AI Agent for Inventory Management & Pricing in a Retail Chain
AI AGENTIC DEVELOPMENT
A large Ukrainian retailer with 120+ stores and an e-commerce platform faced a major operational bottleneck:
Pricing and replenishment decisions were being made manually by category managers across dozens of product categories. Demand forecasting errors systematically led to surplus illiquid stock and chronic stockouts on high-velocity SKUs — costing the business both margin and customer loyalty.
The client approached Artjoker with the task of building an autonomous multi-agent AI system capable of monitoring inventory levels, forecasting demand, dynamically adjusting prices, and triggering procurement processes with minimal human involvement.
15,000+ active SKUs across food, non-food, and seasonal categories required daily pricing decisions
Overstocked items with approaching expiry dates generated $180K/month in markdown losses
Category managers spent 60–70% of their working time on routine data exports and manual price adjustments — instead of strategic work
The existing ERP system had no predictive layer — only reactive reporting
Stockout rate on top-selling SKUs stood at 8.3%, directly suppressing revenue
Pricing rules were scattered across Excel files with no consistency between stores or regions
Artjoker designed and deployed a multi-agent AI system integrated with the client's existing ERP and PIM:
Analyzes sales history, seasonality, local events, and external signals (weather, holidays, competitor promotions) to forecast demand at the SKU level 7–30 days ahead.
Uses a RAG-pipeline enriched with category manager notes and supplier lead-time windows.
Automatically triggers purchase orders and warehouse transfers based on forecasted demand, current stock levels, and configured safety-stock thresholds.
Monitors competitor pricing, stock levels, and market signals to recommend and auto-apply price changes via ESL API and Shopify integration.
LangGraph-powered coordination layer that manages agent-to-agent communication, resolves conflicts between pricing and replenishment decisions, and maintains a shared state across all agents.
Dashboard interface for category managers to review, approve, reject, or override agent recommendations — with full audit trail and reasoning logs.
Stack
-
AI / LLM
Claude Sonnet, fine-tuned forecasting models
-
Agent Framework
LangGraph, custom orchestration layer
-
Data Integration
REST API + webhooks to ERP (1C), PIM, WMS
-
Price Updates
ESL API (electronic shelf labels), Shopify integration
-
Competitor Monitoring
Custom scraping agents, Oxylabs proxy infrastructure
-
Infrastructure
AWS ECS, PostgreSQL, Redis
-
Observability
LangSmith, custom audit dashboard
-
Notifications
Slack Bot, email digest
-
Metric
- Stockout rate (top SKUs)
- Markdown losses (overstock)
- Manager time on routine pricing
- Price update cycle
- Procurement errors
- Gross margin improvement
-
Before
- 8.3%
- $180K/mo
- 65%
- 2–3 days
- ~120/mo
- —
-
After
- 2.1%
- $41K/mo
- 12%
- Real-time
- ~9/mo
- —
-
Change
- 74%
- 77%
- 81%
- Instant
- 93%
+2.4 pp
The agents do the work that used to require three full-time analysts. Our managers now focus on supplier negotiations and category strategy — the system handles all the routine.
-
Multi-agent orchestration with conflict resolution between pricing and replenishment decisions
-
Real-time ESL integration for instant in-store price synchronization
-
RAG-enriched forecasting — combining structured ERP data with unstructured manager knowledge
-
Human-in-the-loop architecture: agents handle routine tasks, escalating edge cases to humans.
Kashcheiev Maksym
Head of Business Development
contact us:
USA
Estonia
Ukraine
Norway
