AI in Accounting: 78%
A medium-sized machine builder from Swabia has been operating a live AI accounting system since February 2026. The company processes 14,000 incoming invoices per quarter with three accountants, using an end-to-end pipeline from OCR to posting and DATEV export. The rate of unadjusted entries stands at 78 percent, compared to zero a year ago. This story is neither a panacea nor an efficiency miracle, but a honest record of decisions that other SMEs can now follow, without repeating the same mistakes.
07.05.2026
Key Takeaways
- 78% Dark Booking is Achievable: Cleanly set up pipelines from OCR, classification, and rule-based posting can achieve this rate in six to nine months in the DACH Mittelstand. Anyone promising 95% is selling a vision, not a product.
- The Challenge is Account Frame Maintenance: SKR-04 or SKR-03 alone are insufficient. The AI requires a mapping layer between supplier behavior, cost centers, and general ledger accounts. This layer requires ongoing maintenance, not just setup, and costs six to twelve hours per month.
- DATEV Remains an Interface, Not a Platform: The value creation of the AI layer occurs before the DATEV export. Tax advisors remain the corrective measure, not the bottleneck. Those who misassign responsibilities will fail due to the four-eye principle.
Related:E-Rechnungspflicht: Umstellung bis Jahresende 2026 / Cybersecurity Tools for SMEs Without IT Teams
What a KI Bookkeeping Can Truly Achieve for SMEs
What is a KI Bookkeeping? A KI Bookkeeping integrates OCR receipt processing, classification models, and language models with rule-based accounting to form a continuous pipeline. The system learns from historical transactions, automatically suggests general ledger accounts, cost centers, and tax keys, and exports the results in a revision-safe format to financial accounting systems like DATEV, Sage, or Lexware. Employees then review exceptions, rather than manually entering each transaction.
The mechanical engineering company based in Schwaben encountered a classic bottleneck in 2025. Three bookkeepers manually processed approximately 14,000 incoming invoices per quarter, along with travel expenses, vendor management, and dunning. The team was overwhelmed, with one position vacant, and the market for bookkeepers in the Schwaben region had been dry for 18 months. Management faced two options: outsourcing to a tax consultant who also had staffing issues or building their own pipeline.
The decision was made to develop a modular pipeline. OCR through a DACH provider, classification using a fine-tuned model based on Llama-3.2-8B, rule-based accounting in a Python layer, and DATEV export via the official XML interface. Six months of development, three months of parallel operation, and then the switchover. The first twelve weeks were uncomfortable. Incorrect postings increased, and the team lost trust. It wasn’t until week 16, when the training data had sufficient volume, that the mood improved.
Three Levers for SMEs to Act on Now
The interesting decisions in such a project are not those in the quarterly report, but those in the third week, which no one hears about. Three levers have proven to be the most reliable and effective in several DACH SME projects.
What Works, What Doesn’t in SMEs
What Doesn’t Work
- Supplier master data without uniqueness. If the same supplier is listed in three different ways, AI will learn to make permanent errors.
- Tax consultants integrated as bottlenecks rather than corrective measures. Anyone who approves every invoice before booking loses pipeline efficiency.
- Generic cloud LLM solutions without DACH data context. GDPR-compliant operation is mandatory, otherwise, accounting will lead to a discussion about order processing with data protection.
- Management’s expectation of 95% dark posting in three months. This sabotages the project by creating undue pressure.
What Works
- Modular pipeline instead of a complete solution. Each component can be replaced individually, keeping vendor lock-in manageable.
- Rules plus ML, not rules or ML alone. High-volume suppliers with rules, long-tail with the model, error rate significantly decreases.
- Active involvement of the accounting team in the setup. The employees know the special cases that no model can learn from historical data.
- Quarterly re-training on new invoices. The error rate in the pilot decreased by 35-50% per quarter without new architecture.
Where the Lever Now Stands
Three SME profiles will yield the clearest added value over the next twelve months. Manufacturing businesses with high incoming invoice volumes from the traditional mid-sized sector, as labor shortages hit hardest there. Service providers with complex travel and expense systems, as classification and booking per invoice are the most expensive there. And family businesses with generational transitions, as a modernized accounting process significantly eases the handover. Those without these triggers should continue with their existing accounting systems and should do so rather than pursuing innovation for its own sake.
A candid observation to conclude. The friction in such projects rarely comes from the technology itself but rather from ownership issues. Who in the accounting team is responsible for model maintenance? Who maintains the chart of accounts? Who decides when a long-tail supplier transitions to high-volume regulation? Three responsibilities that are still undefined in most mid-sized enterprises. Those who clarify these early will enter live operations faster. Those who wait for a complete tooling solution that handles everything will wait two years longer than necessary. This applies to AI accounting just as much as the question of what a cybersecurity setup in an SME without its own IT team actually needs in day-to-day operations.
Frequently Asked Questions
How long does it realistically take for an AI accounting system to become productive?
It typically takes six to nine months from the start of the project to the first productive dark bookkeeping. The first three months are focused on data preparation, followed by three to four months of setup and pilot testing, and then parallel operation. Those who rush the process risk losing the trust of the accounting team if the error rate is high in the initial weeks.
Which models are suitable for classification in medium-sized enterprises?
For DACH SMEs with 5,000 to 50,000 transactions per year, locally operated fine-tuned 8B to 13B models like Llama-3.2 or Mistral-Small are sufficient. For those preferring DSGVO-compliant cloud hosting, German providers such as Aleph Alpha or OpenGPT-X offer suitable options. Generic US cloud LLMs are technically possible but often raise data protection concerns during initial discussions with the data protection officer.
How does the tax consultant view AI accounting?
In most pilot projects, the tax consultant remains a corrective rather than a bottleneck. They review quarterly and annual financial statements, clarify special cases, and assume final responsibility. Operational bookkeeping runs in the client’s system, and the consultant receives the data through DATEV or a similar interface. Those who allow the consultant to review every document are not fully understanding the pipeline logic.
What does a typical setup cost for an SME?
For an SME with 10,000 transactions per year, setup costs range from 45,000 to 90,000 Euros, depending on whether the implementation is internal or external. Monthly running costs range from 1,500 to 4,000 Euros for OCR licenses, model hosting, and maintenance. The break-even point compared to hiring an additional full-time accounting position typically occurs between the 14th and 22nd month for most setups.
About the Author
Angelika Beierlein is COO at Evernine. She has held leadership roles across media and tech-adjacent sectors and regularly writes about operational topics where structures matter more than slogans. She values honest retrospectives over three offsites and sees them as more valuable for meaningful change.
More from the MBF Media Network
Source Title Image: Pexels / Mikhail Nilov (px:6963098)