# Get useful data out of difficult documents.

Document intelligence

Extract fields and tables, validate totals and route uncertain records to review with the source attached.

## From document to reviewed record

Illustrative workflow.

- Supplier: Matched against the supplier register
- Invoice reference: Checked for an existing record
- Tax and totals: Reconciled before accounting export



## A reliable record starts with a reviewable field.

### Document intake

Handle document types, duplicate files, extraction limits and the source record associated with each upload.

### Field validation

Check extracted values against required formats and business rules before they reach downstream systems.

### Reviewer workflow

Route uncertain fields with the relevant source passage so a person can correct the result in context.

### Asynchronous processing

Keep uploads responsive while long extraction tasks run in a queue with visible progress.

### Source retention

Associate results with the source version and apply the agreed retention rules to both.

### Downstream delivery

Submit validated values through explicit contracts and reconcile the resulting business record.

## Extraction is not the same as validation

### Copy a model response into the system

An incorrect value becomes an apparently valid record, with no source available to the reviewer.

### Validate the record before release

Preserve the document reference, highlight exceptions and export only after the required checks and approvals.

## Extraction needs a review path

Scanned invoices, forms and correspondence rarely arrive in one reliable format. Connect every extracted value to its source, validate it against business rules and give uncertain fields a clear owner.

Illustrative scenario, not a customer case study.

An accounts team extracts totals and line items from supplier documents in varied Australian invoice formats.

Preserve the original page and bounding boxes. Validate extracted fields against arithmetic and schema rules, then route uncertain or inconsistent records to a reviewer.

Verification: Measure field accuracy, document-level acceptance and the rate of incorrect automatic acceptance separately.

## What needs attention in your system?

Select the areas you want to discuss. The HTML page can download your selections.

- [ ] Document intake: Handle document types, duplicate files, extraction limits and the source record associated with each upload.
- [ ] Field validation: Check extracted values against required formats and business rules before they reach downstream systems.
- [ ] Reviewer workflow: Route uncertain fields with the relevant source passage so a person can correct the result in context.
- [ ] Asynchronous processing: Keep uploads responsive while long extraction tasks run in a queue with visible progress.
- [ ] Source retention: Associate results with the source version and apply the agreed retention rules to both.
- [ ] Downstream delivery: Submit validated values through explicit contracts and reconcile the resulting business record.

## Do all documents need manual review?

No. Agree which validated cases may proceed automatically and which exceptions require a reviewer. Use field-level evidence rather than one confidence score for the entire document.