OCR and document AI are the first step of almost every back-office automation. Invoices, receipts, purchase orders, IDs, bank statements and forms arrive as PDFs and photos, and nothing downstream can happen until their data becomes structured fields. This comparison covers the four services we see most often in document automation projects: Amazon Textract, Google Document AI, Azure AI Document Intelligence and Nanonets. Every price was read on the vendor's own pricing page on November 17, 2026.
In short. Azure AI Document Intelligence is our pick for most business document automation (8.3/10): 500 free pages a month, reading at $1.50 per 1,000 pages and prebuilt invoice, receipt, ID and contract models at $10 per 1,000 pages. Amazon Textract (8.1/10) is the natural choice on AWS, with plain text extraction at $0.0015 per page and forms and tables priced per feature. Google Document AI (7.9/10) fits Google Cloud teams, with Enterprise Document OCR at $1.50 per 1,000 pages and custom extractors at $30 per 1,000 pages. Nanonets (7.6/10) is the easiest no-code option, with $50 of free credits and workflows priced per block run, which Nanonets estimates at under $2 per invoice end to end.
OCR is not the hard part anymore
Reading characters from a clean PDF is a solved problem. The hard parts are what comes after: knowing which number is the invoice total, matching a supplier name to your records, handling a crooked photo of a receipt, and deciding what to do when the model is unsure. That is why modern services sell three tiers: plain OCR, prebuilt models for common documents, and custom models trained on your own documents.
The right choice depends less on raw accuracy than on three questions: which cloud you already use, how standard your documents are, and whether you have developers or want a no-code tool.
How we score: the PURIST Fit Score
| Criterion | What it measures | Weight |
|---|---|---|
| Extraction quality for business documents | Prebuilt models for invoices, receipts, IDs, forms and tables | 25% |
| Price at small and mid volume | Real cost for 1,000 to 50,000 pages a month | 25% |
| Ease of implementation | Time to a working extraction, with or without developers | 25% |
| Automation readiness | API quality, confidence scores, fit with workflow tools | 25% |
Source: compiled by PURIST (purist.online), November 17, 2026.
The four services side by side
| Service | Extraction for business documents | Price at small and mid volume | Ease of implementation | Automation readiness | PURIST Fit Score |
|---|---|---|---|---|---|
| Azure AI Document Intelligence | 9/10 | 8.5/10 | 7/10 | 8.5/10 | 8.3/10 |
| Amazon Textract | 8.5/10 | 8/10 | 7/10 | 9/10 | 8.1/10 |
| Google Document AI | 8.5/10 | 7.5/10 | 7/10 | 8.5/10 | 7.9/10 |
| Nanonets | 8/10 | 6.5/10 | 9.5/10 | 6.5/10 | 7.6/10 |
Source: compiled by PURIST (purist.online), November 17, 2026.
These are editorial scores based on published pricing and capabilities, not benchmark results. Accuracy depends heavily on your documents: test with 50 of your own before choosing.
Pricing at a glance (checked November 17, 2026)
| Service | Plain OCR | Prebuilt business models | Custom models | Free tier |
|---|---|---|---|---|
| Azure AI Document Intelligence | Read: $1.50 per 1,000 pages up to 1 million, then $0.60 | $10 per 1,000 pages (document, layout, receipt, invoice, ID, W-2, 1098, health insurance card, contract) | Custom extraction $30 per 1,000 pages; custom classification $3 per 1,000 pages | 500 pages per month |
| Amazon Textract | Detect Document Text: $0.0015 per page for the first million pages a month, then $0.0006 | Analyze Expense and Analyze ID APIs; Analyze Document priced per feature: tables $0.015 per page, forms $0.05 per page (then $0.04 after 1 million), queries $0.015 per page | Custom queries adapters | 3 months for new AWS customers, including 1,000 pages a month of Detect Document Text and 100 pages a month of Analyze Expense |
| Google Document AI | Enterprise Document OCR: free up to 1,000 pages, then $1.50 per 1,000 pages up to 5 million, then $0.60 | Specialized processors | Custom extractor $30 per 1,000 pages up to 1 million, then $20; form parser $30 per 1,000 pages | First 1,000 OCR pages |
| Nanonets | Included in workflow blocks | Included in workflow blocks | Included in workflow blocks | $50 in credits, no card required |
Source: compiled by PURIST (purist.online), November 17, 2026.
Nanonets prices workflows rather than pages: simple blocks cost $0.02 per run, standard AI blocks $0.10 and complex AI blocks such as data extraction $0.30, with a typical invoice workflow running 4 to 6 blocks. After the free credits, its page lists $100 per month for 100 credits, with volume discounts of up to 40%.
Azure AI Document Intelligence: the best prebuilt coverage for business documents
Microsoft's service offers the widest set of ready-made models for the documents small and mid-size businesses process every day, at a clear price.
What you pay. The free tier covers 500 pages a month. Reading costs $1.50 per 1,000 pages up to one million pages, then $0.60. All prebuilt models, including invoice, receipt, ID, W-2, 1098 tax forms, health insurance cards and contracts, cost $10 per 1,000 pages. Custom extraction and custom generative extraction cost $30 per 1,000 pages, custom classification $3, and query fields $10. Microsoft notes that these are estimates and actual pricing depends on your agreement and region.
Where it is strongest. Invoices and receipts out of the box, with no training, at one cent per page.
Where it is weaker. You need an Azure account and a developer, or an automation platform, to call the API.
Best for: accounts payable, expense and onboarding workflows, especially in Microsoft-centric businesses.
