Intelligent Invoice Approval with DATEV Integration for Planning Firms

The Planning Firm and the Invoice Flood

A specialized planning firm with seven engineers — focused on renewable energy projects like biogas, wind, and PV installations — manages approximately 250 active projects simultaneously. That means: 400 to 700 incoming invoices per year. Suppliers, subcontractors, authorities, material vendors. Every invoice must be checked, categorized, approved, and entered into accounting software DATEV.

The reality was barely manageable. Invoices arrived by email. One person printed them, filed them in folders, shipped them physically to project managers for approval. After return, they were printed again — this time for DATEV. Two printouts per invoice. Hundreds of sheets per month. And manual DATEV entry? That was time-consuming and error-prone.

Particularly problematic: zero visibility into approval status. No way to see which invoices were "stuck." No data on average processing times. And no chance to ever get a picture of actual costs per project — everything scattered across different people, folders, spreadsheets.

The Strategy: From Paper to Platforms

The goal was ambitious: fully automated invoice processing. No printing. No manual DATEV entry. Direct data flow from invoice receipt to accounting system. The technical foundation would be the Microsoft Power Platform and intelligent DATEV integration via XML.

Phase 1 — Invoice Receipt:

Invoices arrive in the accounting mailbox. A Power Automate flow checks the email for attachments (must be a PDF with "invoice" in the name or from the known scanner). If yes: the PDF is automatically uploaded to SharePoint and tagged with metadata. The filename is standardized (Date_Vendor_Amount).

The invoice is simultaneously scanned with AI Builder. The machine learning model automatically recognizes vendor names, invoice numbers, amounts, tax rates, and due dates. This data lands in a structured table — a "sandbox" data source in Dataverse, the Power Platform's internal database.

Phase 2 — Department Assignment and Approval:

Based on extracted data and mapping rules, Power Automate routes the invoice to the correct project manager or department head. A Teams Approval is sent — with all relevant information already pre-filled. The approver sees:

• Invoice number and amount
• Vendor
• Due date and any available early payment discount information
• Project the invoice is assigned to
• The scanned invoice itself (as PDF link)

One click approves. Optionally, the approver can add a comment or make corrections if the AI extraction was off.

Phase 3 — DATEV Integration via XML:

After approval, the smartest part happens: Power Automate automatically generates an XML file in DATEV standard. This XML contains all accounting information: vendor number, invoice number, amount, tax amount, date, due date, account coding (based on project and cost type). The XML is transmitted directly to DATEV.

DATEV reads in the XML and creates the invoice automatically in the system. No manual data entry. No typos. The invoice is immediately documented in DATEV and ready for payment processing.

Phase 4 — Archiving and Transparency:

The approved invoice is archived in SharePoint, marked as "DATEV-processed." A message back to the approver informs them: "Invoice approved and transferred to DATEV." The invoice file is indexed with all extracted metadata — meaning it's fully searchable later.

Simultaneously, all this data feeds into Power BI dashboards integrated in Teams. Management sees at a glance: how many invoices came in this month? Average processing time? Which projects have the most vendors? What accounting delays exist?

Eliminating Waste

From a Lean perspective, the system is a textbook example of waste elimination:

• Transport: Invoices no longer travel physically. Zero distance.
• Motion: Nobody has to get up, search for folders, or print.
• Waiting: Invoices don't "sit on desks" anymore — they're immediately routed.
• Rework: AI-driven extraction reduces errors. If corrections are needed, they happen digitally.
• Overproduction: No more duplicate printing. No redundant spreadsheets.

The planning firm had to manually touch about 2,000 invoice pages per year. That's gone.

The Numbers

After implementation (approximately 6 weeks for correct AI model calibration):

• Manual Processing Time: reduced from approximately 25 minutes per invoice (printing, routing, data entry) to 3 minutes — an 88 percent reduction
• DATEV Entry Accuracy: from circa 94 percent to 99.2 percent (remaining 0.8 percent are edge cases like wrong currencies or unknown vendors)
• Paper Consumption: from 2,000+ pages/year to approximately 150 (only for archive print if needed) — minus 92 percent
• DATEV Processing Time: from average 4 days to under 6 hours
• Cost per Invoice: from approximately €1.50 (printing, time, error handling) to €0.12 system costs

More importantly: management now has real-time data on cost flows and vendor performance for the first time. The transparency is similar to what we achieved in a general invoice approval workflow handling 340 invoices per month — though without the DATEV integration.

Scalability and Maintenance

The infrastructure is deliberately decentralized. A new vendor? Mapping rules are updated, AI models self-correct through feedback loops. A new project with different cost accounts? A new mapping entry in Dataverse. The flow remains unchanged.

The XML interface to DATEV is standardized — if DATEV updates, only the XML templates need adjustment, not the entire logic.

Error handling runs automatically: if a PDF isn't readable or AI extraction is too uncertain (confidence score below 70 percent), those invoices automatically land in a queue for manual review. An "Escalation Channel" in Teams shows the team which invoices need attention. Typically that's less than 5 percent.

What Comes Next?

The planning firm is already planning the next phase: apply the same intelligence to outgoing invoices to clients. Draft cost estimate PDFs could be automatically converted, numbered, and shipping-managed. Project completion invoices could be semi-automatically generated based on approved vendor invoices.

The core insight: once an automation infrastructure is built, it's not the end — it's the beginning. It creates the foundation for continuous improvement and true digital enterprise planning.

---

Book a free strategy call