Categorisation is one of the most valuable features in personal finance software, and also one of the most overused. A good category structure makes a noisy transaction feed legible at a glance. A bad category structure — or a good one applied too aggressively — replaces the underlying record with a simplified summary that nobody can audit.
The first problem with category-led thinking is that most spend does not fall cleanly into a single bucket. A meal at a café might be groceries, a social event, or a work expense depending on who was there and why. A hardware-store purchase might be home maintenance, a hobby, or a gift in disguise. Forcing every transaction into one tag makes the system tidier than reality.
The second problem is that categories drift. Personal definitions of 'eating out' or 'transport' shift over time, and the software's default taxonomy rarely matches the user's. Once a user starts disagreeing with their own past categorisations, the historical chart they relied on loses meaning. The line goes up or down for reasons that have more to do with labelling than behaviour.
A more durable approach treats categories as a working interpretation rather than a fixed truth. The underlying transaction is permanent; the category is editable, optional, and ideally explained. If a user can see why a payment was placed in a particular bucket — a merchant pattern, a recurring rule, a manual override — they can also see when that reasoning no longer applies.
Granularity is another tension worth getting right. Too few categories hide useful patterns; too many create an admin burden that most users will eventually abandon. The pragmatic middle is around eight to twelve categories at the top level, with optional sub-tags for anyone who wants to go deeper. That structure tends to survive contact with real life.
Categorisation also benefits from being asymmetric. Recurring payments deserve precise categories, because they repeat and accumulate. One-off transactions can be categorised more loosely, because the cost of misclassifying a single coffee is genuinely low. Software that applies the same effort everywhere wastes both its attention and the user's.
The point of categorisation is to make the underlying records easier to read, not to replace them. The records are the asset. The categories are the index. A useful index respects the source it is indexing — which means staying honest about uncertainty, leaving the source easy to consult, and never pretending to know more than it does.
Key takeaway
Categories are a useful lens on financial data, not a replacement for it. When the lens becomes the subject, the picture distorts.