CIOPages

CIOPages · State of Play · Intelligent Document Processing

State of Play 2026: Intelligent Document Processing

Google, Microsoft, AWS, ABBYY and UiPath now use generative models to extract document data, Coupa bought Rossum, and France's e-invoicing mandate took effect. Where the market stands in 2026, and what each kind of product costs where prices are public.

Published November 20, 2026

Intelligent document processing (IDP) software turns documents such as invoices, forms, claims and contracts into structured data that other systems can use. It reads the text with optical character recognition (OCR), classifies each document, extracts the fields a process needs and validates them. Newer products use large language models and vision-language models to extract fields from layouts they were not trained on.

Where the market stands

Microsoft renamed Azure Form Recognizer to Azure AI Document Intelligence in July 2023. The service is now called Azure Document Intelligence in Foundry Tools. Kofax renamed itself Tungsten Automation on January 16, 2024.

France's e-invoicing reform took effect on September 1, 2026. From that date, every French business must receive invoices in electronic form when the supplier is required to issue them, and large and mid-sized companies must issue them. Small and micro businesses must issue them from September 1, 2027. The French tax authority says an image PDF sent by email is not an electronic invoice. Invoices between businesses must pass through an approved platform.

The EU adopted its VAT in the Digital Age (ViDA) package on March 11, 2025. Since April 14, 2025, member states can introduce mandatory e-invoicing under specific conditions. Digital reporting requirements apply to cross-border business-to-business transactions from July 1, 2030.

An invoice that arrives as a structured e-invoice needs no extraction. Estimate how much of your invoice volume these mandates will move to structured formats before sizing an invoice-capture purchase.

What changed at the vendors

  • AWS. Amazon Bedrock Data Automation became generally available on March 3, 2025. AWS says it uses generative AI to turn documents, images, video and audio into structured formats, and returns confidence scores and visual grounding.
  • Google. A Document AI Custom Extractor version powered by Gemini 2.5 Pro became generally available on September 3, 2025. Versions powered by Gemini 3 Pro, Gemini 3 Flash, Gemini 3.5 Flash and Gemini 3.1 Flash Lite entered preview between January and September 2026. In February 2026, Google said its legacy Document AI processors would be discontinued on June 30, 2026.
  • Microsoft. Azure Content Understanding in Foundry Tools reached general availability with API version 2025-11-01. It uses generative AI and covers documents, images, audio and video. Microsoft says Document Intelligence customers need no migration.
  • ABBYY. Vantage 3.0 launched on January 20, 2026, with direct integration with large language models. It connects to Azure OpenAI for prompt-based extraction. Customers choose whether the model receives the document image or ABBYY's OCR text.
  • UiPath. On April 22, 2026, UiPath made IXP (Intelligent Xtraction and Processing) available on Google Cloud Marketplace. It said Gemini will become the default third-party model for new IXP projects.
  • Coupa. On May 12, 2026, Coupa announced it had acquired Rossum. The two companies had already partnered on invoice automation for accounts payable teams.

Five kinds of product

The products fall into five kinds. One invoice process can use more than one.

  • Cloud document services. Azure Document Intelligence and Azure Content Understanding, Google Document AI, Amazon Textract and Amazon Bedrock Data Automation. They are APIs billed through the cloud account. Microsoft describes Document Intelligence as ideal for structured documents with common templates, and Content Understanding as suited to unstructured documents and varying layouts.
  • Specialist IDP platforms. ABBYY Vantage and Tungsten TotalAgility. Tungsten says TotalAgility ingests, classifies, extracts and validates documents, coordinates people, processes and AI agents, and deploys in a public or private cloud or on premises.
  • Model-first platforms. Hyperscience Hypercell includes ORCA (Optical Reasoning and Cognition Agent), its vision language model framework. Hyperscience's Fall 2026 release, announced September 29, 2026, added ORCA 2 with zero-shot table extraction and combines specialized and frontier models. Rossum's extraction runs on a transactional large language model trained on tens of millions of documents. Coupa says it keeps learning from each customer's documents.
  • IDP inside automation suites. UiPath IXP combines document understanding with communications mining inside the UiPath automation platform.
  • Building on foundation models. Teams can prompt a large language model directly. Microsoft's comparison notes that this route returns no confidence scores or grounding unless the team builds them.

