Month two comes in better categorised because by then the tracker has something to learn from: the corrections you made in month one. The first import has no rules and no history of yours, so more rows need a decision. After you have reviewed it once, the same merchants are recognised the next time. That is the honest expectation: the first review is the most work, and each month after asks less of you.
What happens to a row when you import it
SpendixAI categorises in stages, and each stage only handles what the earlier ones could not.
- Your own rules come first. If you have said that a certain merchant always belongs in a certain category, that wins.
- Next comes your history and similar merchants. If you have categorised that merchant before, or one that looks much like it, the earlier choice is reused.
- Then, only if AI is switched on, an AI model handles what is left. It sees the normalised merchant name and an amount band, not your name, account numbers, full statement text or balances.
- Anything still unmatched is held for your review.
You control the third stage. AI can be switched off for individual features, given a monthly AI budget, or turned off entirely. When it is off, unmatched items simply wait for you.
Why is the first import the hardest?
On day one the tracker knows nothing about how you spend. Stage one has no rules, stage two has no history, and so almost everything depends on defaults, similar merchants and, if you allow it, AI. Some rows will be wrong, and some will be in review. That is normal and it is the cost of teaching the system, once.
Think of it as training a new helper in the house. The first week you explain where everything goes. By the second month they already know the milk vendor, the electricity bill and the pharmacy, and only the new shop needs a word.
What your corrections do
Every correction you make becomes history. If you move a transaction from Shopping to Gifts, the next time that merchant appears it can follow. For merchants that show up under messy names, such as a payment handle or a long string of letters, you can add a merchant alias so every variant maps to one clean name. For merchants that should always go to one category, add a rule, which outranks everything else.
| You do this in month one | What it changes in month two |
|---|---|
| Correct a category on a merchant | That merchant is recognised from your history |
| Add a rule for a repeat merchant | The rule applies before anything else |
| Add an alias for a messy name | Different spellings group under one name |
| Leave AI on, review its suggestions | Fewer rows need a decision, with your corrections feeding history |
What month two will not do
It will not be perfect. New merchants, one-off purchases, small vendors who show up as personal names, and anything unusual will still land in review. The staged approach reduces the pile; it does not remove it. Treat any promise of flawless automatic categorisation, from any tool, with suspicion. A realistic goal is a short review each month instead of a long one.
A simple month-two checklist
- Import the new statement into the same account as before, so history and account stay linked.
- Look at what ended up in review first; that is where your time matters.
- Spot-check a few categorised rows from your biggest merchants.
- Correct any wrong ones and add a rule if the mistake would repeat.
- Skim the dashboard or insights page for anything that looks odd.
If month two is still a long slog, it usually means the first review was rushed, or a few large merchants have inconsistent names. Add aliases for those and the improvement tends to show up in the next import.
Useful ideas start with a clearer picture of your own spending.
Explore your reports

