KPMG Australia boosts productivity with generative AI agent for employees and clients

KPMG Australia developed and deployed KymChat, a generative AI chat agent, to help its 10,000 employees access, summarize, and generate insights from both internal and external unstructured data sources. Leveraging Azure OpenAI (GPT-3.5, GPT-4), Azure Cosmos DB for MongoDB vCore (with vector search), Azure App Service, and Azure Container Registry, KymChat was rapidly scaled across the business. The deployment resulted in a substantial leap in search quality (from 50% to 91%) and reduced average response times to under a second. About 70% of employees adopted the tool, which has processed over 120,000 queries. KymChat is now being expanded for client offerings and to other KPMG member firms worldwide. The solution also includes user onboarding, prompt engineering training, and hackathons to encourage new use cases.

The project was enabled by close collaboration with Microsoft engineering teams and involved significant integration between cloud, containerization, and advanced AI search capabilities. KymChat was designed for robust data security and privacy, ensuring sensitive information stays within the firm.

The impact has included up to 50% productivity improvement in areas such as document drafting, communication, sales proposal development, and research, creating more time for client interaction and fostering a culture of continuous learning.

1. 01Employees struggled to efficiently surface insights from large amounts of unstructured internal and external data.
2. 02Traditional search and information retrieval was slow and yielding low-quality results, with a search quality baseline around 50%.
3. 03Frequent manual and repetitive tasks reduced productivity in document drafting, communication, proposal development, and research workflows.
4. 04Scaling AI solutions led to increasing response times and inconsistent result quality.
5. 05A need for secure, organization-controlled deployment of AI tools to meet privacy requirements.

1. 01Developed and launched KymChat, a generative AI agent, using Azure OpenAI (GPT-3.5 and GPT-4).
2. 02Integrated Azure Cosmos DB for MongoDB vCore with vector search for semantic understanding and rapid retrieval of unstructured data.
3. 03Deployed backend code via Azure Container Registry and Azure App Service for scalable, containerized delivery.
4. 04Incorporated prompt engineering training and user onboarding for employees.
5. 05Close collaboration with Microsoft engineering to optimize architecture and solution reliability.

1. 01Search quality increased from 50% to 91%.
2. 02Average response times reduced to under one second across use cases.
3. 0370% employee adoption rate with 120,000+ queries processed in the first phase.
4. 04Productivity increased by up to 50% in key workflows.
5. 05Expanded to client offerings and global KPMG firms, demonstrating high scalability.

• Improved search quality for users
• Reduced response times across use cases
• Expanded AI chat agent into client offerings
• Scaled AI solution across global firms