Pull up your AI spend for last quarter. You can probably find it in about thirty seconds — a line item, a per-seat number, a total you feel reasonably good about.

Now find the rest of it.

The subscriptions on personal cards, coming back through expense reports as software, or professional development, or nothing at all. The department that signed up for its own tool because waiting wasn’t an option. The four people paying twenty dollars a month each for four different models, all of which do roughly the same thing, none of which the company owns.

That’s the second bill. It doesn’t arrive. It doesn’t get approved. And it is not the expensive part.

The expensive part is what doesn’t compound

Here’s the thing that should bother a finance leader more than the duplicate subscriptions.

Workplace research tracked by the U.S. Chamber’s small business AI guide shows unchecked AI tools carry high failure rates — and Qualtrics research puts a number on it: customer-facing AI applications fail at roughly four times the rate of automated back-office tasks. Read that again. The tools people reach for on their own, without a policy or a review step behind them, aren’t just unmanaged. They’re failing at a materially higher rate — quietly, one desk at a time, with nobody in a position to notice.

Four people using four different AI tools are not four times as productive. They’re four silos. Someone in operations figures out how to cut a recurring three-hour task down to forty minutes — and that knowledge stays in their head, in their tool, on their account. Someone in finance solves an almost identical problem six weeks later, from zero, in a different product. Nothing accumulates. Nothing transfers. Nothing gets better across the company because it got better for one person.

You are paying for individual productivity and receiving exactly that: individual productivity, unrepeatable, unmeasurable, and walking out the door if that person leaves.

That’s before the parts that don’t show up as spend at all. Company data sitting in tools you don’t control. No volume leverage on pricing, because you’re buying in ones instead of in blocks. Per-seat rates a fraction of what you’d negotiate as an organization. And no line of sight — none — from any of it back to a recovered hour or a recovered dollar.

This is what makes AI sprawl uniquely bad as a category of spend. Most uncontrolled spend is merely wasteful. This kind is wasteful and it’s compounding against you, because the longer it runs, the more the value — and the risk — gets locked into places you can’t reach.

The five numbers you can’t produce

Try this. Right now, without asking anyone, answer these five:

How many AI models are running in your business? Not licensed. Running. Including the ones on personal accounts.

What percentage of your workforce is using AI? Weekly, not ever.

What percentage of that usage is governed — sanctioned tool, under your control, with a policy behind it?

What percentage is charged back to a department, a cost center, a budget owner?

What percentage is tied to a stated business outcome rather than an individual’s preference?

If you’re like most SMBs, you can’t answer any of the five with confidence. Some leaders can’t answer the first one within a factor of two.

Sit with that for a second. This is a category of spend and risk that is running inside your business — touching client data, absorbing budget, shaping how work gets done — and there is no number attached to any of it. You would not accept this in any other line of the business. You’d have found it in a review a long time ago.

The reason it hasn’t been found is that nobody owns it. It didn’t come in through procurement. It came in through people trying to do their jobs. That’s not a discipline problem on your team. It’s a missing map — and a missing map isn’t anyone’s fault until it stays missing.

You can’t fix what you can’t see

Notice that none of this argues for spending more. It argues for seeing what’s already being spent.

That’s the good news buried here, and it’s the reason this is worth an hour of your time rather than a project plan. Most companies that finally look at these five numbers discover two things at once: they’re spending more than they thought, and they’re capturing less of it than they thought. Which means the first move isn’t a purchase. It’s consolidation — one sanctioned path, volume pricing, gains that stack, and a policy that gives people permission to use the thing you’ve actually paid for.

Companies routinely find they were already paying for AI capability inside licenses they hold today, while separately reimbursing staff for personal subscriptions to do the same work. That’s not a technology problem. That’s a visibility problem, and visibility is cheap.

The hours are there — most SMBs are sitting on four to eight hours per knowledge worker, per week, in repetitive admin and information search. The spend is there too. Right now, neither one is on a report you’ve seen.

We’ve spent 27 years helping Tucson businesses find the gap between what they’re paying for and what they’re actually getting — usually in servers, backups, and networks. The same blind spot is showing up in AI, and it responds to the same fix: someone objectively looks, and hands you the numbers.

Come find your five numbers

We’re running a working session on exactly this: what shadow AI is costing SMBs, how to find the five numbers inside your own business, and what to do with them once you have them.

Not a product demo. Not a keynote about transformation. A method for producing the numbers, and a walkthrough of what most companies find when they run it.

Webinar: The Shadow AI Audit — What’s Really Running in Your Business

45 minutes. You’ll leave with the five-number framework, the questions to ask, and a clear read on the cost and exposure sitting inside your company right now.

Built for finance and operations leaders at companies between 25 and 200 people.

Can’t make it live? Register anyway and we’ll send you the recording and the audit worksheet.

Questions before then? Reach our team at (520) 877-3033 or sales@integratedaxis.com — we’re at 6147 E Grant Rd in Tucson, and we’re happy to talk through what your five numbers might look like before you ever sit through a slide.