Duration: 22:50
🧾 Analytical Summary
🚀 The Core Problem: Supply Chain Friction at the ERP Boundary
Patrick, co-founder of Procuros, a Hamburg-based startup specializing in AI-driven supply chain automation, addresses a fundamental challenge facing businesses using Odoo and other ERP systems: while internal operations run smoothly, interactions with the outside world create massive friction and manual work.
The current reality involves two problematic approaches. First, many companies rely on email and Excel exchanges with third-party logistics (3PL) providers and partners—shipping notifications sent via email, inventory updates returned as spreadsheets, and constant manual data entry. Second, companies attempt traditional EDI (Electronic Data Interface), a technology from the 1980s that requires expensive one-to-one connections for each trading partner, hasn't improved in decades, and remains slow to implement.
The central question: "To AI or not to AI?"—where does AI actually make sense for optimizing supply chains, and where does it fall short?
🧠 Procuros: An AI Automation Collaboration Layer
Procuros positions itself as an AI automation collaboration layer that sits in front of ERP systems like Odoo, handling external interactions with suppliers, retail partners, 3PLs, and other trading partners. Data flows through Procuros—orders from customers, shipping notes, invoices—and the system harmonizes and automates the back-and-forth, eliminating manual friction.
The presentation showcases three specific, tangible examples of AI automation in action, demonstrating both capabilities and limitations.
📄 Use Case 1: Automated Order Processing with Agentic AI
Traditionally, when a B2B partner sends a PDF order, processing involves multiple manual steps:
- Manual data entry teams transcribe orders into the ERP
- Staff check inventory availability and make decisions
- Team members coordinate with fulfillment centers
- Customer service contacts buyers about substitutions for out-of-stock items
- Team leads oversee the entire process
This creates bottlenecks, errors, and delays. Procuros replaces this with what Patrick playfully calls "magical AI data processing"—extracting, interpreting, enriching, and harmonizing order data, then automatically inserting it into the ERP. Automation eliminates most manual steps, creating a unified flow.
The system uses an agentic architecture with specialized agents:
Trade Partner Recognition Agent: Identifies buyer and supplier from document structure.
Line Item Interpretation Agent: Understands industry-specific formats—in food, line items include product name, expiration dates, and packaging units; in fashion, orders contain size/color matrices across multiple SKUs.
The power comes from contextual intelligence. Agents leverage:
Master Data: Product catalogs inform which items exist and how they map to order descriptions.
Previous Documents: Historical order patterns guide interpretation.
Industry Context: Different industries (food, fashion, etc.) have distinct document structures.
User Instructions: Users add natural language rules like "Whenever a Swiss supplier sends an order, apply X, Y, Z," and agents incorporate these directives.
This approach validates the provocative principle: "If a human can process it, given enough context, AI can do the same." The system can demonstrate immediate value by processing orders in real-time as users watch.
💰 Use Case 2: Reconciling Bulk Payments and Remittance Advice
When suppliers receive bulk payments from retail partners (often monthly), they also receive a remittance advice—a document explaining which invoices are covered by a payment (e.g., €200,000 allocated across 20-30 invoices). Current process involves:
- Manual data entry matching remittances to invoices
- Checking each invoice status (paid, outstanding, partial)
- Identifying discrepancies (customer paid €80 but invoice was €100)
- Manually drafting emails to resolve payment issues
Procuros automates this with two capabilities:
Automatic Reconciliation: Matches remittance advice line items to invoices using both deterministic matching (invoice numbers) and intelligent inference when data is ambiguous.
Guided Resolution: When discrepancies arise, an AI agent suggests responses directly on the document. Users simply click approval to send claim requests or inquiries to trading partners.
The user interface displays reconciled invoices (perfect matches), items needing review, and discrepancies with one-click resolution options. The system even features a Clippy-style agent asking, "Do you want me to draft a claim message to your trade partner?"—making complex financial reconciliation dramatically faster.
📊 Use Case 3: Master Data Submission to Retailers
Submitting product master data (products, prices, specifications) to retailers is notoriously cumbersome:
- Retailers demand data in specific proprietary formats
- Suppliers manually create Excel sheets matching retailer templates
- Retailers manually process submissions into their PIM/ERP systems
- Constant back-and-forth for missing fields, incorrect formats, or data that doesn't make sense
Procuros addresses this through:
Data Checks and Transformations: Verifies required fields are present and uses AI to validate data makes sense (e.g., color names are legitimate, not random strings).
AI-Powered Data Enrichment: Translates supplier terminology to retailer terminology. For example, fashion retailers may only accept standardized colors (white, blue, red), but suppliers use creative names like "moon" or "snow white." AI maps these automatically. Similar enrichments handle size conversions and product relationships.
Intelligent Validation: The system checks logical relationships—recommended retail price should exceed purchasing price; if columns are swapped, AI flags the error.
Initially, users confirm AI suggestions (confirming "white" is correct), but over time the system becomes fully agentic, requiring no human intervention. The user experience transforms from tedious manual work into a "magical moment" where complex data transformations happen automatically.
