# Document extraction is not document automation. > Reading an order is only the first step. The work is complete when the details have been checked, any changes approved and the correct order saved in the business system. Canonical URL: https://marmation.com/review/document-extraction-is-not-automation/ By Vikram Roy · 2026-11-12 Reading information from a document does not complete the work it started. The product codes need checking, a person may need to approve a change, and the right order needs to be saved. A supplier sends an order as a PDF. A model reads the customer name, items and quantities. The demo appears complete. But one quantity has an unclear unit, two descriptions could refer to different stock items and the order-management system requires approval before a substitute can be used. ## A readable field is not a valid transaction Text extraction is only the first step in a transaction. Automation must determine whether information is valid, which source is authoritative, when a person must review it and how the result is saved without creating a duplicate. ## Design the route from document to confirmed result Count correctly completed orders and the time spent resolving problems. Text-reading accuracy alone misses the work your team still has to do. Agree which kinds of document the process handles and link each value to the original. Check the details against business rules and official customer or product records. Send unclear cases to a reviewer, record approval and connect the confirmed order to the receiving software. Check what was actually saved, and make it possible to retry or reprocess a request without losing its history or creating a duplicate. ## Use the difficult documents in the test Test with low-quality scans, outdated templates, multi-page tables, contradictory totals and duplicate submissions. Capture which fields were machine-extracted and which were corrected. Preserve an audit trail so the business can understand what happened after approval. ## When a managed extractor is sufficient For standardized, low-risk documents, a managed extraction service may be adequate. A custom system is justified when the business rules, UX and integrations are distinctive enough to create enduring value. ## Start with one document family and one action Scope the first document family and one downstream action. Define a baseline and the cost of errors before proposing a broad automation program. **The decision:** Fund the complete path from an incoming order to a checked result in the business system. Judge it by correctly completed orders and the effort still needed to resolve problems. ### References and context [RSM middle-market survey](); [NIST AI RMF](). ## Sources - [2026 Middle Market AI Survey]() — RSM. Survey of 1,030 US and Canadian middle-market executives at organizations already using AI; 86% integrated in operations; 36% fully embedded across core processes. Limit: This is a snapshot of self-reported responses from the firms surveyed, not a result for every middle-market business. Read the findings in the context of when the survey was conducted. - [Generative AI Profile for AI RMF]() — NIST. Reference governance and risk design principles for gen AI. Risk management reference for generative AI Limit: This is a risk-management reference, not a certification or evidence that Marmation has been audited or complies with it. ## Related reading - [Make intelligence part of the work](https://marmation.com/review/operations/) - [Your ERP is a system of record. It is not the entire workflow.](https://marmation.com/review/erp-is-not-the-entire-workflow/) - [Your knowledge assistant cannot repair an ungoverned knowledge base.](https://marmation.com/review/knowledge-quality-is-ai-quality/)