Here's a finding worth reading twice.
The 2026 QuickBooks AI Impact Report — conducted with the University of Chicago across 34,000 small businesses — found that 77% of respondents now use AI on at least a regular basis, up sharply from 48% in mid-2024. Among them, 41% report revenue increases they attribute to AI, and 74% say their productivity has improved.
On the surface, that's a strong endorsement. Nearly three quarters of small businesses using AI say it's making them more productive. That's a big number.
Here's what the fine print says.
When researchers asked how businesses were measuring those improvements, the answers went quiet. More than half said their evidence was "a general feeling that their business was better." Less than half were tracking specific metrics. The productivity numbers were based on self-reporting, not time studies. The revenue attributed to AI was based on correlation — things improved while AI was in use — not controlled measurement that could separate AI's contribution from everything else happening at the same time.
What the headline said: AI is working for small businesses.
What the data actually said: most small businesses believe AI is working, and most of them have no way to know for certain.
That's a different story.
The Math Behind the Feeling
To understand why this matters financially, start with the numbers.
The average small business that's genuinely adopted AI is running somewhere between three and eight AI tools. Subscriptions tend to run $30 to $150 per tool per month. Put that together and you get a monthly AI budget somewhere in the range of $200 to $600 — call it $400 as a reasonable midpoint.
That's $4,800 per year. On a feeling.
Gartner estimates global AI spending will reach $2.52 trillion in 2026. A substantial share of that is flowing through small and mid-sized businesses, most of which, by the data above, have no mechanism to verify what they're getting back.
Forrester, projecting into this same landscape, predicts that 55% of AI projects this year will not meet their intended goals. Not because the technology failed. Because success was never defined before the project started. You can't miss a target you never set.
The bill is real. The proof is soft. That combination is worth examining before the next renewal cycle.
Why Nobody Set Up Baselines Before Starting
It's worth pausing here to note something: this isn't a story about carelessness. The failure to measure AI ROI is almost universal, and it's not because small business owners are unsophisticated.
The problem is timing. To measure what AI saves, you need to know what you were spending before AI. That means tracking the baseline — how long tasks took, what your error rate was, what your campaign conversion rates looked like — before you adopted the tools.
Almost nobody did this. Why would they? Nobody told them to. The pitch for AI tools is not "here's a measurement framework to set up before you buy." The pitch is: "sign up, connect your accounts, start saving time." The subscription was easy. The baseline measurement was inconvenient. So it got skipped.
Now, eighteen months in, you have a stack of tools, a stack of invoices, and no clean before-picture to compare against. Proving ROI without a baseline isn't impossible, but it's genuinely hard — hard enough that most people settle for the feeling instead.
The feeling is not useless. If your business seems to be running better, something is probably working. But "seems better" doesn't hold up in a budget conversation. It doesn't tell you which tool is driving the improvement. And it doesn't tell you what to cut if you're looking to reduce spend.
The Five Numbers That Replace the Feeling
The researchers at the Institute of Business AI published a framework that addresses this directly. It doesn't require going back and reconstructing a baseline from scratch. It requires picking three tasks, getting a stopwatch, and starting from now.
Number one: time per task.
Choose three recurring tasks where you use AI. Record, with an actual timer, how long each takes now. Then do the same task without AI and measure again. The difference, multiplied by your hourly rate, is your documented time savings per occurrence. If a task that used to take 90 minutes now takes 20, and you're worth $100 an hour, that's $116.67 of value each time you run it. Write that number down.
The stopwatch part matters. Self-reporting on time is famously unreliable — we consistently underestimate how long things take and overestimate our productivity gains. The only way to get a real number is to measure it.
Number two: output quality.
Track what percentage of AI-generated content requires only minor edits versus a full rewrite. If 80% of your AI output needs minor cleanup, the tool is delivering real value. If 80% needs a full rewrite, you've changed your workflow without saving time — you've just replaced one type of work with another.
This matters because many businesses are getting AI output that's structurally correct but tonally wrong, or on-topic but missing key details. Editing that output to a usable standard takes nearly as long as writing from scratch, which means the productivity gain is close to zero even though the tool feels like it's doing something.
Number three: revenue per AI-supported activity.
Compare revenue from AI-assisted campaigns against revenue from equivalent human-generated campaigns, with matched customer lists and timing. This is the most rigorous of the five, and it's not always possible — but when you can run it, it's the most valuable. It tells you not just whether AI is saving time but whether it's changing outcomes.
If your AI-assisted email campaign and your human-written email campaign produce the same revenue, that's a clear result: AI is saving you time, not generating incremental revenue. That's still worth something. But it's different from generating revenue, and the two shouldn't be counted the same way.
Number four: error rate.
If you're using AI for data-heavy tasks — bookkeeping, data entry, report generation — track how often you're correcting errors now versus how often you corrected errors before. A lower error rate is real, measurable value. A higher error rate means AI has added overhead even as it's handling more volume. An unchanged error rate means AI didn't improve this process, and any subscription cost assigned to it needs a different justification.
Number five: tool cost versus documented value.
Add up what you spend on AI subscriptions each month. Then, using the numbers from the first four measures, estimate the documented value those tools produce: time saved plus any measurable revenue impact. If the documented value exceeds the cost, you have a clear case. If it doesn't — or if you can't complete the calculation because you're missing inputs — that's the answer itself.
The target ratio varies by business, but a 3:1 or 4:1 return on documented value versus cost is a reasonable threshold for a tool to justify its subscription.
What Usually Happens When You Run This
The businesses that have gone through a measurement exercise like this tend to land in the same place: two or three tools produce clear, documented value. The rest produce the feeling.
That's not an argument against AI. The two or three tools with clear returns are often significant — they may represent hours per week and meaningful cost savings or revenue impact. But the remaining tools in the stack are often generating cost without generating proof, and without measurement, you can't tell which is which.
The stack audit most small businesses have been avoiding — the one where you actually look at what each tool is costing against what each tool can document — tends to reveal that the answer is simpler than feared. Keep what you can prove. Let go of what you can only feel.
The Question Hiding in the Data
The 34,000-business QuickBooks finding is not a condemnation of AI. The majority of those businesses are probably right that AI is helping them in some real way. The problem is that "right" and "can document" are not the same thing, and the distance between them matters when you're making budget decisions, when you're trying to figure out what to scale, or when someone asks you whether the investment was worth it.
The businesses best positioned for 2026 are not the ones who adopted AI fastest. They're the ones who know what their AI is actually doing — specifically, measurably, on a spreadsheet.
That knowledge starts with five numbers and a stopwatch.
It takes about an afternoon. It's worth doing before the next billing cycle.
The Useful Daily covers practical AI for small business owners. If this was useful, share it with someone who's been meaning to run the audit.