Almost every personal finance product offers categorisation. Almost none do it well enough that users stop arguing with it. The defaults are too granular or not granular enough. Merchant matching is brittle. Edge cases like split bills, refunds, and ambiguous purchases get handled poorly. Over time, most users either give up on categories or develop elaborate manual overrides.
Good categorisation starts by being honest about confidence. Some transactions are unambiguous — utility bills, insurance premiums, payroll deposits. Others are inherently uncertain. A system that labels both kinds with the same visual weight teaches the user to distrust the certain ones along with the uncertain ones. A small confidence cue, used sparingly, restores the difference.
It also matters how the system handles disagreement. When a user re-categorises a transaction, a good system asks one short question: just this one, or all similar in future? That single decision turns a manual correction into a durable rule, and turns the user from a janitor into a teacher.
Granularity needs to be tuned to behaviour, not appearance. The categories that benefit most from being precise are the ones that recur, because precision compounds across many transactions. The categories that benefit least are the ones that contain rare, varied transactions, where fine-grained labels add admin without adding signal. Most products apply the same level of detail to both, which is part of why categorisation feels so heavy.
Naming matters more than people expect. 'Food' covers too much; 'Eating out' versus 'Groceries' captures behavioural difference. 'Shopping' is so broad it usually needs to be broken up to be useful. The right names are the ones the user would use to describe their own spending, not the ones a generic taxonomy provides.
There is also the question of where categorisation does not belong. Some transactions — large one-offs, internal transfers, refunds — are not really expenses and should not be forced into expense categories. A good system has somewhere for these to live without distorting the rest of the picture.
The deepest test of a categorisation system is whether users keep using it after the novelty wears off. Most do not. The ones that do tend to share these properties: honest about uncertainty, easy to correct, asymmetric in effort, and named the way users actually talk. That is a higher bar than most products clear, and worth the work.
Key takeaway
Categorisation is the most reviewed and least respected feature in personal finance software. A few principles separate the systems that hold up from the ones users eventually abandon.