Automatic categorisation is never perfect on day one. What makes it improve is feedback, and the most direct feedback is a correction. When a transaction lands in the wrong category and you change it, you are not just fixing one entry; you can be teaching the app how that merchant should be treated in future. Knowing how to make that correction count turns a small annoyance into a lasting improvement.
What a correction teaches
The app learns from the choices you make. If a merchant is filed under entertainment and you change it to groceries, that choice is a signal. Repeat it consistently and future payments to that merchant are far more likely to land in groceries. This is how a directory of learned merchants becomes accurate for your life specifically.
Consistency is the key word. A correction applied once and then reversed next time teaches nothing. The clearer your pattern, the faster the system settles.
Correcting with purpose
Before changing a category, ask a quick question: is this a one-off or a pattern? The answer decides how you correct.
- **A pattern.** The merchant will usually belong in the new category. Make the change, and let the correction guide future entries. If a rule suits better, create one.
- **A one-off.** This purchase was unusual for that merchant. Change only this transaction, and do not expect the default to shift.
Many people treat every correction the same way and end up with a muddled directory. Naming the situation to yourself first keeps the teaching clean.
A short example
A person pays a general store regularly, always for household supplies. The store is first filed under shopping. They correct three consecutive payments to household. From then on, new payments arrive filed as household, because the pattern is now clear. Later, they buy a gift there once. They change that single entry to gifts, knowing it is an exception, and the household default stays unchanged.
When corrections do not stick
If a merchant keeps returning to the wrong category despite your corrections, look for competing instructions. A rule might be assigning a category and overriding the learned behaviour. The rule editor has a setting about whether rules override your manual choices, and checking it can explain surprises. Also check whether the merchant exists under several names, since each spelling learns separately.
Using corrections alongside the directory
Combine corrections with a visit to the merchants screen. After a month of fixing individual entries, open the directory and confirm that the defaults match what you have been choosing. Where they differ, set the default directly, which is quicker than waiting for the system to infer it.
| Situation | Best response |
|---|---|
| Wrong category, will recur | Correct, and consider a rule |
| Wrong category, once only | Correct that entry alone |
| Merchant has several spellings | Tidy the directory first |
| Corrections get overridden | Check rules and their override setting |
A correction diary for tricky merchants
Some merchants never settle: a large store that sells everything, a payment app that hides the actual payee, a shared account used by several people. For these, a few lines of notes help. Write down the rule you decided on, such as supermarket purchases above a certain amount are household. Following your own written rule makes the corrections consistent, and consistency is what the learning depends on.
A modest ambition
No one needs perfect categorisation. A realistic target is that most transactions arrive right and the rest take a few seconds to fix. Each purposeful correction moves the system a little closer to that, and over a few months the workload falls without any special effort.
Useful ideas start with a clearer picture of your own spending.
Explore your reports

