Invoice verification, posting, and reports with AI

Fundamentals and Core Processes in Accounting Automation

1. Introduction and Architecture

• Why Automate Accounting in 2026? Benefits, Risks, and Regulatory Requirements.

• Mandatory e-invoicing in Germany: Obligation to receive e-invoices effective January 1, 2025; obligation to issue e-invoices effective January 1, 2027 (for companies with annual revenue exceeding 800,000 EUR) and effective January 1, 2028 (for all B2B transactions). Formats: XRechnung, ZUGFeRD 2.0.1.

• Make 2026: Triggers, modules, AI agents (native modules for Claude), error handling, repetition logic.

• Make-Pricing: Core, Pro, Teams, Enterprise. Comparison of AI agents' credit usage versus traditional workflows.

• System integration: e-mail IMAP, SMTP), cloud storage (Google Drive, SharePoint), ERP (DATEV, SAP, Lexware Office, sevDesk), payment providers (Stripe, Adyen), webhooks, and APIs.

2. Data Collection and Preprocessing

• Workflow sources: emails, upload folders, scanners, portals, Peppol inbox.

• Automatic extraction of invoice data (supplier, IBAN, VAT ID, line items) using OCR and Document AI: Mindee, Klippa, Rossum, Hypatos.

• Validations: Check required fields, verify totals, validate IBANs, and prevent duplicates.

• Workshop 1: Automated Capture of Incoming Invoices with Make. Objective: Convert PDF and ZUGFeRD files into structured data and log them.

3. Classification and Account Assignment Proposal

• Rules for suppliers regarding accounts (for example, account 3120) and keyword-based logic.

• AI assistance with Claude (Opus 4.7 and Sonnet 4.7): Heuristics for account identification and anomaly detection.

• Workshop 2: Creating automatic journal entry suggestions. Objective: Convert data into an ERP-compatible format (DATEV or SAP).

4. Data Quality, Auditing, and Error Handling

• GoBD-compliant logging of every transformation, including a timestamp and hash.

• Resolving errors through quarantine processes and automated follow-ups with suppliers.

• Daily reports and alerts in the event of deviations.

Automation of Payments, Reports, and Regulatory Updates

5. Automated payment matching

• Input: Bank statements, Stripe and Adyen exports.

• Matching strategies: Exact match (amount and purpose), fuzzy search (for example, +/- 3 days), and Claude-based heuristics.

• Creation of reconciliation proposals, discrepancy lists, and automatic triggers for the dunning process.

• Workshop 3: Automated Payment Matching. Objective: Automate the process of matching incoming payments with open items.

6. Monthly Reports and Daily e-mail Summaries

• Automation of key metrics: revenue, costs, margin, cash flow.

• Generation of automatic HTML and PDF reports, optionally as a Claude Artifact.

• Scheduled or on-demand delivery, with versioning and target audience control.

7. Automated Regulatory and Exchange Updates

• Distribute daily reports on legal and market-related developments.

• Sources: Regulatory agencies, legal portals, stock market and market news.

• Workshop 4: Morning Compliance Briefing. Objective: Set up an end-to-end workflow for daily compliance updates using Claude and Make.

8. Security, Compliance, and Audit-Proofing

• GDPR-compliant workflows: Minimize data, store it securely, and implement data deletion policies.

• Secret management for API keys, access control, and logging.

• Audit-proof documentation of all changes with immutable logs and version control.

• The EU AI Act 2026 in Practice: AI Literacy effective February 2, 2025; GPAI effective August 2, 2025; Annex III (High-Risk) effective December 2, 2027.

• EU hosting for Claude: Anthropic API with DPA and AWS Bedrock Frankfurt. Notes on hallucinations and control mechanisms.

• Cowork as a desktop tool for the team: Prepare workflows locally and transfer them to Make in a controlled manner.

In our seminars, we explicitly address the data protection requirements for the use of AI in accounting. We explain GDPR-compliant deployment options (EU hosting via Azure, AWS Bedrock, etc.), point out risks such as hallucinations, and provide specific recommendations for operational implementation.