What each costs where prices are public

Figures are list prices as published on October 17, 2026. Microsoft, Google and AWS state theirs in USD. Azure and AWS figures are for the US East region. Rossum shows a dollar sign without naming a currency.

  • Azure Document Intelligence. Read costs $1.50 per 1,000 pages, falling to $0.60 above 1 million pages. Prebuilt and layout models cost $10 per 1,000 pages. Custom extraction costs $30 per 1,000 pages and custom classification $3. Monthly commitment tiers cost less per page.
  • Azure Content Understanding. Asynchronous content extraction costs $1 per 1,000 pages for OCR, $5 for layout and $0.01 for digital files that need neither. Generative features add contextualization at $1 per 1,000 pages, or $3 for advanced contextualization. Model tokens are billed separately on the customer's Microsoft Foundry deployment.
  • Google Document AI. Enterprise Document OCR costs $1.50 per 1,000 pages, falling to $0.60 above 5 million. Custom Extractor and Form Parser cost $30 per 1,000 pages, falling to $20 above 1 million. The Custom Extractor price is the same for generative and custom-model versions. Layout Parser costs $10 per 1,000 pages. The invoice parser costs $0.10 for each block of up to 10 pages in a document. Each deployed custom processor version also costs $0.05 an hour to host.
  • Amazon Textract. Text detection costs $1.50 per 1,000 pages for the first million pages in a month and $0.60 after that. Over the same first million pages, Forms extraction costs $50 per 1,000 pages, and Tables and Queries $15 each.
  • Amazon Bedrock Data Automation. Standard output for documents costs $0.01 per page. Custom output costs $0.04 per page for blueprints of up to 30 fields, plus $0.0005 per page for each field beyond 30.
  • Rossum. The Starter plan starts at $18,000 a year. Business, Enterprise and Ultimate plans are quoted. Rossum prices on page or document volume and workflow complexity, with a one-year minimum contract.

Per-page prices leave out the cost of people reviewing exceptions. Model the cost per document that completes with no human touch, including review time, before comparing quotes.

What a deployment requires

Check where documents are processed. Google says some Gemini-powered Document AI versions use a global endpoint and are not compliant with data residency standards. Requests sent to their US and EU endpoints might be routed anywhere in the world. Among its document tools, Microsoft lists Document Intelligence containers as the only option today for on-premises or air-gapped deployment.

Plan for model versions to change. Document Intelligence API v2.1 retires on September 15, 2027, and v3.0 on March 30, 2029. Google said Custom Extractor version pretrained-foundation-model-v1.4-2025-02-05 would no longer be accessible from February 5, 2026. Its upgrade path for a fine-tuned extractor, in preview since March 31, 2026, fine-tunes a new processor version on a newer base version.

Test the confidence scores, not only the accuracy. Build an answer key for a sample of your worst documents: poor scans, unseen layouts, handwriting and other languages. Measure field-level accuracy, and check whether low-confidence flags catch the wrong values before they reach the ERP.

Start with one high-volume document type. Set confidence thresholds and the review queue, connect the output to the system that consumes it, and add document types once that loop is stable.

Questions to settle first

Settle these before choosing a product.

