Classify business documents using their text and metadata

01

Finding a document is often easier than knowing what it is

Organizations accumulate contracts, reports, policies, invoices, correspondence, project documents and other files, and someone eventually needs to determine:

• What type of document is this?
• Which department owns it?
• Which project does it belong to?
• What retention category applies?
• Which workflow should receive it?

These are classification decisions.

02

Train with information you already have

StayCharted AMT doesn't have to be the system that extracts document contents. Your existing application can provide text or metadata such as:

• Document title
• Filename
• Extracted text
• Description
• Sender
• Existing metadata fields

That data can be used to train the model. For example:

03

Your organization determines the categories

Generic document AI may recognize broad document types. But businesses frequently need more specific distinctions. For example, a legal organization might distinguish:

• Client Agreement
• Engagement Letter
• NDA
• Vendor Contract
• Amendment
• Statement of Work

Another company might care about completely different categories. StayCharted AMT learns the taxonomy represented in your training data.

04

Add it to a document workflow

StayCharted AMT can sit between document ingestion and whatever application manages the resulting file.

1. 1Document
2. 2Text / Metadata Extraction
3. 3StayCharted AMT Prediction API
4. 4Document Category
5. 5DMS / Workflow / Repository

StayCharted AMT handles the prediction. The surrounding application handles document storage, access controls and workflow.

05

Document classification API

Send prepared document text or metadata to a published StayCharted model through the Prediction API. Use the same input fields as training. Your application receives the category and confidence, then files, routes or requests review according to your own workflow.

POST https://amt.staycharted.com/api/v1/models/{id}/classify
Authorization: Bearer amt_live_…

{ "input": "Invoice 88724: freight charges, payment due in 30 days" }

→ { "label": "Invoice",
    "confidence": 0.94,
    "needsReview": false,
    "modelName": "Document types" }

Train on reviewed examples of these document types. Your existing system extracts text from PDFs or scans and retains the original file. StayCharted classifies the supplied information; it does not extract fields, index the file store or verify a signature.

Document classification API: requests and responses →

06

One focused role

Train and test document categories in StayCharted AMT, then use the predictions in your existing document system.

07

Classify logistics documents

Train on reviewed bills of lading, proofs of delivery, invoices and customs papers. Keep extraction and shipment matching in your existing systems.

Logistics document classification: the step-by-step guide →