E-Rechnung 2025 with AI: How the German Mittelstand Meets the Mandate and Makes More of It

Since 1 January 2025 every German B2B company must be technically able to receive structured E-Rechnungen and process them in a GoBD-compliant manner. There is no transition period on the receiving side. The transition is only on the sending side, where the deadline for companies above EUR 800,000 prior-year revenue lands on 1 January 2027¹.

Most Mittelstand companies treated the receiving deadline as a tick-box exercise: install a tool, archive the XML, move on. The companies that quietly took the second step (let AI do the booking suggestion, the format conversion, and the GoBD archival routing) are now processing invoices at one fifth of the cost and one fifth of the time, while the rest are still losing two AP staff per quarter to the workload.

This guide is for the CFO, head of finance, or Geschaeftsfuehrer at a German Mittelstand company who has already met the receiving requirement and wants to turn the next 18 months of mandate ramp-up into a margin lever rather than a compliance bill.

Where the E-Rechnung Mandate Stands in Late 2026

The Wachstumschancengesetz introduced obligatory E-Rechnung for domestic German B2B transactions from 1 January 2025, with a tiered transition period for the sending side¹. Knowing exactly which deadline applies to which entity is the foundation for any AI investment that follows.

• 1 January 2025 - Receiving obligation applies in full to every German B2B company, with no transition period. Recipients must be able to accept and GoBD-archive structured E-Rechnungen¹.
• 1 January 2025 to 31 December 2026 - Issuing transition period. Senders may still use legacy paper or PDF for outbound invoices.
• 1 January 2027 - Issuing obligation begins for companies with prior-year revenue above EUR 800,000¹.
• 31 December 2027 - End of all transition periods. Companies up to EUR 800,000 must also issue E-Rechnungen from 2028.
• Kleinunternehmer (Section 19 UStG) - Exempt from the issuing requirement entirely¹.
• Public sector - Already obligated since 2020 under the e-invoicing directive, often via the OZG-XRechnung path.

XRechnung, ZUGFeRD, and EN 16931 in Plain Language

The German mandate accepts E-Rechnungen that conform to the European EN 16931 standard, which in practice means XRechnung, ZUGFeRD from version 2.0.1 (excluding the Minimum and Basic-WL profiles), and equivalent formats issued by other EU member states¹.

XRechnung

• Pure XML - No PDF, no human-readable layer in the file itself. The viewer renders the XML for the human reader.
• Public-sector heritage - Originated as the obligatory format for invoicing the German federal administration; now standard for B2B too.
• Validation strict - KoSIT publishes the official validator and a tightly defined business-rule set.
• Ideal for - High-volume, automated processing where humans rarely look at the invoice directly.

ZUGFeRD 2.0.1+ (Hybrid)

• Hybrid PDF/A-3 + XML - The XML is embedded in a PDF/A-3 container that displays as a regular PDF for humans.
• Profiles - EN 16931 (default), Extended, XRechnung. The Minimum and Basic-WL profiles do NOT meet the German mandate.
• Best of both - Machine-readable structure plus the visual layer that AP staff and approvers expect.
• Ideal for - Mid-market companies where AP, approvers, and Steuerberater all want to see the invoice as a PDF.

EN 16931 across borders

• EU-wide standard - The same data model underlies XRechnung (DE), Factur-X (FR), FatturaPA (IT), KSeF (PL).
• PEPPOL - The cross-border network for exchanging EN 16931 invoices, increasingly used by EU exporters.
• Practical implication - An AI pipeline built around EN 16931 handles incoming invoices from suppliers in any EU country with the same logic.
• Watch out for - Country-specific extensions and profiles that are valid only in their member state.

GoBD Archival: What the BMF Actually Requires

The 14 July 2025 BMF letter updated GoBD to reflect the E-Rechnung requirements, with one core message: the structured XML part must be archived in its original electronic form, unchanged, machine-readable, retrievable, and GoBD-compliant for the full eight years².

What GoBD-compliant archival means

• Original-form preservation - The XML as received, byte-identical, with no manual modification.
• Unalterability - Storage that prevents any change after archive entry, with cryptographic or filesystem-level controls.
• Retrievability - Any invoice from the last eight years must be retrievable within the audit window the tax authorities define.
• Machine readability - The XML must remain parseable for the full retention period, not just visualised.
• Audit trail - Every read, every export, every retention-period extension logged.
• Visualisation alongside - Where the format is hybrid, store both the XML and the PDF rendering; for pure XRechnung, store the XML and a deterministic visualisation.

