Published August 31, 2026 · Altvina Insights · 6 min read

The AI Waste Audit a Small Firm Can Run in an Afternoon

A one page worksheet for tallying what your AI tools cost, how often anyone actually uses them, and what the same work would cost by hand. Your own numbers only, no industry benchmark, about an afternoon's work.

Print last month's card statement and highlight every AI subscription on it. Most people doing this for the first time lose the thread partway down the page, because a couple of the charges bill under a parent company name nobody recognises. Not one client's story, just the ordinary shape of a first pass at this.

The statement is the least interesting number in the exercise. What matters more is what each tool was bought to fix, and whether that gap is still open.

Paid AI is narrower than the noise, and it stacks

Two figures worth holding side by side before you tally your own.

Intuit QuickBooks' AI Impact Report reads anonymised payment records rather than asking owners what they use. On that basis, only 12% of US businesses with observed payment records paid for dedicated AI tools between 2021 and 2025, far below survey-reported usage rates. Plenty of usage happens on free tiers, personal logins and browser tabs that never touched the company card.

The JPMorganChase Institute looked at payments moving through business banking accounts instead of at what owners report. It found the share of small businesses paying for three or more distinct AI services climbed to 9% in 2025, up from a negligible share just a few years earlier.

So a minority pay for any of this, and inside that minority, stacking three or more services now shows up in banking data. Firms in that group sit ahead of most of their market.

Being ahead is also the exact point where nobody has gone back to check whether the first purchase still fits.

The worksheet: six columns, one page

Give every tool its own row. Include the ones bought for a single project, the ones bundled into a bigger platform bill, and the one somebody expensed personally in March.

Tool and monthly price. Annual plans divided by twelve.
What it was bought to fix. One sentence describing the actual problem, not the product category.
Who uses it. A person's name, not a department.
Uses in the last thirty days. From an admin log if there is one, from asking if there is not. Write "unknown" where that is the honest answer.
Which task it replaced, and the hours that task took by hand each month. Estimate it, but estimate it out loud with the person who used to do it.
Loaded hourly cost of whoever did it by hand. Salary plus employment costs, divided by working hours. A rough figure is fine as long as the same one runs all the way down.

Column two stalls most of these audits for a while. If nobody can say what a tool was bought to fix, you have a finding before any arithmetic.

Three sums

Sum one: cost per use

Monthly price divided by uses in the last thirty days. Write it beside the row.

Some rows come out at pennies. Others come out higher than the task itself is worth, which is the whole reason to do the division instead of eyeballing the invoice. A modest monthly price used four times a month is not cheap.

Sum two: the hand-time comparison

Hours by hand per month, multiplied by the loaded hourly cost. That figure is what the manual version costs you. Put it next to the monthly price and subtract.

A positive number means the tool buys back more than it charges, on your figures. A negative one means the opposite, on your figures. Nobody else's numbers are involved.

Sum three: the unknown column, annualised

Add up the monthly price of every row where nobody could name a regular user or an original problem, then multiply by twelve.

That total is the cleanest number on the page. No interpretation, no debate, and the one to carry into the next budget conversation.

No benchmark, deliberately

You will want to know whether what you pay is normal. Skip the question. No credible average tells a small professional services firm whether its transcription spend is sensible, and a comparison against a number you cannot verify mostly gives you permission to leave things alone.

Your two columns are the comparison. Cost per use, and cost against the hand version. Enough to decide with.

Reading your own answer

Four outcomes tend to fall out of the page.

Cheap per use, real user, clear original problem. Leave it alone and stop thinking about it.

Expensive per use, but the use matters. Sometimes those four uses a month are the four that keep a client. Keep it, and write down why, so next year's version of this audit is shorter.

Nobody uses it. The easy cancel. Do it the same afternoon, while the sum is still in front of you.

Heavily used, but the gap it was bought for has closed. The interesting row, and the one that never shows up on a spend report. The tool works fine. Nobody re-cut the work around it, so people are doing a careful job of something the firm no longer needs done that way.

What the audit usually surfaces

Tools outliving the gap they were bought to close is a pattern that shows up often in fractional operations work. Somebody buys in a bad month, the bad month passes, the subscription renews quietly, and the habit built around it hardens into how the work gets done.

Offered as our judgment rather than as a finding: the waste is rarely the subscription. The waste is the task nobody re-drew after the tool arrived. Cancelling a line item takes two minutes. Noticing that three people still assemble a report by hand for a meeting that stopped mattering months ago takes an afternoon with a worksheet.

Run the six columns. Do the three sums. A couple of dead subscriptions falling out is a good afternoon, and you are done.

Checking which of the two you have is free. The Routing Hub diagnostic separates a workflow problem from a hiring problem from a knowledge-transfer problem, in a browser, in a few minutes. Some people read the answer and go and fix it without us, which is rather the point of leaving it there.

Should the tools turn out fine while the work around them is the problem, that question is bigger than a worksheet answers, and it is the one we spend our time on through the Altvina Blueprint, a paid fixed scope diagnostic that ends in a Bottleneck Diagnosis and a Recommended Path Forward with every option on it, including doing nothing. Either way, the numbers are yours before anyone else sees them.

Content and Accuracy Disclaimer

This article was drafted with AI assistance and reviewed by the Altvina team. We rigorously fact-check all content to ensure reliability.

Should you notice any inaccuracies or outdated information, please contact us so we can correct it. Your feedback helps us maintain high standards of accuracy and transparency.