To keep AI costs from creeping up in an expense app, you need three controls: a monthly budget that warns you, a hard cap that stops spending when reached, and an off switch for features you do not need. SpendixAI has four AI modes, a monthly AI budget and a hard cap, so you decide what is allowed before the first import, not after the first surprise.
Why gig workers are exposed to creeping AI cost
Your statements are busy: dozens of small credits from platforms, lots of fuel stops, many one-off merchants. The more unfamiliar merchants a month contains, the more work falls to the AI step. A quiet month and a festival-season month can cost very different amounts, and income is already uneven, so costs should not be.
How SpendixAI sorts transactions before AI is used
Categorisation runs in stages. First your own rules, then your history and similar merchants, and only if AI is on, an AI model handles what is left. It uses the normalised merchant name and an amount band, not your name, account numbers, full statement text or balances. That order is the key to cost control: the more rules and history you build, the less is left for AI.
A worked example with made-up numbers
Say a month brings 320 transactions. Your rules match 190, history and similar merchants match another 85, which is 190 + 85 = 275. That leaves 320 - 275 = 45 for AI or review. Next month you add rules for the ten most common leftovers, which clear 22 of them. Now 23 are left, roughly half. Corrections you make also become history, so each month the leftover shrinks.
| Month | Transactions | Matched by rules and history | Left for AI or review |
|---|---|---|---|
| First | 320 | 275 | 45 |
| Second | 310 | 287 | 23 |
What are the controls?
- Off per feature: switch off any AI feature you do not use, and unmatched items are held for review instead.
- Monthly AI budget: a ceiling you set so use is tracked against it.
- Hard cap: a stop, not just a warning.
- Off entirely: no AI at all; everything unmatched waits for you.
- Four AI modes: choose how much AI is allowed to do.
Check the settings screen for exactly what each mode does and how the cap is counted; this page does not list them.
What should my budget be?
Decide what you would be comfortable paying for categorisation help in a month and treat it as a business cost like data recharge. Say your comfort figure is ₹150. Set the budget below it and the hard cap at it. AI credit packs are sold separately, so extra credits are a choice you make, not something that happens to you.
What if the cap is reached midway through the month?
When a hard cap stops AI work, unmatched items are held for review rather than lost. That is a feature, not a fault: you decide whether to spend your own ten minutes on them or raise the cap. For a gig worker, a sensible response is to look at what is waiting, because the stuck items are exactly the merchants you have not yet ruled. Add rules for the repeats, correct the rest by hand and the held list clears without any extra AI spend. If the same merchants sit there month after month, the real fix is a rule, not a bigger budget.
Check that the notification settings send you a warning before you reach the limit, so you can choose rather than react.
A monthly check for gig workers
- Look at how many items were left for AI or review.
- Create rules for the merchants that keep recurring, such as a petrol pump or platform payout.
- Confirm the budget and cap still fit the month.
- Correct wrong categories, so history improves.
- Switch off features you did not use.
Useful ideas start with a clearer picture of your own spending.
Explore your reports

