Retail Supply Chain 2025

Flowr — Retail Supply Chain
Automation

Transitioning large-scale supermarket supply chain operations from manual coordination to end-to-end agentic AI workflows.

The challenge

A coordination-heavy, manual process at scale

Retail supply chains in large supermarket operations involve continuous, high-volume manual workflows spanning demand forecasting, procurement, supplier coordination, and inventory replenishment. Despite investment in ERP systems and analytics platforms, the decision-making and coordination layers remained predominantly manual — executed by separate teams across disconnected systems, with no automated integration layer. Exceptions such as stockouts, supplier delays, and demand spikes were detected reactively, limiting the organisation's ability to respond proactively at scale.

Before — Manual workflow
manually analyse sales data & forecast demand per outlet category manager manually check stock levels, detect low stock & spoilage store & warehouse staff manually calculate order quantities & raise purchase orders procurement staff manually email/call 3,000+ suppliers, confirm availability procurement staff manually plan DC dispatch & allocate stock to outlets DC coordinator All steps performed manually · No automation layer · Delays compound at every handoff
After — Flowr agentic workflow
FLOWR — AGENTIC AI SUPPLY CHAIN WORKFLOW AGENT 01 Demand Forecasting Agent Sales data · Loyalty signals · Seasonal per SKU MCP Server fetch sales, loyalty data Purchase & Sales Data Historical · Real-time <forecast data> AGENT 02 Inventory Monitoring Agent Real-time stock · Low stock & spoilage detection MCP Server fetch inventory data Inventory Data Stock levels · Spoilage signals replenishment signals AGENT 03 Procurement and Order Agent Order quantities · Supplier selection · PO gen MCP Server fetch supplier catalog Supplier Data Catalog · Pricing · Lead times draft purchase orders AGENT 04 Supplier Coordination Agent Transmit POs · Confirm availability & slots Communication Channels ERP · Outlook · EDI · Supplier portals confirmed orders + schedule AGENT 05 DC Replenishment Agent Stock allocation · Route sequencing · Dispatch MCP Server fetch fleet availability Fleet Data Vehicles · Routes · Capacity replenishment plan AGENT 06 Exception Handling Agent Anomaly detection · Prioritised alert gen Monitoring System Dashboards · Alerts · Escalation publish alerts Human role: supervise, validate, steer — not execute LLM CONSORTIUM — Llama-3 · Mistral · Qwen (fine-tuned) · GPT-OSS · Ollama
Outcomes

What the transition delivered

↓
Reduced coordination overhead
Manual handoffs between teams eliminated. Agents pass structured outputs directly across workflow stages.
⇌
Improved demand-supply alignment
Continuous per-SKU, per-outlet forecasting replaced periodic spreadsheet analysis across the outlet network.
◎
Proactive exception handling
Exceptions detected and escalated before impact. Stockouts, supplier delays, and spoilage risks surfaced in real time.
Tech stack
OpenAI Agents SDK Claude Code GPT-OSS Llama-3 Mistral Qwen LoRA / QLoRA Ollama MCP LM Studio
Research paper
Flowr — Scaling Up Retail Supply Chain Operations Through Agentic AI →