's fee or a delayed decision. It's also exactly the order the AI Accelerator Diagnostic works in: find where the friction actually sits across the business first, then match tools and training to that, rather than the other way round.

The 23% of leaders who can't say what their AI spend achieved didn't get there by using AI badly. They got there by buying it before answering the question that would have told them whether to.

Go wisely.