Ask a CFO what the company spends on cloud hosting and you get an answer in minutes. Ask what it spends on AI and the honest answer, in most Thai enterprises today, is: nobody actually knows.
That is not a failure of any one team. It is a structural problem. AI spend does not arrive as one line item — it accumulates across personal subscriptions employees expense individually, team-level tool budgets approved outside IT, API usage billed per token inside a product nobody thinks of as "an AI cost," and AI features quietly switched on inside SaaS tools the company already pays for.
Three questions, and why they go unanswered
How much are we spending? Spend is split across vendor invoices, personal reimbursements and consumption-based API bills that scale with usage nobody is watching in real time.
Who is spending it? Without a shared control point, usage is attributed to a department budget at best, not to the team, project or model actually driving the cost.
What did the model see? This is the question that turns into an incident. A near miss — someone pasting confidential information into a public model that should never have left the building — is usually the moment this stops being a finance question and becomes a board question.
What visibility actually requires
Seeing the spend is not a reporting problem you solve with a dashboard bolted on afterwards. It requires a control point that every request passes through, so usage can be attributed at the point it happens — by team, by project, by model — including the usage nobody has been assigning to anyone.
From there, two things follow. You can set rules that are enforced at the request itself, not in a policy document: which teams use which models, at what limits, with what data. And you keep a record — which model touched which data, when, and under whose approval — ready for the moment a parent company, a board, or an auditor actually asks.
Where this fits with adoption, not against it
Governance built this way does not mean locking AI down. It is what makes the three-tier adoption model possible in the first place — you cannot run a controlled allowance for exploring users, or justify a subscription for a champion, without first being able to see what each tier is actually costing.
We deploy this as NextBrain, inside your own cloud, so your prompts and your data never transit anyone else's infrastructure. Most engagements start with a short assessment that establishes what you are actually spending today — which is usually the number that gets the project approved.