What Document Intelligence does

Document Intelligence turns authorised business records into classified, structured and usable information. Use it for financial and legal records, policies and other supported document classes. Originals, processing results, field decisions and output receipts remain distinct records connected by their source context.

Your available features depend on your role, plan, tenant configuration and connected services. A visible page does not grant permission to change a policy, approve a result or release an output.

Find the right workspace

Open Document Intelligence from the application menu. Overview summarises operational work. Library holds original files and the processed document register. Processing contains jobs, bulk redaction, intake channels and delivery operations. Review & QC contains extraction review, indexing, quality control, exceptions, financial comparison and work allocation. Insights contains analytics and supported validation tools. Configuration contains schemas, flow designs, domain packs, policies and service controls.

Assurance offers Production, Codebase and Pipeline perspectives on operational state and processing configuration. Codebase is an administrator view of document configuration, not an external Git repository scanner.

Complete your first document

  1. Choose a representative, non-sensitive sample and an authorised source or upload route. Confirm that the file has a current malware-clearance result.
  2. Select the relevant published classification/extraction schema or permitted automatic strategy. Check OCR, review and output settings before starting.
  3. Run processing and open the resulting document in Review & QC. Compare extracted values against the original and locate the supporting source evidence.
  4. Correct or accept fields, resolve required values and validation issues, then use Finish review. Resolve separate indexing or QC gates if shown.
  5. Inspect the eligible output and, where configured, create a delivery preview. Confirm the resulting transfer receipt and check that the receiving system can use the output.

Read completion correctly

Processing finished means machine work has reached its recorded outcome. It does not mean extraction is approved, indexing is confirmed, sampled QC has passed or delivery has succeeded. Check each stage independently.

Confidence is a recognition signal, not a guarantee of correctness. Missing information must remain missing or be resolved by an authorised reviewer; do not invent values to complete a record.

Run a small acceptance exercise

Choose authorised representative samples: native text, a scan, a difficult document and a non-matching class. Record expected classes/values before running. Check source clearance, selected schema/version, extracted values, review completion, index/QC gates and output receipt separately. Agree who owns exceptions and delivery failures before increasing volume.