The mistake most Mittelstand companies make

The single most common GoBD failure pattern is "we print the invoice and store the PDF". The BMF letter is explicit: printing is not sufficient. The original XML must be preserved. Companies that built their archive around a PDF-only routine in 2024 are now reworking the archive in 2026 to add the XML side.

The AI Invoice Processing Pipeline End to End

An AI-augmented E-Rechnung pipeline has six clean stages, each of which can be measured and improved in isolation. The structured part of the E-Rechnung itself does not require AI to read; AI matters most for the legacy formats, the booking suggestion, and the exception path.

1. Channel intake

• Email mailbox - The dominant intake path; AI classifies attachments by format (XRechnung, ZUGFeRD, PDF, image, other).
• PEPPOL or supplier portal - Direct structured intake for suppliers connected over network channels.
• EDI - For long-standing supplier relationships that already exchanged EDIFACT or similar.
• Paper - Scanned and routed through an OCR-then-IDP path during the transition.

2. Format detection and routing

Each incoming invoice is classified as native E-Rechnung, hybrid (PDF with embedded XML), legacy PDF, or paper. The structured cases bypass AI extraction entirely; only the legacy and paper paths need an AI extraction step.

3. Validation

For native and hybrid E-Rechnungen, validation against the EN 16931 schema and the German business rules (KoSIT validator) is a deterministic step that gates the flow. Invalid invoices route to an exception queue with the validator output attached for the supplier-relationship team.

4. Booking suggestion

The AI booking suggestion reads the structured invoice data, the chart of accounts, the supplier history, and the pending purchase orders, and generates a booking proposal for DATEV or the relevant accounting system. This is the highest-value AI step in the pipeline. DATEV reports its KI-Buchungsservice nearly tripled revenue in 2025⁹.

5. Approval routing

• Auto-approval - For matched POs and trusted suppliers below threshold, the booking goes straight to posting.
• One-click approval - For amounts above auto-approval threshold but below the formal approval limit, the AI prepares the case for one-click human sign-off.
• Full approval workflow - For high-value or unmatched cases, the standard multi-step workflow runs.
• Exception - Validation failures, supplier disputes, and policy breaches route to a human reviewer with full context.

6. GoBD-compliant archival

The archival step preserves the original XML in immutable storage, indexes the invoice in DATEV or the accounting system, and writes the audit log. This is the step the BMF audit will examine first.

6 High-ROI AI Use Cases on Top of the Mandate

The mandate forces every German B2B company to upgrade their invoice receiving infrastructure. The companies that build six AI use cases on top of that infrastructure recover the implementation cost within a year and unlock benefits the mandate alone never delivers.

1. Booking suggestion against the chart of accounts

• Inputs - Structured E-Rechnung data, chart of accounts, supplier history, prior bookings for similar invoices.
• Action - Generate a booking proposal with confidence score for review or auto-posting.
• Realistic impact - 60-80% of invoices booked without manual entry; KPMG reports nearly half of companies with AI invoice processing achieve average ROI above 40 percent¹³.

2. Legacy format extraction during the transition

• Inputs - PDF, image, or scanned paper invoices that arrive during the 2025-2027 transition.
• Action - AI extracts the structured fields and converts the invoice into the same internal format as native E-Rechnungen.
• Realistic impact - Extends the AI pipeline to 100 percent of inbound invoices, not just the structured share.

3. PO matching and three-way verification

• Inputs - Invoice, purchase order, goods receipt, contract terms.
• Action - AI matches the three documents, flags discrepancies, and proposes resolution paths.
• Realistic impact - 70-85 percent of PO-linked invoices auto-matched; manual exception handling drops by 50 percent.

4. Duplicate and fraud detection

• Inputs - Invoice number, supplier, amount, date pattern, IBAN, behavioural signals across history.
• Action - AI flags duplicate invoices, suspicious IBAN changes, anomalous amounts, and known fraud patterns.
• Realistic impact - Detection of issues that manual review consistently misses, particularly mid-volume duplicates from ERP migrations.

