AI tool sprawl: how 30 people get 40 subscriptions
7 minutes read
A 30-person team rarely decides to buy 40 AI tools. It happens one personal signup at a time. Here's how sprawl builds, what it costs, and how to inventory it.
Nobody sits down and decides to buy 40 AI tools for a 30-person team. It never happens as a decision. It happens as 40 small ones, spread across a year, each of them reasonable on its own. Someone expenses ChatGPT Plus. A designer tries a new image tool. Two people on the same team sign up for the same writing assistant a month apart without knowing. By the time anyone counts, there are more subscriptions than people, and no one can say what half of them are for.
That is AI tool sprawl. It is not a discipline problem or a sign of a careless team. It is the default outcome of a fast-moving category meeting a normal expense process, and it costs more than the line items suggest.
How sprawl happens
Sprawl builds one signup at a time, and every step is individually sensible.
A tool solves a real problem for one person, so they sign up with a personal card and expense it. It works, so they keep it. A teammate hears it's good and signs up separately, because there's no shared account to join. A new model launches, someone wants to try it, and trying it means another subscription rather than switching an existing one. Nobody is doing anything wrong. There's just no single place where "we already pay for something that does this" is a question anyone can answer.
The result compounds because AI tools overlap heavily. Three different subscriptions might all wrap the same underlying model. Two might do nearly the same job. But from inside any one person's workflow, they look distinct, and the duplication is invisible until someone pulls every AI charge into one view.
What sprawl costs
The subscription total is the smallest part of the cost. Sprawl is expensive across three axes at once.
| Cost | What it looks like | Why it's easy to miss |
|---|---|---|
| Direct spend | Duplicate and overlapping subscriptions, unused seats, forgotten renewals | Scattered across personal cards and expense reports, never totalled in one place |
| Data exposure | Client or company data flowing through tools no one vetted for residency or retention | Each tool was adopted individually, so none went through review |
| Shadow risk | Access that stays live after someone leaves; spend that never appears in any budget | The tool isn't on any central list, so offboarding and finance both skip it |
Direct spend is the obvious one, and even that is usually underestimated because the charges are scattered. But the two below it are what turn sprawl from a budgeting annoyance into a real governance gap. A client's data can pass through a tool that was never checked for where it stores information or how long it keeps it. When someone leaves, their access to a tool nobody knew about doesn't get revoked, because it isn't on any list to revoke. This is the quiet half of sprawl: not the money, but the spending and access no one can see. It's closely related to the broader problem of shadow AI, where usage happens entirely off the books.
Counting the direct spend, roughly
The direct number is worth making concrete, because "a few subscriptions" and "our AI line" feel small until they're added up. Take the 30-person team from the top of this article.
Say a third of them - 10 people - expense one premium AI assistant each at roughly £20 a month. That alone is £200 a month, £2,400 a year, before anyone has bought a specialist tool. Add a handful of team subscriptions to writing, image, and coding tools at £30-£100 a month each, and a few forgotten seats on plans that were bought for a project that ended. The subscription total lands somewhere in the low thousands a year without a single "expensive" purchase on it.
The point of the exercise isn't the exact figure - it's that no one holds it. Ten £20 charges on ten different expense reports never get seen as one £2,400 decision, so the decision never gets reviewed. Sprawl's direct cost hides in how the spending is shaped, not in how large any piece of it is.
The inventory exercise
You can't fix what you can't see, so the first step is not consolidation. It's counting. The goal is a single list of every AI tool the team actually pays for and uses, which almost no growing team currently has.
Run it in four passes:
- Pull the financial trail. Export the last three months of card statements and expense reports, and flag every charge that is or might be an AI tool. Include the small ones - a lot of sprawl hides in $20 monthly line items.
- Ask the team directly. Financial records miss anything on a free tier or a personal account used for work. A one-question survey - "which AI tools do you use for your job?" - catches what the statements don't.
- Group by what the tool does, not what it's called. List each tool next to the job it does. This is where duplication becomes visible: three separate subscriptions that all do first-draft writing, two that do the same summarising.
- Note where the data goes. For anything touching client or sensitive work, record what you actually know about where it processes and stores data. The blanks in this column are your risk list.
The output is one table: tool, owner, monthly cost, job it does, and data sensitivity. Most teams find the exercise itself is the wake-up - seeing the count in one place lands harder than any warning about it.
Consolidating without breaking workflows
Once the list exists, the fix is not to cut everything to one tool. It's to remove duplication and bring the rest under one roof, without ripping working tools out of people's hands.
Work through the inventory in this order:
- Cut exact duplicates first. Where two tools do the same job, keep the better one and cancel the other. This is the fastest saving and the least disruptive, because nobody loses a capability.
- Consolidate overlapping model access. Several subscriptions wrapping the same underlying models can usually collapse into one point of access. Teams genuinely do need more than one model - but they need them assigned deliberately by role, not accumulated by accident.
- Keep genuinely distinct tools, but centralise the account. A specialist tool that does one job well is worth keeping. The change is moving it off a personal card onto a shared, visible account, so it shows up in spend and gets revoked at offboarding.
- Set one rule for new tools. Sprawl regrows the moment the inventory is done if nothing changes upstream. A single lightweight approval step - "check the list before signing up for something new" - is enough to stop the next round.
The aim is one view of what's paid for, who uses it, and where the data goes, without taking a single working tool away from the people using it.
Where Hebno fits
Hebno is the single roof the consolidation step points at. One interface to every major model, so overlapping subscriptions collapse into one access point. Spend is attributed by team and person, so the inventory stays current instead of decaying the moment you finish it. Access is tied to membership, so offboarding closes the shadow-risk gap. The inventory exercise above is worth running whatever you use next, but its point is to get everything into one place, which is what Hebno is for. See how it works on our pricing page.
