# GPC Asia Pacific — Client POC Definition Prompt
## Agentic AI for Retail Auto Parts: Inventory, Demand, Promotions & Virtual Assistant

**Date:** March 27, 2026
**Client:** GPC Asia Pacific (gpcasiapac.com)
**Engagement Type:** Proof of Concept (POC)
**Industry:** Retail — Automotive Parts & Accessories
**Status:** PHASE 1 — Discovery & Definition
**Prepared by:** Kyndryl Agentic AI Practice

---

## EXECUTIVE SUMMARY

GPC Asia Pacific is one of the region's leading distributors of automotive parts, accessories, and workshop equipment. With thousands of SKUs, a complex multi-warehouse distribution network, seasonal demand spikes, and high-volume promotional campaigns, GPC AsiaPac faces significant operational challenges where Agentic AI can deliver measurable, high-ROI outcomes.

This POC Definition Prompt defines the scope, use cases, agent architecture, and success criteria for a multi-agent AI system addressing GPC AsiaPac's core retail operational challenges.

**Key Framing:** *"Your ERP gives you data. Kyndryl helps you turn data into autonomous retail decisions and outcomes."*

---

## CLIENT PROFILE

| Attribute | Detail |
|-----------|--------|
| **Organisation** | GPC Asia Pacific (gpcasiapac.com) |
| **Industry** | Retail — Automotive Parts, Accessories & Workshop Equipment |
| **Business Model** | B2B (trade / workshop accounts) + B2C (retail consumers) |
| **Geographic Footprint** | Australia, New Zealand, Singapore, and broader Asia Pacific |
| **Key Brands** | Repco, NTG, Rare Spares, Motion Industries (APAC) |
| **Operations** | Multi-warehouse distribution, franchise retail network, e-commerce |
| **Primary AI Opportunity** | Inventory intelligence, demand signal aggregation, promo error prevention, customer self-service |

---

## STRATEGIC AI OPPORTUNITIES

### **Use Case 1: Intelligent Inventory Management Agent** 📦

**Executive Framing:** Eliminate stockouts and overstock — autonomously rebalance inventory across the network.

**Current State Pain Points:**
- Manual reorder processes with static min/max thresholds
- Warehouse stock imbalances — overstock in some DCs, stockout in others for same SKU
- Excess working capital locked in slow-moving parts
- No real-time visibility across franchise & company-owned stores

**AI-Powered State:**
- Agent continuously monitors stock levels across all warehouses and stores
- Automatically triggers inter-warehouse transfer recommendations when imbalances detected
- Raises purchase orders for replenishment based on reorder logic + demand signals
- Flags dead stock and generates clearance recommendations
- **Human-in-the-Loop:** Manager approves transfers/POs before execution

**Agent Capabilities:**
- **Stock Intelligence:** Real-time SKU visibility across all nodes
- **Reorder Automation:** Dynamic reorder point calculation (replaces static rules)
- **Inter-DC Balancing:** Identifies transfer opportunities vs. new supplier orders
- **Dead Stock Detection:** Flags SKUs with >90 days no movement
- **Supplier Integration:** Drafts PO recommendations with lead time awareness

**Expected Outcomes:**
- ✅ 20–30% reduction in stockout events
- ✅ 15–25% reduction in overstock / excess inventory value
- ✅ 10–20% improvement in inventory turn rate
- ✅ 40% reduction in manual reorder effort

---

### **Use Case 2: Demand Forecasting Agent** 📈

**Executive Framing:** Predict what parts will be needed, where, and when — before customers ask.

**Current State Pain Points:**
- Forecasting driven by historical averages, missing seasonal and event-driven demand
- No integration of external demand signals (weather, vehicle registration trends, racing calendar, harvest seasons for agricultural parts)
- Promotional uplift poorly estimated, leading to under/over-ordering
- New product introductions lack demand baseline

**AI-Powered State:**
- Agent ingests internal sales history, promotional calendars, seasonal indices, and external market signals
- Generates SKU-level, location-level 4/8/12-week rolling forecasts
- Identifies emerging demand trends (new vehicle models driving fitment parts demand)
- Integrates with inventory agent to pre-position stock ahead of forecast demand spikes

