A monthly AI budget caps how much AI categorisation your tracker can use in a month. Set it by estimating how many rows are left after your rules and history have done their work, since only those go to the AI model. When the budget runs out, the rest are not guessed; they are held for review and wait for you.
Why cap AI use at all?
- Cost control: AI credit packs are separate from your plan, so a cap stops usage running past what you meant to spend.
- Predictability: you know the most that can be used in a month.
- Privacy comfort: less AI use means fewer items ever reach the model.
- Habit: a cap nudges you to write rules, which are free and firmer.
In SpendixAI you can switch AI off per feature, give it a monthly AI budget, or turn it off entirely. The budget is the middle option.
What goes to the AI model, and what does not?
Categorisation runs in stages. Your own rules come first, then your history and similar merchants. Only what neither can place goes to the AI model, and then only if AI is on. The model sees the normalised merchant name and an amount band, not your name, account numbers, the full statement text or balances. That means the number of rows that reach it depends mostly on how good your rules and history are, not on how long your statement is.
How do I size the cap?
Work from the leftovers. Look at a recent import and count how many rows were not placed by rules or history. Those are what AI would be asked about. Then allow a margin. Here is a worked example with invented numbers.
| Step | Example | Result |
|---|---|---|
| Rows on a typical month's statements | 180 | 180 |
| Placed by your rules | 110 | 70 left |
| Placed by your history | 30 | 40 left for AI |
| Margin for a busier month, say 50 percent more | 40 x 1.5 | 60 rows |
So you would want a budget that covers about 60 rows a month. Check your settings screen for how the budget is expressed and what each item uses, because the unit and the price depend on your own account. Expect the figure to drop over time: as you write rules and your history grows, fewer rows are left for AI.
What happens when the cap is reached?
The rest of that month's unmatched items are held for review. Nothing is saved with a guessed category and nothing is lost. You can work through the queue yourself, which also teaches the tracker: your corrections become history, and any repeat merchant can become a rule. You can raise the budget if you decide it was set too low, or buy an AI credit pack if you want more headroom.
What about the optional features that read a document?
Some optional features, such as reading an unfamiliar statement layout, process the uploaded document itself. Decide on those separately from categorisation. If you prefer, leave them off and upload CSV or Excel files, which are usually easier to read.
A setting checklist
- Count the unmatched rows from your last import.
- Add a margin for a busy month.
- Set the monthly AI budget to cover that number.
- Write rules for any merchant that keeps reaching the model.
- Review the held items once a week when the cap is reached.
- Revisit the budget after two months.
Useful ideas start with a clearer picture of your own spending.
Explore your reports