  • Which document types and volumes will an e-invoicing mandate turn into structured data within the contract term?
  • Who will build and run the review screen, exception routing and accuracy monitoring: your team or the vendor?
  • Where may documents and prompts be processed, and does any document type require on-premises processing?
  • What does the model version roadmap look like, and who retests accuracy when a version is retired?
  • What share of documents must pass with no human touch, at what error rate, before the pilot counts as a success?
Sources. Microsoft Learn, What is Azure Document Intelligence in Foundry Tools?: https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/overview; Microsoft Learn, Azure Form Recognizer client library for .NET: https://learn.microsoft.com/dotnet/api/overview/azure/AI.FormRecognizer-readme; Microsoft Learn, What is Azure Content Understanding in Foundry Tools?: https://learn.microsoft.com/en-us/azure/ai-services/content-understanding/overview; Microsoft Learn, Choose the right Azure AI tool for document processing: https://learn.microsoft.com/en-us/azure/ai-services/content-understanding/choosing-right-ai-tool; Microsoft Learn, Pricing for Azure Content Understanding in Foundry Tools: https://learn.microsoft.com/en-us/azure/ai-services/content-understanding/pricing-explainer; Microsoft, Azure Retail Prices API, Azure Content Understanding meters: https://prices.azure.com/api/retail/prices?$filter=contains(productName,%27Content%20Understanding%27)%20and%20armRegionName%20eq%20%27eastus%27; Microsoft, Azure Retail Prices API, Azure Document Intelligence meters: https://prices.azure.com/api/retail/prices?$filter=contains(productName,%27Document%20Intelligence%27)%20and%20armRegionName%20eq%20%27eastus%27; Tungsten Automation, "Kofax is now Tungsten Automation", January 16, 2024: https://www.tungstenautomation.com/press-releases/2024/kofax-is-now-tungsten-automation; Tungsten Automation, TotalAgility: https://www.tungstenautomation.com/products/totalagility; impots.gouv.fr, "À partir de quand suis-je concerné par la réforme de la facturation électronique ?": https://www.impots.gouv.fr/professionnel/questions/partir-de-quand-suis-je-concerne-par-la-reforme-de-la-facturation; impots.gouv.fr, "Est-ce qu'une facture envoyée par mail est une facture électronique ?": https://www.impots.gouv.fr/professionnel/questions/est-ce-quune-facture-envoyee-par-mail-est-une-facture-electronique; impots.gouv.fr, Je passe à la facturation électronique: https://www.impots.gouv.fr/professionnel/je-passe-la-facturation-electronique; European Commission, VAT in the Digital Age (ViDA): https://taxation-customs.ec.europa.eu/taxation/vat/vat-digital-age-vida_en; AWS, "Amazon Bedrock Data Automation is now generally available", March 3, 2025: https://aws.amazon.com/about-aws/whats-new/2025/03/amazon-bedrock-data-automation-generally-available; AWS, Amazon Bedrock User Guide, Bedrock Data Automation: https://docs.aws.amazon.com/bedrock/latest/userguide/bda.html; AWS, Amazon Bedrock pricing: https://aws.amazon.com/bedrock/pricing/; AWS, Price List API, Amazon Bedrock offer file (US East): https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/current/us-east-1/index.json; AWS, Amazon Textract pricing: https://aws.amazon.com/textract/pricing/; AWS, Price List API, Amazon Textract offer file (US East): https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonTextract/current/us-east-1/index.json; Google Cloud, Document AI release notes: https://docs.cloud.google.com/document-ai/docs/release-notes; Google Cloud, Document AI pricing: https://cloud.google.com/document-ai/pricing; Google Cloud, Document AI product page, pricing section: https://cloud.google.com/document-ai; ABBYY, "ABBYY Launches Vantage 3.0 Directly Integrating with Generative AI", January 20, 2026: https://www.abbyy.com/company/news/abbyy-launches-vantage-3-genai-integration/; UiPath, "UiPath Brings its AI Document Processing Solution to Google Cloud Marketplace with Gemini-Powered Automation", April 22, 2026: https://www.uipath.com/newsroom/uipath-ixp-gemini-automation-google-next; Coupa, "Coupa Acquires Rossum to Accelerate End-to-End Autonomous Spend Management", May 12, 2026 (as published by Thoma Bravo): https://www.thomabravo.com/press-releases/coupa-acquires-rossum-to-accelerate-end-to-end-autonomous-spend-management; Rossum, Pricing: https://rossum.ai/pricing/; Hyperscience, "Hyperscience Advances Hypercell to Unlock Enterprise Data for the Agentic AI Era", September 29, 2026: https://www.hyperscience.ai/newsroom/hyperscience-advances-hypercell-to-unlock-enterprise-data-for-the-agentic-ai-era/; Hyperscience, "Hyperscience Unveils Winter 2025 Release, Advancing Agentic Document Automation for the Enterprise", October 15, 2025: https://www.hyperscience.ai/?p=13335

Ask to contribute

Vendors in this market can ask to answer the six questions for State of Play 2026: Intelligent Document Processing. Use an email address on your company’s own domain: answers come from the company itself, not from an agency. We email a link to confirm the request.

Keep reading

Buyer Guide
Intelligent Document Processing (IDP)
Worksheet
Project Charter Worksheet

Answer guided questions and download a complete, formatted Project Charter document.

Job 04
Control technology spend

AI is billed per use, so its cost grows with adoption while the rest of the budget is under review.

State of Play 2026: Low-Code and No-Code Platforms