The hypothesis
Mid-market manufacturers live with a dangerous dependency: 2-3 procurement veterans hold all the supplier relationships in their heads. When Sarah from purchasing is out sick, nobody knows which suppliers need daily follow-up, which ones respond only to specific language, or which purchase orders are at risk of missing production schedules. The manual burden is crushing - 240 supplier follow-ups per week across 85 active purchase orders, consuming 28 hours of expert time.
We believed an agent could eliminate this tribal knowledge dependency by systematically chasing supplier confirmations, delivery updates and exception notifications. The hypothesis: automate 87% of routine supplier follow-up whilst maintaining the relationship nuance that prevents supply disruptions.
What we tested
We built a 4-component supplier follow-up agent that integrates directly with the client's ERP system. The agent monitors purchase order status, generates contextual follow-up communications, tracks supplier response patterns and escalates genuine exceptions to human procurement staff.
Purchase Order Monitor: Scans the ERP for POs approaching confirmation deadlines, delivery dates or milestone triggers. Feeds a stream of supplier engagement requirements to the next stage.
Communication Generator: Creates supplier-specific follow-up messages based on purchase order details, supplier history and communication templates. Adapts tone and urgency based on supplier response patterns and order criticality.
Response Classifier: Processes incoming supplier emails, phone logs and portal updates. Extracts confirmation dates, delivery commitments and exception flags. Routes clear confirmations to ERP updates, ambiguous responses to human review.
Escalation Router: Identifies genuine supply risks that require human intervention - supplier capacity issues, material shortages, delivery delays impacting production schedules. Prioritises escalations by production impact and supplier relationship sensitivity.
We tested with 4 weeks of live purchase order data: 340 active POs across 67 suppliers, representing £1.8M in committed spend. The agent processed 960 individual follow-up interactions during the test period.
The architecture
ERP INTEGRATION ZONE AGENT PIPELINE
┌─────────────────────┐ ┌─────────────────────┐
│ │ │ [1] PO Monitor │
│ Purchase Orders │───────────────▶│ • Deadline scan │
│ • PO status │ │ • Milestone check │
│ • Delivery dates │ │ • Supplier queue │
│ • Supplier data │ └─────────────────────┘
│ │ │
└─────────────────────┘ ▼
┌─────────────────────┐
┌─────────────────────┐ │ [2] Comm Generator │
│ │ │ • Message craft │
│ Email / Portal │◀──────────────▶│ • Supplier adapt │
│ • Outbound msgs │ │ • Urgency tone │
│ • Response capture │ └─────────────────────┘
│ │ │
└─────────────────────┘ ▼
┌─────────────────────┐
│ [3] Response Class │
│ • Email parse │
│ • Commitment ext │
│ • Exception flag │
└─────────────────────┘
│
▼
┌─────────────────────┐
│ MATCH/EXCEPTION │
│ DECISION │
└─────────────────────┘
│ │
MATCH │ │ EXCEPTION
▼ ▼
┌─────────────────────┐ ┌─────────────────────┐
│ │ │ [4] Escalation │
│ ERP Updates │◀───────────────│ • Risk assess │
│ • Delivery confirm │ │ • Human queue │
│ • Status changes │ │ • Priority rank │
│ • Milestone log │ └─────────────────────┘
│ │ │
└─────────────────────┘ ▼
┌─────────────────────┐
│ Human Review │
│ • Supply risk │
│ • Relationship │
│ • Negotiation │
└─────────────────────┘
What worked
The agent successfully automated routine supplier follow-up whilst preserving the relationship management that prevents supply chain disruptions.
- Response classification achieved 91% accuracy on supplier commitment extraction from emails and portal updates
- Communication adaptation by supplier type reduced average response time from 3.2 days to 1.8 days
- Exception escalation correctly identified all 23 genuine supply risks during test period with zero false negatives
- ERP integration posted 798 delivery confirmations automatically, eliminating manual data entry backlog
- Supplier relationship scores remained stable - no complaints about automated communication tone or frequency
- 17% of supplier responses required human interpretation due to non-standard confirmation language
- New supplier onboarding still requires manual communication template creation and tone calibration
- Complex multi-line PO changes triggered false exceptions, requiring escalation rule refinement
What we learned
- Tribal knowledge can be systematised if you capture the decision patterns, not just the data. The agent learned which suppliers respond to urgent vs. gentle follow-up, which confirmation language indicates genuine commitment, and which supplier behaviours signal supply risk.
- Supplier relationship preservation requires communication adaptation, not just automation. The 1.8-day response improvement came from matching follow-up style to supplier preferences - formal purchase orders for enterprise suppliers, casual check-ins for local fabricators.
- Exception escalation must err on the side of human review for supply chain decisions. The 162 escalated cases included 23 genuine supply risks and 139 false positives. Better to over-escalate than miss a production-stopping delay.
- ERP integration transforms follow-up from administrative burden to supply intelligence. The agent is useful for chasing confirmations, but the real value comes from automatically updating delivery schedules in the production planning system, giving operations teams real-time visibility into supply constraints.
Potential for our clients
Experiment status
The supplier follow-up agent is production-ready for manufacturers with established supplier bases and standardised purchase order processes. The communication adaptation and response classification components require 2-3 weeks of supplier interaction data to calibrate effectively. Exception escalation rules need customisation based on production criticality and supplier relationship sensitivity.
Operations directors managing 40+ active purchase orders with 3-5 procurement staff should evaluate this capability. The typical engagement is a Workflow Sprint to map supplier communication patterns, define escalation criteria and configure ERP integration points. Most implementations achieve 75%+ automation within 6 weeks of deployment.