Nobody decided to buy fourteen AI subscriptions. That is the important thing to understand about how it happens. There was no meeting, no strategy, no budget line labeled "AI sprawl." Five teams each solved a version of the same problem within about a year of each other, every purchase was individually defensible, and the total was never anyone's job to look at.
Fourteen at around two hundred people is the number we keep landing on when a Scan inventories what is actually being paid for. Three of them overlap materially. The direct cost is real but it is not the expensive part.
The dollars
The subscription line is the easy half. It shows up on a card statement, someone can cancel it, and the saving is bankable within a month.
What the line item hides is that overlapping tools are rarely paid for once. The same capability arrives bundled inside a suite license somebody already renewed, then again as a point solution, then a third time as seats on a platform bought for something else entirely. You are not buying three tools. You are buying one capability three times and using about seventy percent of each.
That is a rounding error at ten people and a real number at two hundred — and it compounds, because every one of those contracts renews.
The decisions
Here is the part that does not appear on any statement.
Fourteen tools means fourteen places where work happens and none where it accumulates. Every one holds a slice of context: this one saw the support conversation, that one drafted the follow-up, a third scored the account. No single one saw the case end to end, and none of them can tell you what happened afterwards.
The second column is why we treat sprawl as a strategy problem rather than a procurement one. A company running fourteen tools is not overspending by a bit. It is failing to accumulate anything, in a period when accumulated outcome data is the only durable advantage available to a business that does not train its own models.
Fourteen tools is not a spending problem with a decision-making side effect. It's a decision-making problem that happens to have an invoice.
What we actually recommend
Not rip and replace. That answer is almost always wrong, and it is usually the answer a vendor is being paid to give.
The Scan inventories every AI subscription, shows the overlap in dollars per month, and then sorts the tools into three buckets:
- Keep and integrate. It fits your workflow off the shelf. Buying it was the right call. We wire it in rather than rebuild it, and we say so plainly.
- Keep for now, plan around. It works, but it is holding context that should be accumulating somewhere you own. Fine to run, not fine to build your future on.
- Replace. You are renting one hundred percent of a tool for about a seventy percent fit, in a workflow specific enough that the gap is where the value was.
Most inventories come back mostly bucket one and two. If a tool fits, buy it — we will tell you so, and it is a cheaper answer than anything we could build.
The ones worth replacing are the ones where the workflow is the differentiator, and you have been renting a generic version of the thing that makes you different.
Where to start on Monday
Pull the card statements and list every AI-adjacent line, including the ones bundled into suites. Next to each, write the workflow it touches. Then count how many distinct workflows you have and how many tools are pointed at each.
Any workflow with three or more is where your money and your context are both leaking. That is your first conversation — and it is a much better first conversation than "what should we do about AI."
Get the next one first.
Field notes from the lab on building AI systems that hold up in production — sent when we publish. No noise.
