Document automation software in 2026: 12 tools compared

Since 1 September 2026, every VAT-registered business in France, whatever its size or legal form, must be able to receive e-invoices through an approved platform (plateforme agréée). Large companies and mid-caps (ETI) must also issue them. SMEs and micro-businesses have until 1 September 2027 at the latest to start issuing. E-reporting, which covers sales to consumers and international transactions, follows the same calendar.

An e-invoice is a structured file in one of three accepted formats: UBL, CII, or a mixed format such as Factur-X, which pairs structured data with a readable image. The DGFiP FAQ points out that a plain PDF sent electronically does not count as an e-invoice under the reform.

An approved platform (formerly called PDP) is a private operator registered by the French State for three years, renewable. It is the mandatory intermediary for sending and receiving invoices, and it passes data to the tax authority. Businesses choose freely, can use several and can switch at any time. Since 1 September 2026, invoices also carry four new mentions: the customer's SIREN number, the type of transaction, the option to pay VAT on debits where it applies, and the delivery address if it differs from the customer's.

What the reform changes for invoice extraction

Invoices between VAT-registered businesses established in France arrive as structured data through the platforms once the issuer falls under the obligation, and the issuer's platform extracts the tax data itself. For that flow, OCR is no longer needed. Every other incoming document still arrives as a PDF, a scan or paper.

On the generation side, invoices now come out of your invoicing software connected to an approved platform, as soon as the obligation to issue applies to your business. A document generator keeps its role for quotes, contracts, certificates and legal documents.

"Document automation" names two needs that run in opposite directions. Generation produces documents that leave your company. Extraction reads documents that come in from outside and pulls data out of them for your software.

The heaviest volumes often sit in regulated work: documents and exhibits in law firms, patient files, or tenant files and mandates in real estate. For tools built for contracts and legal documents, see also our comparison of the best AI tools for legal work.

When your documents fall outside the models vendors provide, or when extraction has to trigger further actions (filing, follow-ups, a CRM update), another route is to call an AI model or a cloud document-reading service directly, billed per page or per token.

Claude processes each PDF page twice: as text, typically 1,500 to 3,000 tokens per page, and as an image. The cost therefore depends on the model you pick and the number of pages. The 600-page limit per request drops to 100 pages on models with a context window under one million tokens. Per-model prices are in our guide to Claude API pricing.

These services read the document. You still have to build the confidence threshold, the human review step, the connection to your software and the audit trail. That is the job of a custom AI agent or an n8n workflow. If the goal is to answer questions from your documents rather than pull fields out of them, enterprise RAG is the right approach. To connect extraction to the rest of your stack, see also the best AI automation tools.

Document extraction carries a specific risk: the documents it processes contain personal data by nature. ID documents, payslips, medical records and proof of address are exactly what extraction is used to read.

Where data is hosted and who owns the vendor are two separate questions. Rossum hosts in Frankfurt by default and has belonged to Coupa since the May 2026 announcement. Klippa, originally Dutch, belongs to Germany's Doxis. Mindee is independent and based in Paris, but its EU processing zone is not offered on every plan. Ask both questions and get the answers confirmed in the data processing agreement.

An agent that reads documents from outside your company also raises its own security questions, covered in our guide to AI agent security.

When extraction has to fit into a wider business process (sorting files, follow-ups, CRM updates), the tool alone is not enough, and that is the work Lumyniq does. Transparency: we publish this site and are not in this comparison, because we do not sell document software.