Categorization with suggestions

Where a suggestion comes from

The app builds its reference base from your own already-categorized entries: for each one it keeps the original file description, the amount, the date and the category you chose. On top of that base, it works in two complementary ways.

1. By similar text

The new entry's description is normalized and compared against the already-categorized ones. This is not an equality lookup: the app measures how alike the texts are and only considers a suggestion when the similarity is high. That is what makes "SHELL STATION 4471" find "SHELL STATION 0192" without confusing it with "SHELL MARKET".

2. By recurrence pattern

Beyond the text, the app looks for repeating expenses. When it finds the same kind of entry showing up month after month, with a similar amount and around the same day, it recognizes that pattern and uses it to suggest — even when the description varies more than usual. This is what handles tuition, subscriptions and utility bills, where the bank's text changes but the behavior doesn't.

For a pattern to be recognized, it must have repeated a few times and must not have long gaps between occurrences. An expense that happened twice is not a pattern yet; one that disappeared for half a year has stopped being one.

What a suggestion brings along

A suggestion is not just the category. It also carries what you added to the similar entry in the past:

• The category — the entry's destination.
• The title you wrote — when you replaced the bank's raw description with something readable, that version comes along.
• The marker — if that entry had a marker you defined, it is proposed too.

The on-screen flow

1. Load the suggestions. The app walks the unclassified list and computes each item's proposal. On big lists this takes a few seconds.
2. Review before applying. There is a filter to see only the items that got a suggestion — and only the ones that didn't. Start with the ones that did: that block clears fast.
3. Apply in bulk. One command applies every loaded suggestion at once.
4. Handle the rest in groups. For what is left, select similar entries and categorize them together, setting shared descriptions for the whole group.

Categorizing a group in one go

The bulk categorization screen asks for a category and up to three descriptions to apply to every selected entry. The description each entry already has is not lost: it joins the list as an item called Original text, and the arrows beside it choose where it sits — before the first description you typed, between two of them, or after the last.

That is why the box labels skip a number when the Original text sits in the middle: with it in second place, the boxes read "Description 1", "Description 3" and "Description 4". The numbers describe the order the descriptions will come out in on the entry, not the position of the box on screen.

There is also a Marker field for the whole batch, stamping the same marker on every selected entry — the shortest way to tag everything you have just recognized as being about the same thing. It depends on the Transaction marker switch, under Advanced options.

When a suggestion is wrong

Nothing is lost. Under "Classified Entries" you filter, select and undo the categorization — the items return to the unclassified queue and can be redone correctly. Since the reference base is made of what is categorized, fixing a mistake also improves the next suggestions.

Next steps