5. Exception triage and supplier communication

• Inputs - Validation failures, missing data, format issues, supplier-side errors.
• Action - AI drafts the supplier email asking for the corrected invoice, including the specific validator output and the format requirement.
• Realistic impact - Supplier-error resolution time cut from days to hours; AP team focuses on substantive disputes only.

6. Cash-flow reporting and forecast

• Inputs - All invoice data, payment terms, historical payment behaviour, supplier criticality.
• Action - Real-time cash-flow projection that updates as invoices arrive, plus payment-priority recommendation.
• Realistic impact - The CFO dashboard moves from monthly to weekly; payment timing optimised for early-payment discounts and supplier relationships.

DATEV, SAP, Lexware, Sage: Integration That Holds

The accounting platform you already use determines the cleanest integration path. The Mittelstand reality is that 70 to 80 percent of companies run DATEV or a DATEV-tax-advisor relationship, with SAP, Lexware, Sage, and Microsoft Dynamics covering the rest⁹.

DATEV-centric setups

• DATEV E-Rechnung Online - DATEV's native receiving and sending product, suitable as the channel layer.
• DATEV Automatisierungsservice Rechnungen - The AI booking-suggestion service, DATEV's strongest 2025 AI growth driver⁹.
• DATEV Unternehmen Online (DUO) - Cloud-based document and approval flow that extends naturally for AI agent integration.
• DATEV Connect online - The API layer that lets a custom AI agent read and write into DATEV without screen scraping.

SAP-centric setups (S/4HANA, ECC)

SAP customers typically deploy SAP Document and Reporting Compliance for the receiving and validation layer, plus the AI agent on top through standard BAPI or OData APIs. The pattern from our SAP and AI agents guide applies directly to invoice processing.

Lexware, Sage, Microsoft Dynamics

These platforms cover much of the smaller-Mittelstand segment. Each has a native E-Rechnung receiving option (Lexware Office, Sage E-Invoicing, Dynamics 365 Business Central) and an API or partner-app layer for the AI agent integration. The architectural pattern is identical to DATEV: keep the platform as the system of record, layer the AI for booking suggestion and exception handling on top.

The 90-Day E-Rechnung + AI Pilot

A 90-day pilot takes one company entity, one major incoming invoice channel, and the booking-suggestion use case from current state to production. The shape that works in the Mittelstand keeps DATEV or your existing accounting platform as the system of record.

Phase 1: Audit and Baseline (Weeks 1-4)

1. Week 1: Inbound invoice census - Pull 90 days of inbound invoices and classify by format (E-Rechnung native, hybrid, PDF, paper) and supplier.
2. Week 2: Archive review - Audit current GoBD archive against the 14 July 2025 BMF requirements, with a focus on original-XML preservation.
3. Week 3: Booking workflow map - Document the current booking workflow from receipt to posting, including approval steps and exception handling.
4. Week 4: Baseline KPIs - Lock today's cost per invoice, cycle time, auto-booking rate, exception rate, and Steuerberater handover quality.

Phase 2: Build and Shadow Mode (Weeks 5-8)

5. Week 5-6: Pipeline build - Connect data sources, configure the GoBD-compliant archive, integrate with DATEV or SAP, build the AI booking-suggestion layer with a 200 to 400 invoice evaluation set.
6. Week 7: Shadow mode - Agent processes real inbound invoices in parallel with the human team but does not post bookings. AP staff and Steuerberater review divergences.
7. Week 8: Calibration - Tighten policies based on shadow-mode misses. Validate GoBD archive integrity through Steuerberater review. Add edge cases to the evaluation set.

Phase 3: Go Live with Human-in-the-Loop (Weeks 9-12)

9. Week 9: Soft launch - Live for one supplier segment. Auto-booking only for high-confidence matched-PO invoices; human review for the rest.
10. Week 10-11: Expand intents - Add additional supplier segments and the second use case (typically PO matching or duplicate detection).
11. Week 12: Measure and decide - Compare cost per invoice, cycle time, auto-booking rate, and Steuerberater handover quality against baseline. Decide on next channel and use cases.

DSGVO, EU AI Act, Steuerberater, Betriebsrat

E-Rechnung sits at the intersection of tax law, data protection, AI regulation, and the Steuerberater relationship. Each of the four corners has a clear obligation that the implementation must respect.