⚠️ AI Limitations: Guardrails and Realism
Patrick emphasizes critical AI limitations in supply chain contexts:
Hallucinations and Non-Determinism: AI models produce slightly different outputs for identical prompts. In an interactive demonstration, Patrick asks attendees to query their favorite LLM twice with "How would you use AI to optimize supply chains?" The results differ each time. This is problematic when supply chains require deterministic outcomes—shipping notes with specific unit numbers and quantities must reach recipients exactly as specified. AI inventing additional products on an order creates real operational problems.
Solution: Implement guardrails preventing AI from taking excessive action or exercising unwarranted creativity.
Middle-to-Middle, Not End-to-End: AI excels at simple use cases (a basic order form dropped into ChatGPT yields 90% accuracy) but struggles with complexity. More intricate documents cause failures. Production AI requires step-by-step guidance rather than throwing problems at the model and hoping for solutions.
Retrofitting vs. Reinventing: Drawing an analogy to Tesla's autonomous driving evolution from rule-based machine learning (describing every scenario: "if user does this, car does that") to AI-based systems (learning from human training data), Patrick shows a historical image from 1803—the "horseless carriage," literally a carriage with a steam engine instead of a horse. It didn't work. Similarly, bolting AI onto old processes often fails. Some processes must be stripped entirely and rebuilt from the ground up for AI to deliver value.
🤝 The Future: Humans and Agents Working Hand-in-Hand
The vision isn't replacing humans entirely but creating hybrid teams:
Customer Ops: Instead of five people doing data entry, two humans work alongside three "friendly agents" handling repetitive tasks.
Accounts Receivable: Humans perform checks and strategic work while agents automate monotonous processing.
The goal: free humans for strategic, value-adding work—building supplier relationships, finding new prospects, deepening customer engagement—while agents handle routine labor.
In the long run, the implications become more radical: could entire trading processes become fully autonomous? Autonomous systems detecting discrepancies, resolving them, and pulling data from multiple sources without human intervention? Patrick suggests the path isn't an overnight leap but incremental evolution:
- Digitization: Move from manual to digital
- Rule-Based Automation: Implement automated data checks
- Co-Pilot Stage: Agents suggest actions; humans approve
- Full Autonomy: Systems operate independently
📈 Real-World Results: Customer Impact
Procuros works with retailers and suppliers in fashion, beauty, and food sectors—including Bunga, Flaconi, Niche Beauty, Tony's Chocolonely, and Cororo.
Operational Improvements: High-effort tasks reduced by 95%. For Buena, better data quality reduced warehouse cycle time by 50%—products reach stores twice as fast.
Business Outcomes: Customers using Procuros show 2.4-4% faster revenue growth compared to industry benchmarks. While correlation vs. causation remains unclear, companies embracing AI-driven automation scale faster.
💡 Key Takeaways
External interactions with ERP systems like Odoo create massive friction; AI can transform this.
Agentic AI systems with specialized agents and contextual intelligence deliver practical automation today.
Guardrails are essential—supply chains require determinism, and unconstrained AI introduces risk.
Some processes must be reinvented, not retrofitted, for AI to deliver value.
The future is hybrid: humans and agents collaborating, with humans focused on strategy and agents handling routine work.
🧠 Viewpoint: Odoo Perspective
⚠️ Disclaimer: AI-generated creative perspective inspired by Odoo's vision.
What Procuros demonstrates beautifully is the power of Odoo's open API architecture in enabling an entirely new class of integration solutions. We designed Odoo to be the core operational system, but we've always known that the B2B supply chain world is messy—legacy EDI, inconsistent formats, manual processes everywhere. Rather than trying to build everything ourselves, we've created the foundation that allows innovative partners like Procuros to sit in front of Odoo and solve these external friction points with AI. Their agentic approach to order processing, payment reconciliation, and master data management shows how Odoo becomes more valuable when surrounded by intelligent automation layers. This is exactly the kind of ecosystem innovation that accelerates digital transformation—Odoo as the reliable core, AI handling the chaos at the boundaries.
🏢 Viewpoint: Competitors (SAP / Microsoft / Others)
⚠️ Disclaimer: AI-generated fictional commentary. Not an official corporate statement.
Procuros addresses genuine pain points in SMB and mid-market B2B supply chains, and their agentic AI architecture is technically sound for their target use cases. However, enterprise supply chains operate at different scales and complexity levels. When you're processing millions of transactions daily across hundreds of trading partners in regulated industries, you need more than intelligent agents—you need enterprise-grade EDI networks with guaranteed SLAs, comprehensive audit trails, and compliance certifications. The "reinvent vs. retrofit" argument is interesting, but enterprises often cannot simply strip out decades of business logic and start fresh. Legacy EDI exists precisely because it's deterministic and legally defensible. The non-determinism Patrick correctly identifies as an AI limitation is a deal-breaker for pharmaceutical supply chains, aerospace manufacturing, or financial services. While the co-pilot vision is appealing, enterprises require formal change management, compliance validation, and risk mitigation frameworks that go beyond "agents suggesting actions." The question isn't whether AI adds value—it clearly does—but whether this startup-friendly approach translates to environments where supply chain errors have billion-dollar consequences and regulatory implications.
Disclaimer: This article contains AI-generated summaries and fictionalized commentaries for illustrative purposes. Viewpoints labeled as "Odoo Perspective" or "Competitors" are simulated and do not represent any real statements or positions. All product names and trademarks belong to their respective owners.