Case study
From Weeks to Under 15 Seconds: An Enterprise Agentic AI Analytics System for a Leading APAC FMCG Corporation
- Team size: 11 (4 Data Engineers, 3 AI Engineers, 2 Data Stewards, 1 Solution Architect, 1 PMO)
- Development time: 9 months across two phases
Explore more case study
Background
As the FMCG industry across APAC becomes increasingly complex, enterprises must navigate multiple markets, diverse operating models, and varying growth priorities. Bain & Company notes that several emerging trends are reshaping the world’s largest consumer products market, requiring CPG companies to effectively manage the region’s complexity to achieve sustainable growth. McKinsey further highlights the diversity and fragmentation of the APAC landscape through a study of 25,998 consumers across 18 markets, representing approximately 75% of global GDP. In parallel, Gartner emphasizes the growing role of augmented analytics powered by natural language processing (NLP) and conversational interfaces, enabling non-technical business users to access data and generate insights more easily.
In this case, the client is a leading FMCG corporation operating across the APAC region, managing multiple operating companies in more than 20 countries.
To address these challenges, GEM partnered with the client to build an agentic AI analytics system that would allow business users to ask questions in plain language and receive validated, explained, and visualized answers in seconds. The resulting agentic AI analytics system was built on three foundations: a harmonized data foundation, an enterprise semantic layer, and an agentic AI intelligence layer.
Challenges
Data fragmentation and inconsistent definitions across markets
- Data was scattered across many source systems and operating companies, with no common enterprise model to unify it.
- Each market structured its data differently, making cross-market analysis a manual reconciliation exercise.
- Core KPIs were defined inconsistently from market to market, so the same question could produce conflicting answers.
- Large volumes of data were collected, but the effort required to access and trust it limited how often it was actually used.
Slow, technical, and inflexible access to insight
- Business users could not query data directly, every non-standard question required an analyst and a hand-written SQL query.
- Static dashboards answered known questions but could not support open-ended, exploratory analysis.
- Decision cycles stretched from hours to days or weeks, eroding the value of the insight by the time it arrived.
- Any solution had to be enterprise-grade: trustworthy, governed, secure, and accurate enough to support real commercial decisions.
Solution
GEM delivered an agentic AI analytics system that transforms a natural-language business question into a validated, explained, and visualized answer in under 15 seconds. The agentic AI analytics system is structured around three interconnected layers, with a multi-agent reasoning framework at its core.
Layer 1: Harmonized Data Foundation
Data from every operating company and country is ingested and unified under a common enterprise data model, creating a single trusted source for analytics and AI workloads. This foundation enables the agentic AI analytics system to perform consistent cross-market analysis while maintaining governance and scalability.
Layer 2: Enterprise Semantic Layer
The semantic layer standardizes KPIs, dimensions, and business logic across all markets. By embedding enterprise definitions directly into the platform, the agentic AI analytics system ensures that every analysis is based on consistent business rules regardless of geography or business unit.
Layer 3: Agentic AI Intelligence
At the core of the agentic AI analytics system is a multi-agent architecture capable of understanding business intent, planning analyses, querying trusted data sources, generating insights, and creating visualizations. The result is a conversational analytics experience that combines speed, context, and reliability.
Within this layer, a team of specialized agents collaborates on every question:
- Agent Orchestrator: controls the workflow and coordinates handoffs between agents.
- Planner Agent: interprets the user’s question and creates a structured analysis plan.
- Context Agent: retrieves the relevant business context — KPIs, dimensions, and historical knowledge.
- Query Agent: generates and executes optimized data queries against trusted sources.
- Analysis Agent: analyzes the retrieved data to surface trends, anomalies, and key insights.
- Judge Agent (LLM-as-Judge): validates the reasoning, data usage, and results to ensure every response is accurate and reliable.
- Short-Term Memory: holds session context and intermediate results across a conversation.
- Long-Term Memory: stores reusable knowledge so the system improves over time.
Enterprise-grade platform and security
Because the system handles sensitive commercial data across more than 20 countries, security and governance were designed in from the start. The platform runs inside an Azure Virtual Network with Private Endpoints, authenticates users through Microsoft Entra ID, and is protected by an Azure Application Gateway and Firewall. The Judge Agent adds a reasoning-level guardrail on top of the infrastructure controls, validating outputs before they ever reach a user.
Tech stack
- Cloud platform: Microsoft Azure
- Data foundation: Azure Storage (ADLSv2), Azure Data Factory, Databricks, Delta Lake
- Semantic layer: Databricks Metric Views, Azure AI Search
- Agentic intelligence: Azure App Service, Azure OpenAI, Azure AI Search, Redis Cache, Azure SQL Database, Azure Key Vault, LangGraph
- Security & networking: Azure Virtual Network & Private Endpoints, Microsoft Entra ID, Azure Application Gateway & Firewall
Output
- Delivered a production-grade agentic AI analytics system capable of answering natural-language business questions across more than 20 APAC markets.
- Built a harmonized data foundation on Databricks and Delta Lake, unifying data from multiple operating companies under a single enterprise model.
- Implemented an enterprise semantic layer that enforces governed KPI definitions and business logic across all markets.
- Deployed a multi-agent reasoning architecture with Orchestrator, Planner, Context, Query, Analysis, and Judge agents.
- Embedded LLM-as-Judge validation capabilities to ensure every response generated by the agentic AI analytics system is accurate, explainable, and trustworthy.
- Enabled conversational analytics, autonomous SQL generation, automated insight narration, and dynamic visualizations.
Impacts
Operational Efficiency
- Reduced time to insight from days or weeks to under 15 seconds through the agentic AI analytics system, enabling real-time exploratory analytics.
- Achieved 99.8% data retrieval accuracy when validated against trusted enterprise metrics.
- Eliminated analyst bottlenecks for ad hoc reporting and business questions.
- Standardized KPI definitions across all markets, significantly reducing reconciliation efforts.
Business Transformation
- Democratized data access through the agentic AI analytics system, allowing sales representatives, sales managers, and brand leaders to explore enterprise data without SQL or technical expertise.
- Accelerated decision-making by providing immediate access to trusted insights.
- Increased confidence in enterprise reporting through validated and explainable AI-generated answers.
- Established a scalable data and AI foundation for future analytics, automation, and AI initiatives across APAC operations.
Closing remarks
By reducing the distance between a business question and a trustworthy answer from weeks to under 15 seconds, the agentic AI analytics system delivered far more than faster reporting. It created a data-driven operating model across more than 20 markets. Through harmonized data, governed business definitions, and a self-validating multi-agent architecture, the agentic AI analytics system transformed a fragmented data landscape into a single conversational source of truth and established a scalable foundation for the next generation of AI-powered decision-making.
Discover another AI transformation success story: Intelligent shipper and carrier matching platform
