{
  "data": null,
  "kind": "anchor.page",
  "links": {
    "api": "https://www.anchorterminal.com/api/v1/index.json",
    "html": "https://www.anchorterminal.com/niches/document-extraction.fields-from-business-docs/",
    "json": "https://www.anchorterminal.com/niches/document-extraction.fields-from-business-docs/index.json",
    "llms": "https://www.anchorterminal.com/llms.txt",
    "markdown": "https://www.anchorterminal.com/niches/document-extraction.fields-from-business-docs/index.md",
    "mcp": "https://www.anchorterminal.com/mcp",
    "slim": "https://www.anchorterminal.com/niches/document-extraction.fields-from-business-docs/index.min.md"
  },
  "markdown": "Four or more companies' docs show a full fit.\n\n| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |\n| --- | --- | --- | --- | --- |\n| 5 | 12 | 36 | 16 | 10 |\n\n## What counts as done\n\nJSON matching the given schema or the documented prebuilt field list, with line items as arrays, numbers as numbers, a confidence per field and nulls where a field is absent, from one call or one async job.\n\nMeasured, for our own tests: field-level accuracy on a fixed set of invoices and receipts.\n\nHow agents ask for it: get the total, VAT and supplier from this invoice; extract line items from these receipts; fill this JSON schema from the document; parse this CV into fields\n\n## Does the job\n\nThe docs show it meets the test above, and nothing we read says otherwise.\n\n- **Azure Document Intelligence (Microsoft Corporation)**: Prebuilt invoice, receipt, ID, contract and US tax models return documented fields with confidence; there is no JSON schema input, but the success criterion allows a prebuilt field list. acc-12 reports it working for a year after tuning.\n- **Extend API + MCP (CrowdView Inc. dba Extend)**: Extract fills a JSON schema and returns output.value shaped like the schema with per-field ocrConfidence, logprobsConfidence and citations; sync or async runs.\n- **Google Cloud Document AI (Google LLC)**: Invoice, expense, identity, W2 and other prebuilt processors return typed entities with confidence, normalizedValue and properties for line items, and a Custom Extractor takes a schema (English only for generative extraction).\n- **LlamaParse API + MCP (LlamaIndex, Inc.)**: Extract takes a JSON Schema or Pydantic model up to 3,200 fields, supports nullable fields and arrays, and returns confidence scores and citations when confidence_scores and cite_sources are set.\n- **Nanonets API + MCP (Nano Net Technologies Inc.)**: json_options takes a JSON schema or field list, field specs type string, number, boolean, date or array, and metadata returns per-field confidence plus an overall score penalised for truncation.\n\n## Does the job, with a limit\n\nIt does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.\n\n- **Amazon Textract (Amazon Web Services, Inc.)**: AnalyzeExpense returns invoice and receipt fields and line items with Confidence, and AnalyzeID reads US identity documents, but there is no JSON schema input, AnalyzeExpense is English only, and IDs are US only.\n- **Clairic (Cloudflare, Inc.)**: Typed fields come back with numbers as numbers, confidence, p_null and null when absent, but list items can only be text, number or date, so line items as records are not possible; inputs are text or four images, and the model is a preview.\n- **Datalab (Endless Labs, Inc.)**: page_schema takes a JSON schema with numbers and arrays, up to 750 fields, and returns per-field citations and verification status, but no numeric confidence per field is documented and there are no prebuilt models.\n- **Docling (Docling Project (LF AI \u0026 Data Foundation))**: Information extraction takes a string, dict or Pydantic template and fills it with a VLM, but there are no prebuilt models and no per-field confidence; confidence is graded per page.\n- **Invofox (Invofox, Inc. (Delaware), with a registered branch in Spain)**: Typed JSON for invoices, receipts, payslips, ID documents and custom types, with per-field ERR_LOW_CONFIDENCE errors, but confidence is high, medium or low per document, types are set up in the web app rather than passed as a schema, and languages are Latin-script.\n- **LandingAI Agentic Document Extraction (LandingAI, Inc.)**: Extract fills a JSON schema, returns null for absent values and flags partial results with warnings, schema_violation_error and HTTP 206, but no per-field confidence is returned and there are no prebuilt models.