The real cost of shadow AI on your team's budget
8 minutes read
Shadow AI is the pile of personal AI subscriptions your team expenses one card at a time. Here is a worked example of what it really costs, and how to bring it into a single budget.
Your tracked AI spend is one number. What your team actually spends on AI is almost always a bigger one. The gap between them is shadow AI, and most finance leads can feel it long before they can prove it.
Closing that gap starts with sizing it. Here is what shadow AI is, why it stays invisible, and what it tends to cost a growing team.
What counts as shadow AI
Shadow AI is any AI tool your team uses and pays for outside your central budget. A personal chatbot account on someone's own card, a writing assistant bought one seat at a time, an AI feature bundled into a design tool a freelancer expensed: if it is not on a list someone owns, it is shadow AI.
The word borrows from shadow IT, the older problem of staff running software the ops team never approved. Shadow AI moves faster. Signing up takes a card and two minutes, the output is good enough that people keep using it, and nobody thinks to ask permission for something that costs less than lunch. By the time anyone reviews it, the tool is already part of how the work gets done.
It helps to be clear about what this is not. An approved chatbot plan that finance pays for and tracks is not shadow AI, even if half the company uses it. The problem is not that people use AI. The problem is that the spend and the data behind it never pass through anyone who is accountable for either.
Why shadow AI stays invisible
Shadow AI hides because every single charge is small enough to clear an expense report without a second glance. A €20 subscription does not trigger a review. Forty of them, spread across four teams and coded to a dozen different expense categories, never add up in one place, so nobody sees the total. The tools that hide best are the ones that look like something you already pay for.
| Where it hides | How it clears | What it looks like on the statement |
|---|---|---|
| Personal cards | Reimbursed as an expense, not booked as a software cost | A €20 line under "subscriptions" or "other" |
| Unlisted subscriptions | Bought per seat, never added to the central tool register | Nothing, until the annual renewal lands |
| Bundled AI features | Paid for inside a tool you already approved | The old tool's name, at a slightly higher price |
The pattern is the same one that drives AI tool sprawl: decisions made one person at a time, each defensible on its own, never reviewed as a whole. Shadow AI is the spend side of that sprawl.
What shadow AI actually costs
The real cost is usually several times the tracked cost, because the tracked cost only counts the tools you bought centrally. To make that concrete, here is a worked example for a 40-person agency. The figures are illustrative, chosen to show the mechanism rather than to report a survey.
| Tool | People | Monthly each | Monthly total |
|---|---|---|---|
| Personal chatbot accounts | 18 | €20 | €360 |
| Premium writing assistant | 6 | €30 | €180 |
| AI design tool | 4 | €40 | €160 |
| Coding assistant | 3 | €20 | €60 |
| Misc annual tools, monthly | - | - | €100 |
That is €860 a month, or roughly €10,000 a year, none of it in the AI budget. If the same agency thinks it spends €2,000 a year on a single approved tool, its real AI bill is six times what the finance lead reports to the board. The reported figure was not wrong so much as partial. It counted the one tool that went through procurement and missed every tool that did not, which is most of them.
Change the inputs and the headline holds. The number of people carrying a personal subscription matters more than the price of any one tool, because the cost scales with how many quiet decisions went unreviewed, not with the sticker price. A team that has never audited this has no reason to expect the total to be small.
The costs that never reach the invoice
Money is the easiest part of shadow AI to measure, and not the most expensive.
The first hidden cost is data exposure. A personal AI account runs on that tool's consumer terms, not on any agreement your organisation signed and can point to. Client work and internal documents pasted into an unapproved account may be retained or used to train the underlying model, and because the account sits outside your systems, you have no record that any of it happened. If a client later asks where their data went, "one of our people had a personal subscription" is not an answer you want to give. For a team doing client-facing or regulated work, that is a compliance gap you cannot audit, for the plain reason that you cannot see it in the first place.
The second is duplication. When five people each buy their own AI writing tool, you pay five retail prices for something a single team plan would cover at a lower rate per seat. Every one of those purchases also skips the volume discount you would have negotiated by buying once, so shadow AI quietly costs more per seat than the same tools would on a central plan.
How to surface the real number
Start with the expense data you already have, not a new tool. Most of shadow AI is already sitting in your accounting system, filed under the wrong labels. You are not trying to catch anyone out. You are trying to see a total that no one has ever added up, because until now it was never anyone's job to. A focused review, done in an afternoon, turns that scattered spend into a number you can act on.
- Pull three months of card and expense data and search for the AI tools by name. Filter on the vendor names, not the category, because the category is where they got mislabelled in the first place.
- Ask each team lead to list the AI tools their people actually use. Frame it as a budgeting exercise, not an audit. You want the honest list, and you only get that if nobody expects to be in trouble for answering.
- Add the two together and remove duplicates. The total is your real AI spend. The gap between it and your tracked figure is what shadow AI was costing you.
- Sort the list into what to keep and what to consolidate. Some tools earn their place and should move onto the central budget under a named owner. Others are three people paying separately for the same thing, and moving them onto one team plan usually cuts the per-seat price while giving you a single line to track. Anything nobody can justify, you drop.
Do this once and you have the real number. The harder part is making sure it stays real next month, because a one-off audit ages the moment someone expenses the next new tool. That is a budgeting problem, and it is the same discipline covered in how to budget AI usage across teams: give the spend an owner, a limit, and a place it has to run through.
Where Hebno fits
A quarterly audit surfaces shadow AI after the fact. A single point of access prevents it from forming in the first place.
Hebno puts every major AI model behind one interface, so people reach for the tools they need without buying their own. Spend shows up live, by team and by person, against a hard budget that stops at the limit instead of warning you after it. The subscriptions that used to scatter across personal cards become one bill you can see, attribute, and cap. See how the numbers work on our pricing page.
