The finance review that found AI everywhere
Nobody decided to spend this much on AI. It happened one line at a time. Marketing has a writing tool. Sales pays extra for the AI tier of the CRM. The meeting software added an AI notes feature that someone switched on. A few people claim ChatGPT Plus on expenses. A developer's card is paying an OpenAI API bill for an internal tool. And now the Microsoft 365 renewal includes Copilot for a department.
When finance finally adds it up, the total is a surprise, and nobody can say which of these are earning their keep. Several tools do roughly the same job, and some of them have not been opened in months.
Why AI spending sprawls
AI features are sold in small, easy-to-approve amounts, often as upgrades to software you already use, so they rarely go through a proper buying decision. Each one makes sense on its own. Nobody is looking across them all, so overlaps are invisible.
Usage-based API bills add a different problem. They grow with use, often in ways nobody predicted, such as a long document fed in on every request, a loop that retries too often, or a feature that became popular. Without per-feature or per-team reporting, a bigger bill is just a bigger number.
Where the money leaks
| Source | Typical leak |
|---|---|
| AI add-ons in existing software | Paid across all seats, used by a few |
| Personal subscriptions on expenses | Duplicates company tools, with no data controls |
| Overlapping writing and note tools | Several products paid for one job |
| API usage without limits | Bills grow with no view of which feature caused it |
| Pilots never switched off | Keys and subscriptions keep running after the trial ends |
The cost is not only money. Every separate tool is another place your data goes, another set of terms, and another account to close when someone leaves.
How we get AI costs under control
- We build an inventory of AI spend from card statements, expense claims, software invoices and admin consoles, including the AI tiers hidden inside other subscriptions.
- We match each item to usage data where it exists, and to a short conversation with the people using it where it does not.
- We group tools by the job they do and recommend which to keep, which to cut, and where a single company tool can replace several personal ones.
- For API usage, we put a gateway between your applications and providers such as OpenAI and Anthropic, so every call is tagged by application and team, and budgets and alerts apply before a bill arrives.
- Within the applications, we look at the common causes of waste: oversized prompts, repeated calls that could be cached, and expensive models used for simple tasks that a cheaper one handles as well.
- We set up a monthly AI spend report that shows cost by tool, team and feature next to usage, so decisions are made on data.
We do not recommend cutting tools people rely on just to reduce the line count. The aim is to pay for what gets used, once.
What you have afterwards
A single list of every AI tool and API the business pays for, who owns it and what it is used for. Fewer overlapping subscriptions. API bills that can be broken down by feature, with alerts before a budget is exceeded rather than after.
And a simple rule for new AI purchases: check the list first, because the thing someone wants may already be paid for.
For the applications you build yourself, the gateway also makes future changes cheaper. If a provider changes its prices or a new model does the same job for less, you can switch one feature at a time and see the effect in the next report, instead of changing code across several systems and hoping.
Is this your situation?
- Nobody can give you a single figure for what the business spends on AI.
- Staff claim AI subscriptions on expenses.
- Several teams use different tools for the same job.
- An AI API bill has grown and nobody can say which feature caused it.
- AI add-ons were switched on in software you already use without a review.