\n- **Mindee API (Mindee SAS (Mindee Inc. outside the Eurozone))**: Templates for invoice, receipt, resume, payslip and IDs return typed fields with list fields and Certain, High, Medium or Low confidence, but every model must be created in the web platform first, and acc-46 saw new None values crash the Python client.\n- **Mistral OCR API (Mistral AI (RCS Paris 952 418 325))**: Document annotations fill a JSON schema at $5 per 1,000 pages, but acc-38 saw the model loop on number fields and truncate the JSON, and no per-field confidence is returned.\n- **Reducto API + MCP (Reducto, Inc.)**: Extract fills a JSON schema with a documented list of guaranteed keywords and returns citations and usage, but per-field confidence is not shown in the extract response format read, and there are no prebuilt models.\n- **Unstructured API + MCP (Unstructured Technologies, Inc.)**: Extract takes a JSON Schema and returns data in its shape, nulls included, with optional citation metadata, but no per-field confidence; the docs warn not to read missing evidence as confidence.\n- **Veryfi API + MCP (Veryfi, Inc.)**: process_document returns receipt and invoice fields with line_items, and dedicated calls cover W-2, W-8, W-9 and cheques, but per-field confidence and supported countries could not be read, and acc-42 found the MCP server failing until token construction was fixed.\n- **Xiaomi-OCR-0 (Xiaomi Corporation)**: Key information extraction takes a JSON schema and returns a JSON object, but the schema is a template of empty strings with no numeric types documented, and there is no confidence or null signal.\n\n- not this job: Adobe PDF Services / PDF Extract API (Adobe Inc.): No schema parameter and no prebuilt invoice, receipt or CV model in the API; Extract returns layout elements only.\n- not this job: Nemotron 3.5 Parse (NVIDIA Corporation): A layout parser with no schema or field extraction mode.\n- not this job: OpenDocRouter (LlamaIndex, Inc.): Parse only; no schema or field extraction endpoint.\n\n6 listings in the category were added after this reading and aren't marked yet.\n\n## What stops agents finishing\n\n- strict schema output loops or truncates on number fields\n- client library breaks on a response change and stops the pipeline\n- prebuilt models only for US documents or English\n\n## Around it\n\nBefore: mailbox-access.find-and-read, file-storage.agent-upload. After: accounting.capture-bill, crm.upsert-contact, human-in-the-loop.review-output.\n\n## Where this comes from\n\nWeight 16 of 100 in Document parsing \u0026 extraction: 10 of 46 accounts; invoice and receipt pipelines on Hacker News plus Mistral, Mindee, Veryfi and Docling issues\n\nAs data: https://www.anchorterminal.com/api/v1/use-cases/document-extraction.fields-from-business-docs.json, or the MCP tool get_use_case with id document-extraction.fields-from-business-docs. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.\n",
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "attributionRequired": true,
    "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-11-16",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.8",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run",
    "terms": "https://www.anchorterminal.com/terms/#data"
  },
  "page": {
    "breadcrumbs": [
      {
        "name": "Home",
        "url": "https://www.anchorterminal.com/"
      },
      {
        "name": "For agents",
        "url": "https://www.anchorterminal.com/agents/"
      },
      {
        "name": "Niches",
        "url": "https://www.anchorterminal.com/niches/"
      },
      {
        "name": "Pull named fields from invoices, receipts, forms and CVs",
        "url": ""
      }
    ],
    "description": "Pull named fields from invoices, receipts, forms and CVs: 5 companies fully do it, 12 with a limit, 36 carry the tag. Every listing in Document parsing \u0026 extraction with the reason and a source.",
    "facts": null,
    "h1": "Pull named fields from invoices, receipts, forms and CVs",
    "image": "https://www.anchorterminal.com/assets/og/default.png",
    "path": "/niches/document-extraction.fields-from-business-docs/",
    "published": "",
    "section": "agents",
    "title": "Pull named fields from invoices, receipts, forms and CVs | Anchor…",
    "toc": null,
    "updated": "2026-11-15",
    "url": "https://www.anchorterminal.com/niches/document-extraction.fields-from-business-docs/"
  },
  "tokens": {
    "markdown": 1900,
    "slim": 430
  },
  "version": 1
}