**Agent Capabilities:**
- **Time-Series Forecasting:** ARIMA/ML hybrid per SKU/location
- **External Signal Ingestion:** Weather, ANZAC/Easter/school holiday calendars, motorsport events
- **Promo Lift Modelling:** Quantify expected uplift per promotion type and SKU
- **New Product Introduction (NPI) Forecasting:** Analogue-based models for new SKUs
- **Exception Alerting:** Flags forecast vs. actual variance > threshold for human review

**Expected Outcomes:**
- ✅ 15–25% improvement in forecast accuracy (MAPE reduction)
- ✅ 20% reduction in emergency/expedited orders
- ✅ Better promotional inventory pre-positioning
- ✅ Reduced new product launch stockouts

---

### **Use Case 3: Promotional Error Detection Agent** 🔍

**Executive Framing:** Stop revenue leakage and customer trust erosion from pricing and promotion mistakes — before they go live.

**Current State Pain Points:**
- Promotional pricing errors discovered post-launch (wrong price, wrong SKU, wrong dates)
- Catalogue errors (wrong fitment data, wrong image, wrong description)
- Promotion overlaps — multiple discounts stacking unintentionally, margin erosion
- Compliance issues — advertised price vs. POS system price mismatches (ACCC risk in Australia)

**AI-Powered State:**
- Agent automatically scans every promotion configuration before it goes live
- Cross-checks: catalogue data integrity, pricing rules, date validity, margin floor compliance, fitment accuracy
- Flags conflicts between concurrent promotions (unintended stacking)
- Validates POS system prices match published/advertised prices (regulatory compliance)
- **Human-in-the-Loop:** Marketing/category manager reviews and approves flagged items

**Agent Capabilities:**
- **Pre-Launch Promo Scan:** Automated checklist run against every promo record
- **Margin Guard:** Flags any promo taking SKU below minimum margin threshold
- **Conflict Detection:** Identifies overlapping promotions for same SKU/category
- **Fitment Validation:** Cross-checks part fitment data against authoritative parts database
- **Regulatory Compliance Check:** Validates advertised price = POS system price (ACCC, Consumer Law)
- **Catalogue Integrity:** Checks images, descriptions, and part numbers for anomalies

**Expected Outcomes:**
- ✅ 90%+ reduction in post-launch promotional pricing errors
- ✅ Elimination of unintended promotion stacking / margin erosion events
- ✅ Reduced risk of regulatory non-compliance (misleading pricing)
- ✅ Faster, more confident promotional go-live process

---

### **Use Case 4: Virtual Assistant Agent (Trade & Retail)** 💬

**Executive Framing:** Give every trade customer and retail consumer a knowledgeable parts expert — available 24/7.

**Current State Pain Points:**
- High call centre volume for parts lookup, fitment queries, order status
- Workshop (trade) customers frustrated with manual quote and order process
- Retail customers abandoning purchases due to fitment uncertainty ("will this fit my car?")
- Counter staff spending significant time on repetitive lookup tasks

**AI-Powered State:**
- Conversational AI agent handles fitment queries, parts lookup, order status, quote generation
- Trade portal: Agent assists workshop managers with order history, returns, credit account queries
- Retail: Agent guides customers through vehicle-specific parts selection (year/make/model)
- Escalation to human agent with full context when complex queries arise

**Agent Capabilities:**
- **Fitment Query Resolution:** "Will this brake pad fit a 2019 Toyota Hilux SR5?" → Instant answer
- **Parts Cross-Reference:** OEM part number ↔ GPC catalogue number lookup
- **Order Status & Tracking:** Real-time order status, ETA for delivery or click & collect
- **Quote Generation (Trade):** Draft quote from spoken/typed parts list
- **Returns Processing:** Guide customer through returns eligibility and initiate RMA
- **Inventory Availability:** "Do you have X in stock in Penrith?" — Check nearest store/DC
- **Human Escalation:** Seamless hand-off to counter/call centre staff with full context

