Microsoft Store Assistant leverages Azure OpenAI and Semantic Kernel to deliver AI-driven customer support and increase sales
Microsoft replaced a legacy rule-based chatbot on Microsoft Store with Microsoft Store Assistant, powered by Azure OpenAI, Semantic Kernel, and real-time page context. The assistant comprehends Microsoft's extensive product portfolio and intelligently routes complex queries to human sales associates. It manages millions of conversations annually, maintains a 4-star satisfaction rating, and has generated revenue exceeding 140% of its forecast. The multi-expert AI agent architecture uses Semantic Kernel skills coordinated for optimal conversational flow and real-time intelligence from Microsoft's product pages. Azure AI Foundry enables continuous testing and safety evaluations while Azure Cosmos DB, Azure Functions, and Power BI support conversation analysis and metrics visualization.
- Industry
- Retail
- Location
- United States
- Published
- April 2025
Reported outcomes
140%
revenueRevenue & growth
Strategic outcomes
Catalog median for revenue & growth deployments: +34% across 150 reported metrics. Compare benchmarks →
- Legacy chatbot was rule-based, costly to maintain, and poorly rated by customers.
- Human agents were overwhelmed with off-topic or incomplete chats the legacy bot could not handle.
- Updates lagged behind product launches, causing customer dissatisfaction and operational inefficiencies.
- Developed Microsoft Store Assistant using Azure OpenAI GPT-4o models and Semantic Kernel multi-expert orchestration.
- Integrated real-time page context and Azure AI Search product catalog data to enhance responses.
- Implemented Azure AI Foundry for continuous functional and safety testing.
- Used Azure Cosmos DB, Azure Functions, and Power BI for conversation analysis and monitoring.
- Implemented Azure Content Safety for redacting personal information and ensuring safe conversations.
- Manages several million conversations annually with a consistently high customer satisfaction rating (4 stars).
- Exceeded revenue forecasts by 140%.
- Reduced human transfers by 46%, improving operational efficiency.
- Enabled touchless product releases using real-time detailed context integration.
Architecture
The solution uses Semantic Kernel for multi-expert orchestration with a Coordinator that plans and invokes expert skills for each conversational turn. These experts leverage enrichment plug-ins providing context such as chat history, latest user message, current page context, and rules to follow. Real-time product catalog data from Azure AI Search is integrated for accurate product recommendations and support. The system undergoes continuous evaluation and safety testing with Azure AI Foundry. Conversation analytics are implemented with Azure Cosmos DB, Azure Functions, and Power BI.
Sources & evidence1
AI-generated summary. Verify important details with the linked sources before relying on this case.
Explore related AI use cases
Was this useful?