Concept card

SAP Document AI

As of 2026-09-23

Edited by Cédric-Olivier Mâry, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is SAP Document AI?

SAP Document AI (formerly Document Information Extraction) turns PDFs and images into validated JSON fields via pre-trained models, templates or LLM-based schema extraction — and its edition, not the demo, decides which of those you actually get.

SAP Document AI is SAP's managed service on SAP Business Technology Platform (BTP) for extracting structured data from business documents — invoices, purchase orders, payment advices, delivery notes, bank statements, certificates and any custom document type you define. It was called Document Information Extraction (often abbreviated DOX) until SAP's What's New entry of 15 July 2025 announced the rebranding to SAP Document AI. The rename was cosmetic at the technical level: the BTP marketplace tile is still sap-document-information-extraction, the subscription entitlement is still document-information-extraction-application, and the REST API base path is still /document-information-extraction/v1. Expect both names in tickets, SAP Notes and customer landscapes for years.

What problem it solves

Most document-centric processes in an SAP landscape — accounts payable, sales order intake, HR onboarding, supplier certificates — still start with a human reading a PDF and keying fields into S/4HANA, SAP Ariba or a workflow form. Document AI takes the file (PDF, JPEG, PNG, TIFF; XLS/XLSX for payment advices; hybrid PDF/XML e-invoices in Factur-X and ZUGFeRD up to version 2.3), runs OCR, and returns header fields and line items as JSON with a confidence score per field, which a workflow, an integration flow or an agent can then validate and post. It can also enrich the result by matching the document to master-data records you upload (for example, returning the supplier ID that matches the extracted vendor name and address).

The two product families — know which one you are in

The service exposes two distinct surfaces, and confusing them is the first scoping mistake:

Why it matters

It is the renamed Document Information Extraction (15 July 2025): same service keys, same REST path — scoping documents and SAP Notes will use both names.
Edition choice decides capability: generative extraction, instant learning, workflows and channels are not in the base edition.
Premium metering counts pages and fields, not files — a one-page document with more than 50 extracted fields is billed as more than one document.
Since July 2026 the Tools Service API exposes classify/extract/answer/summarize to agents, which puts Document AI inside Joule and agentic designs, not only in BPA flows.

Key points

Renamed from Document Information Extraction on 15 July 2025; technical names (sap-document-information-extraction, /document-information-extraction/v1) are unchanged.
Two surfaces: basic UI + REST (base, embedded, premium) and workspace + OData V4 (embedded and premium only).
Three methods: SAP pre-trained ML, templates, and generative extraction from a schema with processing instructions.
Plans: free/trial (50 pages per 30 days), base (blocks of 100 docs + compute hours), premium (docs, requests, users), embedded (AI units).
2026 additions: confidence ranges, Schema Studio, AI-generated schemas, Tools Service API for agents, retry, OData V4 page ranges.
SBPA bundles base-edition extraction; premium is called through a BTP destination and a reusable 'DocAI – Extract with Schema' package.

Terms used on this page

Schema: Definition of the header fields and line-item fields to extract for a document type, with data types, optional descriptions and processing instructions.
Content schema: SAP-delivered standard schema in the workspace (e.g. BUSINESS_CARD_STANDARD, OCR_STANDARD) that you enable and optionally adapt.
Template: Layout-anchored extraction configuration built from a sample document; on the base edition its use is metered in compute hours.
Instant learning: Embedded/premium feature that learns from user corrections to improve subsequent extractions; not intended for Art. 9 GDPR data.
Data feedback collection: Opt-in feature that lets confirmed documents be used by SAP to retrain its extraction models.
Tools Service API: API (July 2026) that lets agents classify documents, extract information, answer questions and summarize text.
Enrichment: Matching an extracted document to uploaded master-data records and returning the matching record ID.

Sources

Full card available to members. What the full card adds: the full decision framework · the common pitfalls and their fix · the cheat sheet · the code blocks · the facts worth quoting.

Open in the app →