Project led by Bruno Jiménez, now an Akaen partner.
In multi-category e-commerce, managing hundreds of suppliers with different formats is an operational nightmare. This case describes order confirmation automation at one of Central Europe’s largest marketplaces, eliminating a manual bottleneck and achieving 70% touchless processing.
Executive summary
About the customer
A leading multi-category marketplace in Central Europe, with over a million active references and hundreds of suppliers across the continent. Its daily operations depend on processing order confirmations accurately and quickly to meet customer delivery commitments.
The Challenge: Mass Scale and Extreme Format Diversity
The volume was high — an average of 150 order confirmations per day — but the real problem was heterogeneity. Each supplier sent its confirmations in different formats: native PDFs with complex tables, scanned documents, vertical and horizontal layouts, and mixed languages (German, English, French, Italian). These documents reached more than 10 departmental mailboxes without a unified standard.
Traditional template-based OCR was not viable: it would have required maintaining hundreds of extraction templates as suppliers and formats changed. The team spent about 12.5 hours a day on manual processing, an unsustainable cost in the German labour market.
The Solution: Adaptive AI + RPA on SAP
The workflow acts as an intelligent abstraction layer over SAP. It understands document content semantically rather than relying on fixed coordinates, whether supplier A puts the date at the top right or supplier B places it in the footer.
Automated processes (4 key phases)
Confirmations scattered in more than 10 mailboxes, mixed with spam.
A bot monitors all mailboxes, downloads attachments and filters out noise automatically.
Hundreds of formats impossible to cover with OCR based on templates.
Generative AI extracts headers and tables and normalises units, currencies and dates regardless of layout or language.
Manual verification of each confirmation against the original order in SAP.
A robot finds the order in SAP and automatically checks each line’s quantity, price and date.
Discrepancies that required manual search to understand what had changed.
If everything matches, it confirms in SAP. Otherwise, it generates an alert with the discrepancy already analysed (e.g. “Supplier: €12/unit, SAP: €10/unit”).
Operational comparison
- • Manual review of each confirmation.
- • ~12.5 hours a day spent on repetitive processing.
- • Template-based OCR was not viable (+100 formats).
- • Frequent transcription errors.
- • Semantic AI that understands any format.
- • 70% of confirmations processed without human intervention.
- • Automatic line-by-line SAP cross-checking.
- • The team focuses only on exceptions.
The Impact
The geographical context amplifies ROI: with high administrative labour costs in Germany, freeing the team from repetitive tasks has an immediate financial impact:
- Capacity freed up: capacity equivalent to more than 1.3 full-time roles returned to the business (~2,200 hours per year).
- Economic efficiency: over €100,000 in estimated annual savings on direct operating costs.
- Data quality: elimination of manual transcription errors in prices, quantities and dates.
- Scalability: new suppliers are integrated without creating additional templates.
Operational results
- Touchless Processing: 70% of confirmations processed without human intervention.
- Accelerated resolution: The remaining 30% reach the team with the discrepancy already analysed.
- Built-in multilingual support: the system processes documents in German, English, French and Italian without additional configuration.
Conclusion
"When supplier diversity is the norm and not the exception, template-based automation does not scale. Semantic AI makes heterogeneity an operational advantage."
Want to automate supplier management in your operations? .