AI Agent Development Company in Munich

Manufacturing / Procurement / Munich

A Munich-based Mittelstand precision engineering company (EUR 380M revenue, 1,400 employees, manufacturing hydraulic components for construction and agricultural equipment, 4 factories in Germany, Czech Republic, and China, 420 active suppliers) needed to automate its procurement operations. (1) Purchase order volume — the company processed 14,000 purchase orders annually across: raw materials (steel, aluminium, brass — 35% of PO volume), components (seals, bearings, fasteners — 30%), MRO supplies (maintenance, repair, operations — 25%), and services (contract manufacturing, logistics, consulting — 10%). Each PO followed a multi-step process: purchase requisition (from production planning or department request), supplier selection (from approved supplier list, comparing price, availability, and quality rating), PO creation and approval (gyeoljae-style approval chain in SAP — different approval levels based on PO value), order confirmation tracking, goods receipt verification, and invoice matching (3-way match: PO, goods receipt, invoice). (2) Processing bottleneck — the 8-person procurement team processed 14,000 POs manually. Average PO processing time: 3.2 hours (from requisition to approved PO). For rush orders: 6+ hours (requiring phone calls to suppliers for availability, manual price comparisons, and expedited approval routing). The procurement team was the bottleneck in the company's order-to-delivery cycle — production planning frequently waited 2-3 days for materials to be ordered after the requisition was created. (3) Maverick spending — 28% of purchases were made outside the formal procurement process (employees ordering directly from suppliers via email or phone, without PO). Maverick spending: estimated EUR 12M annually without price comparison, contract compliance, or approval workflow. The finance team discovered maverick purchases only when invoices arrived without matching POs — creating reconciliation delays and missed early payment discounts. (4) Supplier management — the company's 420 suppliers were managed through a combination of SAP vendor master data and Excel spreadsheets. No systematic supplier performance tracking (delivery performance, quality ratings, price competitiveness). Supplier consolidation opportunities missed — the company often purchased the same component from multiple suppliers at different prices. (5) Betriebsrat — the Betriebsrat was supportive of procurement automation in principle (it did not directly affect production floor employees) but required consultation per BetrVG Section 90 as a significant technology change.

