Invoice Capture Agent
Read incoming invoices, verify mandatory fields, extract data - archived in GoBD-compliant format.
Reads incoming invoices (PDF, scan, XRechnung, ZUGFeRD), verifies mandatory fields per Paragraph 14 UStG, extracts structured data and archives in GoBD-compliant format. Reduces manual capture to exceptions only.
Score Dashboard
What This Agent Does
Invoice receipt is the highest-volume process in accounts payable. Dozens to thousands of invoices arrive daily - by email, portal, post, EDI. Each must be read, checked, extracted and archived. Today that means: manual review, data entry, duplicate check, filing.
The Decision Layer breaks invoice receipt into eight decision steps. The AI Agent classifies the document type and extracts data from unstructured documents. The rule engine verifies mandatory fields per Paragraph 14 UStG, detects duplicates, validates e-invoice formats and archives in GoBD-compliant format - XML and PDF, with timestamp, immutable.
The result: standard invoices are processed in seconds instead of minutes. The data capture error rate drops to near zero. And accounting only handles the exceptions that genuinely require human judgement.
Micro-Decision Table
Classify document type Invoice, credit note, reminder or quotation? AI Agent
LLM classification of unstructured inputs
Decision Record
Challengeable: Yes - fully documented, reviewable by humans, objection via formal process.
Verify mandatory fields Are all mandatory fields per Paragraph 14 UStG present? Rules Engine Vendor
Deterministic checklist against statutory mandatory fields
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Challengeable by: Vendor
Data extraction Invoice number, amount, currency, service period? AI Agent
LLM extraction from unstructured documents
Decision Record
Challengeable: Yes - fully documented, reviewable by humans, objection via formal process.
Duplicate check Does this invoice already exist in the system? Rules Engine
Database match on invoice number, vendor and amount
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Vendor matching (exact) Can the vendor be uniquely identified? Rules Engine
Exact match via VAT ID or vendor number
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Vendor matching (fuzzy) Which vendor matches when the name differs? AI Agent Vendor
Fuzzy match for name variants or new vendors
Decision Record
Challengeable: Yes - fully documented, reviewable by humans, objection via formal process.
Challengeable by: Vendor
E-invoice validation Is the XRechnung/ZUGFeRD format schema-compliant? Rules Engine
Schema validation against official specification
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
GoBD-compliant archiving Are XML and PDF stored immutably with timestamp? Rules Engine Auditor
Rule-based archiving per GoBD 2025
Decision Record
Challengeable: Yes - rule application verifiable. Objection possible for incorrect data or wrong rule version.
Challengeable by: Auditor
Decision Record and Right to Challenge
Every decision this agent makes or prepares is documented in a complete decision record. Affected parties (employees, suppliers, auditors) can review, understand, and challenge every individual decision.
Prerequisites
- ERP system with API access (SAP, DATEV, Oracle FI or equivalent)
- Inbound channel configuration (email, portal, EDI)
- Vendor master data with VAT ID
- GoBD-compliant document management system
Governance Notes
Not a high-risk system under the EU AI Act. GoBD relevance: high - invoice receipt is the starting point of every tax-relevant process. Paragraph 14 UStG defines the mandatory fields, which the agent checks deterministically. GoBD 2025 requires archiving both the XML and the PDF portion of e-invoices - PDF alone is no longer sufficient. Paragraph 203 StGB is relevant for client data of professional secrecy holders: LLM inference runs exclusively in EU data centres.
§203 StGB-relevant data is encrypted end-to-end and never passed to AI models in plain text.
Process Documentation Contribution
Infrastructure Contribution
The Invoice Capture Agent builds the foundational infrastructure for the entire AP chain. The document classification (invoice vs. credit note vs. reminder) is reused by the Credit Note Agent. The e-invoice validation forms the basis for the Invoice Generation Agent. The GoBD-compliant archiving pattern is used by all Finance agents that process tax-relevant documents.
Related Agents
Account Coding Agent
GL account, cost centre, tax code - automatically coded, with confidence score.
Three-Way Matching Agent
Purchase order, delivery note, invoice - automatically matched, discrepancies routed.
Payment Run Agent
Select due invoices, optimise early payment discounts, generate SEPA XML - with four-eyes approval.
Frequently Asked Questions
Which invoice formats are supported?
PDF (scanned and digital), XRechnung, ZUGFeRD, EDI. The AI Agent extracts from unstructured PDFs and scans. For e-invoices (XRechnung/ZUGFeRD), the structured XML data is processed directly and additionally validated against the schema.
What happens when mandatory fields are missing?
The agent documents exactly which mandatory fields per Paragraph 14 UStG are missing and routes the invoice to the clerk. The vendor can be asked to send a corrected invoice. The defective invoice is still archived in GoBD-compliant format.
How accurate is the automatic data extraction?
For structured e-invoices (XRechnung/ZUGFeRD), accuracy is 100% - the data is read directly from the XML. For unstructured PDFs and scans, the AI Agent typically achieves 95-98% accuracy. Every extraction includes a confidence score; below a defined threshold, manual review is triggered.
Implement This Agent?
We assess your finance process landscape and show how this agent fits your infrastructure.