Amazon Textract: granular pricing and the AWS ecosystem
Textract offers five APIs: Detect Document Text for OCR, Analyze Document for forms, tables, queries and signatures, Analyze Expense for invoices and receipts, Analyze ID for identity documents, and Analyze Lending for mortgage documents.
What you pay. AWS's own examples put Detect Document Text at $0.0015 per page for the first million pages a month, then $0.0006. Analyze Document charges per feature: $0.015 per page for tables, $0.05 per page for forms (key-value pairs), dropping to $0.04 after one million pages, and $0.015 per page for queries. AWS's worked example of 5,000 tax form pages with tables and forms comes to $325. New AWS customers get a three-month free tier.
Where it is strongest. Fine control over cost, because you pay only for the features you call, and native fit with S3, Lambda and other AWS services.
Where it is weaker. Forms extraction at $0.05 per page adds up, and the per-feature model takes some planning to estimate.
Best for: teams already on AWS and high-volume pipelines built by developers.
Google Document AI: strong OCR and custom extractors
Google's service combines Enterprise Document OCR, specialized processors and custom extractors trained on your documents.
What you pay. Enterprise Document OCR is free up to 1,000 pages, then $1.50 per 1,000 pages up to five million pages, then $0.60, with lower rates on Google's one- and three-year savings plans. OCR add-ons cost $6 per 1,000 pages. The custom extractor costs $30 per 1,000 pages up to one million, then $20, and the form parser costs $30 per 1,000 pages.
Where it is strongest. High-quality OCR across languages and layouts, and custom extractors for documents no prebuilt model covers.
Where it is weaker. For standard invoices, Azure's prebuilt model at $10 per 1,000 pages is cheaper than a $30 custom extractor.
Best for: teams on Google Cloud and businesses with unusual document types.
Nanonets: no-code document workflows
Nanonets packages extraction, validation, formatting and export into workflows built without code, which is why non-technical finance and operations teams choose it.
What you pay. No platform fees or seat licenses. You start with $50 in credits, then $100 per month for 100 credits. Each workflow step is a block: $0.02 per run for simple operations such as formatting, routing and export, $0.10 for standard AI such as classification and validation, and $0.30 for complex AI such as data extraction. Nanonets estimates a typical invoice workflow at 4 to 6 blocks, under $2 per invoice end to end.
Where it is strongest. Time to value. A finance team can have invoices extracted and exported to accounting without a developer.
Where it is weaker. Per-document cost is higher than calling a cloud API directly, which matters at volume.
Best for: small finance and operations teams without developers.
Which one should you choose?
If you process standard invoices and receipts: Azure AI Document Intelligence prebuilt models.
If you are on AWS or need plain OCR at the lowest cost per page: Amazon Textract.
If you are on Google Cloud or have unusual documents: Google Document AI with a custom extractor.
If nobody on the team writes code: Nanonets.
How we build document extraction workflows
The extraction service is one step. A reliable workflow has five:
- 1 Intake. Documents arrive from an inbox, a form or a shared folder and are saved with a reference. Storage options are compared in our document management comparison.
- 2 Extraction. The service returns fields with confidence scores.
- 3 Validation. Totals are checked against line items, the supplier is matched to your records, and any field below a confidence threshold goes to a person for review. That human check is what makes the workflow safe.
- 4 Posting. Validated data creates the bill, expense or record in your accounting or ERP system. For bills, see our accounts payable automation comparison.
- 5 Exceptions report. Every week, documents that failed or needed review are summarized so the rules can improve.
Our guide to data entry automation with AI covers the business case, and the n8n HTTP Request node guide shows how to call these APIs from n8n.
Frequently Asked Questions
What is the best OCR software for business documents?
For standard business documents such as invoices and receipts, Azure AI Document Intelligence offers prebuilt models at $10 per 1,000 pages and 500 free pages a month. Amazon Textract is the best choice on AWS, Google Document AI on Google Cloud, and Nanonets if you want a no-code tool.
How much does Amazon Textract cost per page?
AWS's pricing examples list Detect Document Text at $0.0015 per page for the first million pages a month, and Analyze Document at $0.015 per page for tables, $0.05 per page for forms and $0.015 per page for queries, as checked on November 17, 2026.
Is Google Document AI free?
Enterprise Document OCR is free for the first 1,000 pages, then $1.50 per 1,000 pages up to five million pages. Custom extractors and the form parser cost $30 per 1,000 pages.
Can AI extract data from invoices automatically?
Yes. Azure's prebuilt invoice model, Textract's Analyze Expense API and Nanonets' extraction blocks all return invoice fields such as supplier, date, totals and line items. Add a validation step and a human review for low-confidence fields before posting to accounting.
Do I need a developer to use OCR APIs?
For Azure, AWS and Google, a developer or an automation platform such as n8n is needed to call the API. Nanonets is designed for teams without developers.
How accurate is document AI?
It depends on your documents. Clean digital PDFs extract very reliably; low-quality scans and photos less so. Test each service on at least 50 of your real documents and use confidence scores to route uncertain fields to a person.
Sources
Prices were read on the vendors' official pages on November 17, 2026: Azure AI Document Intelligence pricing, Amazon Textract pricing, Google Document AI pricing, Nanonets pricing. Prices change and vary by region: check the vendor's page before you buy.
Want invoices and forms read, checked and posted without typing? Ask for a free written automation audit and we will design the extraction workflow for your documents, in writing.
Tags
Steve founded PURIST, a workflow automation studio working with n8n, Make and Claude AI, and is the point of contact on every project. Prices and features in our articles come from vendor documentation and public pricing pages. How we research and score tools.