We built the procurement AI agent system over 10 weeks with Betriebsrat consultation completed at week 2: (1) Betriebsrat: brief consultation — procurement automation affects the procurement team (8 people) and does not process production employee data. BetrVG Section 90 information provided. Betriebsvereinbarung not required (no employee performance monitoring involved) but an information agreement was signed for transparency. (2) Purchase requisition processing agent: autonomous agent processing purchase requisitions from SAP. Requisitions generated by: SAP MRP (Material Requirements Planning) — automatic requisitions based on production planning and stock levels, manual requisitions from department heads (via SAP Fiori), and maintenance requisitions from SAP PM (spare parts for maintenance orders). The agent processes each requisition: verifies the requested item against the material master (SAP MM), identifies the optimal supplier based on: contracted price (Rahmenvertrag — framework agreement), current spot pricing (for non-contracted items — the agent queries supplier portals and price databases), supplier quality rating (from SAP QM — vendor evaluation scores), delivery capability (estimated delivery time based on historical performance and current supplier load), and geographic preference (German suppliers preferred for critical components — supply chain resilience strategy post-COVID). Supplier recommendation: the agent recommends the optimal supplier with detailed justification — price comparison, delivery timeline, and quality rating. For contracted items with clear Rahmenvertrag pricing: the agent auto-selects the contracted supplier. (3) PO creation and approval agent: autonomous agent creating POs in SAP and routing for approval. PO creation: the agent generates the PO in SAP MM with: correct material numbers, quantities, delivery dates, prices, and Incoterms. All PO fields auto-populated from the requisition and supplier selection. Approval routing: the agent routes POs through the company's approval hierarchy (configured in SAP): POs <EUR 5,000: Abteilungsleiter (department manager) approval — auto-approved if within budget and from contracted supplier. POs EUR 5,000-25,000: Einkaufsleiter (procurement manager) approval. POs EUR 25,000-100,000: Werkleiter (plant manager) approval. POs >EUR 100,000: Geschaeftsfuehrer (managing director) approval. Approval notifications: sent via email and KakaoWork-equivalent (the company used Microsoft Teams) with one-click approve/reject. Approval time: from 1.5 days average to 2.8 hours (approvers receive complete information — they can approve confidently without requesting additional details). Rush orders: expedited workflow — the agent identifies rush requisitions (production-critical materials with <48 hours lead time), contacts the recommended supplier directly (via automated email/EDI with human review), and routes for same-day approval with clear "RUSH" designation. Rush PO processing: from 6+ hours to 1.5 hours. (4) Order tracking agent: autonomous agent monitoring PO fulfilment. The agent tracks: order confirmation receipt (did the supplier confirm the PO within 48 hours? If not: automatic reminder sent), delivery tracking (estimated delivery dates monitored against promised dates — deviations flagged immediately), goods receipt (when materials arrive: the agent matches the delivery against the PO — quantity, quality, and documentation verified), and invoice matching (3-way match: PO, goods receipt, invoice — discrepancies flagged for procurement team review). Exception handling: delivery delays (>2 days beyond promised date): agent sends escalation to supplier and notifies the requisitioner with updated timeline. Quantity discrepancies (>2% deviation): agent flags for quality inspection before goods receipt posting. Price discrepancies (any deviation from PO price): agent flags for procurement review before invoice approval. (5) Maverick spending detection agent: autonomous agent identifying purchases made outside the formal procurement process. Detection: the agent monitors: invoices without matching POs (flagged at invoice receipt), direct supplier communications (email forwarding rules identifying purchase-related communications outside the procurement workflow), and expense reports containing items that should be procured through the formal process. Action: each detected maverick purchase: flagged to the procurement team with analysis (what was purchased, from whom, at what price, and what the procurement process would have cost), routed for retrospective PO creation (regularising the purchase in SAP), and reported in monthly maverick spending dashboard (aggregate — not individual blame, per Betriebsvereinbarung information agreement). Maverick spending reduction: from 28% to 9% of total purchases within 6 months (combination of detection and cultural change — departments know that out-of-process purchases are now visible). (6) SAP integration: all agent actions executed through SAP. Requisitions read from SAP MM (ME51N). POs created in SAP MM (ME21N) via BAPI. Approval workflows managed through SAP (ME28). Goods receipts posted in SAP (MIGO). Invoice matching in SAP (MIRO). Vendor evaluation data from SAP QM. The agents do not create shadow systems — SAP remains the single source of truth for all procurement data.

Procurement efficiency: PO processing time from 3.2 hours to 35 minutes (82% reduction). Rush PO processing from 6+ hours to 1.5 hours (75% reduction). Requisition-to-PO lead time from 2-3 days to same-day (for standard items with contracted suppliers). Approval time from 1.5 days to 2.8 hours (81% reduction — mobile approvals with complete information). Procurement capacity: 8-person team now processes 14,000+ POs with capacity for 22,000+ — supporting company growth without additional procurement headcount. Cost savings: contract compliance improvement (maverick spending from 28% to 9%): estimated EUR 1.8M annual savings (purchases now routed through negotiated contracts with volume pricing). Supplier consolidation: the agent's analysis identified 34 components purchased from multiple suppliers at different prices — consolidation to preferred suppliers saved EUR 420K annually. Early payment discount capture: automated invoice matching enabled faster payment processing — the company captured EUR 180K in early payment discounts previously missed due to slow processing. Price comparison: the agent's systematic price comparison identified EUR 340K in overpriced purchases that had been routinely approved without comparison. Total procurement cost savings: EUR 2.74M annually. Supplier performance: delivery on-time rate: improved from 82% to 91% (systematic tracking and automated follow-up improved supplier accountability). Invoice matching accuracy: from 78% auto-match to 94% auto-match (fewer discrepancies, faster processing). Betriebsrat: information agreement functioning smoothly. Quarterly reports provided. Zero concerns raised. The procurement team's feedback: positive — "the system handles the routine POs so I can focus on strategic sourcing and supplier negotiations." Production impact: materials availability improved — production planning no longer waited 2-3 days for PO processing. Production schedule adherence improved 8% (materials available on time). Revenue impact: faster material procurement contributed to: 4-day reduction in average order-to-delivery cycle, estimated EUR 1.2M in additional revenue from reduced customer lead time quotes. Cost: procurement efficiency: EUR 2.74M annual savings. Production improvement: estimated EUR 1.2M revenue impact. Procurement team capacity: growth supported without hiring (2 avoided hires at EUR 85K = EUR 170K savings). Development cost: EUR 155,000. Payback period: 3 weeks.