AI Procurement Agent for Precision Manufacturing
Hartley Engineering's procurement team was spending 60% of their time on routine purchasing tasks. We built a custom AI agent that autonomously manages supplier sourcing, quote comparison, and purchase order generation.
Results
Impact delivered
£1.4M
Annual procurement savings
Through better quote comparison and supplier negotiation
68%
Purchase orders processed autonomously
Routine orders handled end-to-end without human involvement
74%
Procurement cycle time reduction
From requisition to purchase order generation
22 hrs/week
Strategic time reclaimed
Team time redirected to supplier development and negotiation
The Challenge
What they faced
Hartley Engineering, a Midlands-based precision manufacturing firm with 280 employees, processes over 3,500 purchase orders per month across 200+ suppliers. Their four-person procurement team was buried in repetitive tasks: requesting quotes, chasing supplier responses, comparing pricing across multiple vendors, and manually generating purchase orders. Strategic procurement activities like supplier relationship development, contract negotiation, and cost reduction initiatives were consistently deprioritised. The team estimated that 60% of their working hours were consumed by tasks that followed predictable, rule-based patterns — perfect candidates for automation.
Our Solution
How we solved it
We designed and deployed a custom AI procurement agent that operates autonomously within defined guardrails. The agent monitors inventory levels and production schedules to anticipate purchasing needs, automatically requests quotes from approved suppliers, compares responses against historical pricing and quality data, and generates purchase orders for approval. For routine orders under £5,000, the agent can execute end-to-end without human intervention. Higher-value orders are prepared and presented to the procurement team for a single-click approval. The agent integrates with Hartley's SAP system and communicates with suppliers via email using natural language.
Approach
Step by step
Procurement workflow mapping
Spent two weeks embedded with the procurement team, documenting every step of the purchasing process from requisition to delivery confirmation.
Agent architecture design
Designed a multi-step agent framework with clearly defined decision boundaries, approval thresholds, and escalation paths for exceptions.
Supplier communication module
Built an email-based communication layer that allows the agent to send RFQs, parse supplier responses (including PDF quote attachments), and maintain conversation context.
SAP integration
Developed bidirectional integration with SAP for real-time inventory data, production schedules, purchase order creation, and goods receipt confirmation.
Decision engine training
Trained the quote comparison engine on 18 months of historical purchasing data, incorporating pricing trends, supplier reliability scores, and lead time patterns.
Controlled rollout with increasing autonomy
Started with the agent in advisory mode (suggesting actions for human approval), then progressively increased autonomy for routine categories over eight weeks.
“I'll be honest, I was worried the agent would make mistakes that would disrupt our supply chain. Six months in, it's made fewer errors than we did manually and it's saved us over a million pounds. The procurement team now spends their time on work that actually requires human judgement.”
Sarah Collinson
Operations Director, Hartley Engineering Ltd
Technology
Stack used
FAQ
Frequently asked questions
The agent has clearly defined operating boundaries. Any situation outside its decision parameters — unusual pricing, new supplier requests, specification changes — is automatically escalated to the human procurement team with a full context summary.
Supplier emails are sent from a dedicated procurement address and follow natural business communication patterns. Suppliers are informed that Hartley uses AI-assisted procurement, but the communication quality is indistinguishable from human-written correspondence.
Approval thresholds are fully configurable by the procurement team. Currently, orders under £5,000 for approved suppliers and standard materials can be processed autonomously. All other orders require human approval via the dashboard.
Want similar results?
Book a free strategy call and we'll assess how AI can deliver impact for your business.