How AI Categorization Is Planned to Work in EXPOZOR

EXPOZOR is available now as a live web application. AI-assisted categorization is a planned feature rather than a production capability, and this post explains the product direction I am designing toward.

Rules first, suggestions second

The planned categorization workflow starts with user-defined rules. If you decide that a merchant belongs in a category, that rule should take priority over any AI suggestion.

The intended order is:

1. Exact merchant rules - examples like "Planet Fitness = Health & Fitness."
2. Pattern rules - examples like "AWS = Business / Infrastructure."
3. AI-assisted suggestions - used only when no rule matches.

That keeps AI in the role of assistant, not authority.

Review before relying

AI suggestions can be useful, but they can also be wrong. EXPOZOR's planned workflow keeps review visible so users can confirm, edit, or reject a suggested category before treating it as reliable.

I am not naming an AI provider yet because no production provider has been contracted. When a provider is selected, the Privacy Policy and Subprocessors page should be updated before the feature goes live.

What exists today

Today, the live web application stores account information and the expense records a signed-in user chooses to create or import. It does not collect bank credentials, receipt images, or screenshots. CSV files are parsed in the browser, and only approved normalized rows are sent for storage.

The web app is configured with HTTPS, HSTS, Content Security Policy, X-Frame-Options, X-Content-Type-Options, Referrer-Policy, and Permissions-Policy headers. EXPOZOR does not sell account or workspace data.

Why this matters

Categorization accuracy matters because budgets and spending summaries are only useful when the categories are trustworthy. The product direction is simple: make categorization faster without removing user control.