When AI categorisation is switched off, a transaction that your rules and history cannot place is held for review instead of being guessed. It waits in a queue for you to give it a category. That is by design: without AI, nothing is categorised unless one of your own rules or your own history says so, and the queue is where you decide the rest.
Why are my transactions held?
Categorisation in SpendixAI works in stages. Your own rules come first. Then your history and similar merchants. Only if AI is on does an AI model take what is left. If you turned AI off, whether for a feature, through an exhausted monthly budget or entirely, the last stage does not run, so anything unmatched is held for review. A new merchant, an unusual narration or a payment to an unfamiliar name will land here.
It does not mean something went wrong with the import. The amounts and dates are saved as they were. Only the category is waiting.
What does a typical queue look like?
Say your October card statement has 160 rows. Rules and history place 128 of them. That leaves 32 held. Of those, 9 are one merchant you have never used before, 7 are small UPI payments to different people, and 16 are a mix of new shops. You are looking at about ten minutes of work, not 160 rows.
| Group in the queue | Rows | Best handling |
|---|---|---|
| A new merchant seen several times | 9 | Categorise once, write a rule |
| UPI payments to people | 7 | Categorise by payee, rule for regulars |
| One-off shops and charges | 16 | Categorise by hand, no rule |
| Total held | 32 | About ten minutes |
How do I clear it quickly?
- Sort the queue by merchant so identical names sit together.
- Take the biggest group first, categorise one row, and apply the same to the rest of that merchant if the tracker lets you select several rows.
- Create a rule from any merchant you have just fixed that you expect to see again.
- Merge look-alike spellings with an alias before writing the rule.
- Do the one-offs last, using a general category where a precise one would not change anything.
- Check the queue is empty before you open reports, so the totals are complete.
How do I stop the same items coming back?
The way is rules. Every merchant you categorise by hand is a candidate. A rule catches it next time before the queue ever sees it. Your corrections also become history, so similar merchants start to match without any new rule. After two or three imports, the queue should contain mostly genuinely new things, and that is the right place for your attention.
Should I turn AI back on?
That is your call, and neither answer is wrong. With AI on, the model is shown the normalised merchant name and an amount band, not your name, account numbers, the full statement text or balances, and it fills in what rules and history leave. You can control it per feature, set a monthly AI budget, or keep it off. If you want a middle path, use a small monthly budget: when it runs out, the remaining unmatched items are held for review as before.
A queue-clearing checklist
- Sort by merchant, then by amount.
- Fix the largest repeat group first.
- Write a rule for anything recurring.
- Add an alias if one shop has several spellings.
- Leave one-offs in a simple category.
- Look again next week; the list should be shorter.
Useful ideas start with a clearer picture of your own spending.
Explore your reports