GoBD and tax law

• 14 July 2025 BMF letter - The current authoritative reference for GoBD-compliant E-Rechnung handling².
• 8-year retention - Reduced from 10 by BEG IV; clock starts at end of calendar year.
• Original XML preservation - Mandatory. Printing alone is insufficient.
• Steuerberater sign-off - The Steuerberater retains final responsibility for the books and must approve the GoBD archive design.

DSGVO essentials

• Personal data on invoices - Supplier and contact names trigger DSGVO. Document the lawful basis for processing.
• Auftragsverarbeitungsvertrag (AVV) - Required with the LLM provider, the archival vendor, and the AI agent platform.
• EU residency - Anthropic, OpenAI, Mistral, Google and Aleph Alpha all offer EU-hosted endpoints for invoice processing.
• Audit trail - Log every read, every booking suggestion, every export for the 8-year retention period.

EU AI Act

• Risk classification - A booking-suggestion AI is limited-risk because it does not make creditworthiness, employment, or safety decisions.
• Article 4 (AI Literacy) - Mandatory from August 2026. AP staff and Steuerberater contacts who configure or supervise the AI need proportionate training¹⁹.
• Transparency - Disclose AI involvement in the AP process internally; external transparency is not required for inbound invoice processing.

Betriebsrat

• Co-determination triggers - Section 87(1)6 BetrVG covers technical devices that monitor employee performance²⁰. AP-team performance metrics derived from the AI system trigger this.
• Betriebsvereinbarung - Practical path: covers AP-staff performance use, retention, and the right to override AI suggestions.
• Engage early - Include the Betriebsrat in Phase 1, not after launch.

How Superkind Fits

Superkind builds custom AI agents that sit on top of DATEV, SAP, Lexware, Sage, or Dynamics for end-to-end E-Rechnung processing. The approach is process-first - we sit with your AP team and your Steuerberater before any model is involved - and the agent augments the platform you already use.

• Process-first discovery - We sit with your AP lead and Steuerberater for a full week before any agent is built. No templates, no slide-deck assumptions.
• Sits on top of your stack - Connects to DATEV (Connect online, DUO), SAP (BAPI/OData), Lexware, Sage, Microsoft Dynamics, and custom ERPs through APIs. Your AP team keeps the tools they know.
• GoBD-compliant by design - Original XML preservation, immutable archive, audit-trail design, all reviewed against the 14 July 2025 BMF letter.
• DATEV-fluent - Native DATEV Connect online integration where DATEV is the system of record; works alongside DATEV's own AI services for booking suggestion.
• Sovereign by default - EU data residency, your choice of LLM (Anthropic, OpenAI, Mistral, Aleph Alpha), and on-premises options where regulated industries require them.
• Outcome metrics that matter - We design and report against cost per invoice, cycle time, auto-booking rate, and Steuerberater handover quality.
• Steuerberater-friendly - We support the Steuerberater conversation directly, including walkthroughs of the GoBD archive design.
• Live in 8 to 12 weeks - First production use case in a single channel within a quarter; subsequent use cases add faster because the integration layer is already built.

Decision Framework: Where to Invest in 2026

Every German Mittelstand company has the same 18-month window from late 2026 to the full sending obligation. Use this framework to decide where to invest first.

Related Articles

• AI in Bookkeeping: How the Mittelstand Automates Receipts and DATEV Booking - The companion piece on the broader bookkeeping pipeline that sits behind the E-Rechnung mandate.
• Intelligent Document Processing - The IDP architecture that handles legacy PDF and paper invoices during the transition period.
• AI in Procurement - The procure-to-pay companion that completes the inbound side of the AP pipeline.
• SAP and AI Agents - Integration patterns for SAP S/4HANA, ECC, and Business One that apply directly to E-Rechnung receiving.
• AI as a Compliance Assistant - The broader GoBD, DSGVO, and EU AI Act compliance picture.
• Human-in-the-Loop - The autonomy levels and risk x reversibility matrix for AI booking approvals.

Frequently Asked Questions

Henri Jung

Co-founder of Superkind, where he helps SMEs and enterprises deploy custom AI agents that actually fit how their teams work. Henri is passionate about closing the gap between what AI can do and the value it creates in real companies. He believes the Mittelstand has everything it needs to lead in AI - it just needs the right approach.