Document classification

Every document a client uploads to an organizer — or that you upload from your firm — goes through AI classification. Assure Pro identifies the form type, files the document into the right folder, and tries to match it to an open checklist item.

This is automatic. You don’t trigger it, you don’t pay for it per document, and you can override anything that’s wrong.

The 25 default document types

When a firm is created, Assure Pro seeds 25 document types into your library. These are what the classifier picks from — and what you can pick from when building document items in a template.

Income (11 types)

Deductions (7 types)

Statements (2)

Tax returns and forms (3)

Other (2)

What happens on upload

When a document is uploaded:

1. Assure Pro saves the file and starts a classification.
2. AI looks at the document and picks a type from your firm’s library.
3. The result populates two pieces of metadata on the document:
   • The classified type — for example, 1099-DIV.
   • A confidence score from 0 to 100%.
4. If the document matches an open checklist request (same document type, still pending), it auto-attaches.

The classifier retries a few times automatically if a transient error occurs. If it permanently fails, the document sits unclassified — you can classify it yourself.

Where you see classifications

In organizer review

Each uploaded document in organizer review shows:

• The classified type (a chip).
• The filename.
• A View document link to open the file.
• If confidence was below your firm’s auto-match threshold, the row is flagged for manual confirmation.

In the Documents module

Open Documents from the sidebar. Each document row shows the AI-classified type. Click any row — the Documents drawer opens with the classified type at the top and a Change type button if you need to reclassify.

In the AI Review queue

Documents with low classification confidence (or whose extraction needs review) appear in Documents → AI Review Queue. See Extraction & review for the full review surface.

Reclassifying

If the AI got it wrong:

Reclassifying:

• Updates the classified type to your pick.
• Re-runs the auto-match against open checklist items.
• Records an activity entry naming you as the human classifier.

The document is now treated as human verified for future training and downstream logic.

Managing your document types

Open Settings → Document Types from the sidebar.

The page lists all your firm’s document types organized by category. Each row shows:

• The name.
• Category.
• Description.
• Whether it’s a system type (built-in, can’t be deleted) or a firm-custom type.
• Active or archived state.

You can:

Categories are free-text — you can group your types however you like. The default categories are Income, Deduction, Tax return, and Other, plus whatever you add.

When AI confidence is low

Below a firm-set threshold (default 70%), the document is classified but flagged for review. In organizer review, the row is highlighted amber. In the Documents module, the row has a “Review needed” indicator.

Above 90%, the auto-match runs immediately and the row reads as confirmed.

Between 70% and 90%, the document is auto-classified but the row asks for your one-click confirmation.

You set these thresholds in firm settings — see Settings → Documents → Auto-match confidence (covered in Phase 7).

What classification doesn’t do

A few things to be clear about:

• It doesn’t read tax data. Classification picks a type. Extraction pulls field values (employer name, EIN, wages, and so on).
• It doesn’t take action on the document. That’s the next step.
• It doesn’t decide if a document is required. That’s set by your template’s checklist items.
• It doesn’t validate the tax year. A 2023 W-2 uploaded to a 2025 organizer still classifies as W-2 — you’d notice when the client name and year look wrong in extraction.

Tips

”I keep getting documents that classify as Other”

Two things:

1. Add a new document type that matches what you keep getting (Settings → Document Types → New type).
2. Check that the AI can actually read the document. Some PDFs are scanned at very low quality and the model can’t read them clearly.

”The same document gets classified differently each time”

It shouldn’t. The classifier is deterministic for the same file — same file in, same answer out. If you’re seeing variance across files, it’s because the documents themselves differ (cropping, scan quality, layout). Picking a type once for a borderline document helps your firm’s future matching.

”I want classification to never auto-match”

Set your firm’s auto-match confidence threshold above 100%. Nothing will auto-match. Every upload waits for your manual confirmation.

Permissions

Next

• Extraction & review — pulling field values out of the classified document.
• The document checklist — how requests and classifications connect.
• The Documents module — full coverage in Phase 5.