Automation you can audit: categorization that shows its work
Bao sorts your transactions with a deterministic rules engine, not a machine-learning black box. The same transaction always produces the same category, every rule is one you can read, and every transaction carries a stamp recording where its category came from: UNCATEGORIZED, SUGGESTION, RULE, or MANUAL. So when a category looks wrong, you can always answer the question that actually matters — who decided this, and how do I stop it happening again?
That matters because automatic categorization is quietly won or lost on trust. When an app sorts your spending into neat buckets, you're taking it on faith that the buckets are right. And the moment you catch one that's wrong, with no way to see why, that faith cracks. So we built categorization you can audit.
Deterministic rules, not a black box
A lot of tools categorize with opaque machine-learning models: transactions go in, categories come out, and no one (sometimes not even the people who built it) can tell you exactly why. That can be convenient right up until it's wrong in a way you can't inspect or correct.
Bao takes the transparent path. Its categorization runs on a deterministic rules engine: given the same transaction, it produces the same result every time, following rules you can understand. There's no mystery box guessing at your life, just a set of legible rules doing legible work you can reason about and predict.
If you can't tell why an app made a choice about your money, it's hard to fully trust it.
Every category is stamped with its source
Here's the part we're proud of. Every transaction carries a category source: a small stamp that tells you exactly where its category came from:
- UNCATEGORIZED. Nothing's been assigned yet; it's waiting for you or a rule.
- SUGGESTION. A proposed category you can accept or change. A suggestion, not a decision.
- RULE. Assigned automatically by the rules engine, following a rule you can inspect.
- MANUAL. You set this one yourself, on purpose.
That stamp turns categorization from a black box into a paper trail. You're never left wondering whether a category was your call, a firm rule, or just a soft guess. You can see the difference, which means you can decide how much to trust each one.
MANUAL is sacred. Never overwritten
The most important rule in the whole system is a promise: a MANUAL choice is never overwritten by automation. When you deliberately categorize a transaction yourself, that decision stands. No rule, no suggestion, no future update quietly reaches in and changes it back.
This is what makes the automation safe to lean on. You get the convenience of rules doing the repetitive work, without the anxiety that the app might silently undo something you cared enough to set by hand. The machine helps; you stay in charge.
Learning from the community, privately
Good categorization gets better when it can learn from patterns across many people, and that shouldn't come at the cost of your privacy. So where Bao draws on shared knowledge to improve suggestions, it's designed to be privacy-preserving and k-anonymous: enrichment leans on patterns common to enough people that no individual stands out, rather than on anything that singles you out. You benefit from the crowd without becoming identifiable within it.
Automation that respects you
The whole philosophy here is simple: automation should save you effort without asking you to give up understanding or control. Rules you can reason about, a source stamp so nothing's a mystery, a promise that your manual choices stand, and community learning that never trades away your privacy. That's the shape of it.
Bao 🐼 is happy to do the tedious sorting for you, and equally happy to show its work every step of the way. That's automation you can audit.
If you'd like to see where categorization sits in the wider flow, how Bao works walks through it from linking an account onward, and the features page covers what else runs quietly in the background.