**Expected Outcomes:**
- ✅ 30–40% reduction in inbound call centre volume
- ✅ 25% improvement in online conversion (fitment confidence)
- ✅ 20% reduction in average handle time for complex queries
- ✅ 24/7 trade customer self-service capability

---

## RECOMMENDED AI AGENT ARCHITECTURE

```
┌─────────────────────────────────────────────────────────────────┐
│                    GPC AsiaPac Agentic AI Platform               │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  ┌──────────────────┐    ┌──────────────────┐                   │
│  │  Inventory Mgmt  │    │ Demand Forecast  │                   │
│  │     Agent        │◄──►│     Agent        │                   │
│  │  (reorder, DC    │    │  (SKU/location   │                   │
│  │   balancing)     │    │   forecasting)   │                   │
│  └────────┬─────────┘    └────────┬─────────┘                   │
│           │                       │                              │
│           ▼                       ▼                              │
│  ┌─────────────────────────────────────┐                        │
│  │         Orchestrator Agent          │                        │
│  │  (coordinates, prioritises,         │                        │
│  │   routes to Human-in-the-Loop)      │                        │
│  └──────────────┬──────────────────────┘                        │
│                 │                                                │
│      ┌──────────┴──────────┐                                    │
│      ▼                     ▼                                    │
│  ┌──────────────┐   ┌──────────────────┐                        │
│  │  Promo Error │   │ Virtual Assistant│                        │
│  │  Detection   │   │ Agent (Trade &   │                        │
│  │  Agent       │   │ Retail)          │                        │
│  └──────────────┘   └──────────────────┘                        │
│                                                                  │
│  ═══════════════════════════════════════════════════════         │
│  DATA LAYER: ERP │ POS │ WMS │ Parts Catalogue │ CRM            │
│  ═══════════════════════════════════════════════════════         │
└─────────────────────────────────────────────────────────────────┘
```

### Agent Priority Matrix

| Agent | Business Impact | Implementation Complexity | Recommended Phase |
|-------|----------------|--------------------------|-------------------|
| Promo Error Detection | HIGH — immediate ROI | LOW | **Phase 1 POC** |
| Virtual Assistant | HIGH — customer-facing | MEDIUM | **Phase 1 POC** |
| Inventory Management | VERY HIGH — margin impact | MEDIUM | **Phase 2 POC** |
| Demand Forecasting | HIGH — strategic | HIGH | **Phase 3 POC** |

---

## POC SCOPE — PHASE 1 RECOMMENDATION

### **In Scope (Phase 1 POC — 10 Weeks)**

| Deliverable | Description |
|------------|-------------|
| **Promo Error Detection Agent** | Automated pre-launch promotion scanning with margin guard, conflict detection, and fitment validation |
| **Virtual Assistant (Retail)** | Web-embedded conversational agent for fitment queries, parts lookup, and order status |
| **Data Integration** | ERP (SAP/JDE), POS system, Parts Catalogue API |
| **Human-in-Loop UI** | Manager review dashboard for agent recommendations |
| **KPI Dashboard** | Real-time tracking of promo errors caught, deflections, accuracy |

### **Out of Scope (Phase 1)**
- Full demand forecasting ML pipeline (Phase 2)
- Trade portal virtual assistant (Phase 2)
- Full inventory rebalancing automation (Phase 2)
- Production deployment to all regions (Phase 3)

---

## SUCCESS CRITERIA (Phase 1 POC)

| KPI | Baseline | Target | Measurement Period |
|-----|----------|--------|-------------------|
| Promo errors caught pre-launch | ~5/month manual | 95% automated detection | 4-week pilot |
| Unintended promo stacks | ~3/month | 0 | 4-week pilot |
| Virtual assistant query deflection rate | 0% (no VA) | ≥30% of fitment/lookup queries | 4-week pilot |
| Customer satisfaction (CSAT) for VA interactions | N/A | ≥4.0/5.0 | 4-week pilot |
| Time to generate trade quote | 15–20 min | <3 min | Pilot period |

