What the firm's AI use costs
How an accounting or audit firm can read every AI request and its estimated cost in Achekam, review spending monthly and switch AI off when it must.
By Achekam · · 5 min read · For Accountants & audit firms in South Africa, Nigeria, Kenya
The first AI invoice of the month is rarely the problem. The problem is the month a manager discovers that one engagement team asked the AI to redraft the same set of notes forty times, or that a feature nobody thought expensive has been running against long documents all quarter. By the time the vendor's bill arrives, the money is spent and nobody can say which work it served.
Partners do not need to become experts in tokens and model pricing. They need two things: a record of what was asked of the AI, by whom and at what cost, and a way to stop it if something goes wrong. This article explains what Achekam gives a firm on both counts, where the limits are, and a monthly routine that turns the record into control.
Every request, on the record
The AI Activity Log lists every AI request made in the workspace, with the vendor, the model, the number of tokens and an estimated cost.
That list answers questions that are otherwise guesswork:
- Which vendors and models the firm's work actually goes to, which matters for the firm's own policy on where client information may be sent.
- How much is being used, and whether use is growing steadily with adoption or jumping in a way that needs explaining.
- What it costs, request by request, so unusual spending can be traced back rather than averaged away.
Every AI feature in Achekam runs through the same AI engine, so the log is one place to look rather than a log per feature. The engine can use Anthropic, OpenAI, Google Gemini, DeepSeek or xAI.
Estimates, not invoices
The costs in the log are estimates, worked out from the vendors' list prices. They are not the vendor's bill.
Treat them as a management figure. They are good for spotting trends, comparing months and finding the expensive outliers. They are not a figure to post to the ledger or to recharge to a client as if it were exact. Vendors change prices, apply discounts and bill in their own way, so the estimate and the invoice will rarely match to the cent.
If the firm uses its own API key, the vendor's own billing for that account is the figure to reconcile the log against each month. A large gap between the two is worth a question, either to the vendor or to whoever administers the firm's AI settings.
No spending cap the firm can set, yet
A firm cannot yet set its own monthly limit on AI spending from any screen. The AI engine would enforce one if it were stored, but there is no field in the settings panel to store it. Usage is metered per firm, and a monthly token allowance applies only where the platform owner has set one.
Until a firm-set limit exists, control has to come from people:
- Decide a monthly amount the firm considers reasonable for AI use, based on the first few months of the log.
- Name one person, usually the administrator who manages the AI settings, to read the log each month.
- Agree what happens when spending runs well above the amount: a conversation with the teams responsible, not an automatic cut-off.
Switching AI off
A workspace administrator can switch AI off for the whole workspace. It is a blunt control, and that is its value. If a client objects to AI being used on their file, if a data concern needs investigating, or if spending has run out of control, the firm can stop all AI use at once while the partners work out what to do.
Switching off does not undo anything, and it is not a substitute for a policy. But a firm should know who holds that switch and in what circumstances they will use it, and write both down.
A related point for firms that have not yet added a key: with no AI key at all, every AI feature says so and drafts nothing, rather than appearing to work.
The firm's own key or the platform's
The engine uses the firm's own API key first, and the platform's key only where that is allowed. Administrators set keys and routing under Settings > "AI models, keys, routing & spend".
The choice affects cost control as much as data control:
| Firm's own key | Platform's key, where allowed | |
|---|---|---|
| Whose vendor account | The firm's | The platform's |
| What to reconcile the log against | The vendor's bill to the firm | Usage metered per firm |
A firm that wants a direct contract with the vendor, and a bill it can check line by line, will usually prefer its own key.
What the log will not show you
The log shows requests, not judgement. It cannot tell you whether a draft was any good, or whether a person reviewed it before it went into a file.
Two safeguards sit elsewhere. The engine refuses any task whose type is giving an opinion, reaching a conclusion, approving or signing. And by default it removes South African ID numbers from text before it is sent to an AI vendor, a setting that can be switched off. The fuller picture of what the AI can and cannot see is in what Achekam's AI can see, do and refuse.
A monthly review routine
COSO's guidance on the Internal Control – Integrated Framework includes supporting guidance on monitoring and on technology; a monthly look at AI use is a small, practical example of both. A routine that takes less than an hour:
- Compare this month's estimated cost with last month's and with the firm's agreed amount.
- List the largest requests and find out what work they served.
- Check that the vendors and models used are the ones the firm's policy allows.
- If the firm uses its own key, compare the estimate with the vendor's bill.
- Record the review, any questions raised and the answers, in the firm's quality records.
All of this belongs in the firm's written rules for AI; writing the firm's AI use policy sets out what those rules should cover.
Open the AI Activity Log this week, note last month's estimated total, and decide who will read it every month from now on.
General guidance only, not advice on your own circumstances. Tax and payroll rules change: check the current position with the authority or your adviser.