---

## TECHNOLOGY STACK (Proposed)

| Component | Technology |
|-----------|-----------|
| **AI Framework** | LangChain / LangGraph (multi-agent orchestration) |
| **LLM** | Azure OpenAI (GPT-4o) |
| **Vector Store** | Azure AI Search (parts catalogue embeddings) |
| **Backend** | Python / FastAPI |
| **Frontend** | React + TypeScript (agent chat UI + manager dashboard) |
| **Data Platform** | Azure Data Factory + Azure SQL / Cosmos DB |
| **Integration** | REST APIs to ERP, POS, WMS |
| **Infrastructure** | Azure Kubernetes Service (AKS) |
| **Monitoring** | Azure Monitor + Application Insights |
| **Security** | Azure Entra ID, RBAC, data masking for PII |

---

## ENGAGEMENT TIMELINE (Phase 1 POC — 10 Weeks)

| Sprint | Weeks | Focus |
|--------|-------|-------|
| **Sprint 0** | 1–2 | Infrastructure setup, data access, API agreements |
| **Sprint 1** | 3–4 | Promo Error Agent — core scanning engine |
| **Sprint 2** | 5–6 | Promo Agent — margin guard + conflict detection + UI |
| **Sprint 3** | 7–8 | Virtual Assistant — fitment + parts lookup + order status |
| **Sprint 4** | 9–10 | Integration, KPI dashboard, pilot validation, stakeholder demo |

---

## KYNDRYL TEAM (Proposed)

| Role | Responsibility |
|------|---------------|
| **Engagement Lead** | Client relationship, stakeholder management, commercial oversight |
| **AI Architect** | Agent architecture, LLM selection, integration design |
| **AI/ML Engineer (x2)** | Agent development, LangGraph implementation |
| **Data Engineer** | ERP/POS/Catalogue integration, data pipeline |
| **Frontend Developer** | Manager dashboard, VA chat UI |
| **QA/Test Lead** | Test strategy, pilot validation |

---

## RISKS & MITIGATIONS

| Risk | Likelihood | Impact | Mitigation |
|------|-----------|--------|-----------|
| ERP/POS API access delayed | Medium | High | Early data access agreement; use synthetic data for Sprint 1 |
| Parts catalogue data quality issues | High | Medium | Data quality scan in Sprint 0; agent handles ambiguity gracefully |
| Stakeholder alignment on promo workflow | Medium | Medium | Workshop with marketing/category managers in Sprint 0 |
| LLM accuracy on technical fitment queries | Medium | High | RAG with curated parts catalogue; human escalation fallback |
| Regulatory sensitivity (ACCC pricing) | Low | Very High | Legal review of promo compliance agent outputs in Sprint 2 |

---

## NEXT STEPS

1. **Kick-off Workshop** — GPC AsiaPac stakeholders + Kyndryl team (Week 1)
2. **Data Access Agreement** — ERP, POS, Parts Catalogue API access (Week 1–2)
3. **Discovery Interviews** — Category managers, IT team, call centre team (Week 1–2)
4. **Prototype Demo** — Static HTML prototype of VA + Promo Agent UI (Week 2–3, prior to full build)
5. **Phase 1 POC kick-off** — Sprint 0 begins Week 3

---

## DELIVERABLES CHECKLIST

- [x] `client_poc_definition_prompt.md` — This document
- [ ] `gpcasiapac_discovery_transcript.html` — Stakeholder discovery interview record
- [ ] `gpcasiapac_tearsheet.html` — Executive one-pager (C-suite ready)
- [ ] `gpcasiapac_agentic_ai_poc_spec.html` — Full POC technical specification
- [ ] `gpcasiapac_prototype_spec.html` — Prototype specification (Phase 3)
- [ ] `gpcasiapac_poc_spec.json` — Machine-readable POC spec
- [ ] `index.html` — Table of Contents landing page

---

*Confidential — Kyndryl Internal Use Only | GPC Asia Pacific Engagement | March 2026*