AI assistant for 20,000 insurance agents

Find out how we helped ERGO Hestia use Microsoft Foundry to create an AI assistant for more than 20,000 insurance agents.

ERGO Hestia is a leading insurance company operating on the Polish market, serving millions of individual and business customers. The insurer supports its customers by offering a wide range of products, such as motor, property and health insurance. The company works with tens of thousands of insurance agents.

Existing agent chat, Microsoft Azure tenant

// Solutions

End-to-End Copilot Experience

Copilot Agents

Custom AI Solutions with Microsoft Foundry

Business Metrics Forecasting with Machine Learning

AI Governance

Microsoft Purview: Data Security and Governance

Threat Protection with Microsoft Defender XDR

Azure Cloud and Microsoft 365 security hardening

// Industry

Insurance

// Stack

Microsoft Foundry

Azure

Microsoft Entra ID

The challenge

Agents’ questions about the General Terms and Conditions had to be answered from multi-page documents and overlapping product versions. The target: at least 60% of traffic captured by the chatbot.

The decision to begin working with Onex Group stemmed from the need to address three key challenges:

Handling over 60% of the traffic at the agents’ service centre related to questions about the General Terms and Conditions of Insurance, in order to speed up and streamline the response process.
Analysis of multi-page documents, including general terms and conditions. The knowledge base included product information, marketing materials and highly detailed documents, including the General Terms and Conditions of Insurance with their appendices and regulations.
Versioning of the product offering. The changing product offering resulted in overlapping terms and conditions and coverage scopes for insurance products.

What goals did we set?

Key success KPI: at least 60% of traffic captured by the chatbot.

Improving the efficiency of insurance agents
Improving customer service quality
Reducing the time to market for new products

Our approach

An AI search engine built into the chat agents already use, running on ERGO Hestia’s own Azure tenant. RAG keeps every answer tied to verified source data.

Designing solutions based on real-world business processes
The project began with an analysis of how agents work and how information flows within the organisation. The key objective was to create a solution that would not change the way end users work, but would naturally enhance existing processes. Therefore, the intelligent AI search engine was integrated into the existing chat interface via a dedicated Web API, which significantly reduced the time it took for the organisation to adopt the solution and minimised the need for additional user training.
Architecture based on Microsoft Azure
The solution was designed using a PaaS architecture on the client’s Microsoft Azure tenant, which allowed for full control over the data and compliance with ERGO Hestia’s security policies.
RAG (Retrieval-Augmented Generation)
The use of the RAG (Retrieval-Augmented Generation) approach helped reduce the risk of generating incorrect information, as the model generated responses based solely on verified source data. The solution was also tailored to the organisation’s security requirements. The authentication and authorisation mechanisms were based on Microsoft Entra ID and Managed Identity, which enabled the secure deployment of the system in a production environment and ensured compliance with applicable security standards.

Solution concept

Two sample questions and the answers the assistant gives, from inspection deadlines to how the terms classify garden fixtures.

Sample questions and answers

“How often should a customer have their electrical installation inspected?”

The customer must inspect the technical condition of the electrical installation every 5 years. The inspection should cover the operational condition of connections, fittings, safety devices and electric shock protection measures, as well as the insulation resistance of conductors and the earthing of the installation and equipment. The inspection must be performed by individuals who hold the qualifications required to supervise the operation of equipment, installations and power grids.

“What category does a lawn sprinkler system and solar-powered lights fall under?”

A lawn sprinkler system and solar-powered lights are considered part of garden architecture. Garden architecture includes, among other things, irrigation systems and garden lights, which are considered garden fixtures as defined in the General Terms and Conditions.

Results

68%

of the traffic received regarding the General Insurance Terms and Conditions

~90%

of conversations with no reported issues

20,000

users can use the solution

Following the implementation of an intelligent AI search engine, measurable business results were achieved, which led to a change in the agents’ work model and the way enquiries are handled within the organisation.

In practice, the traffic acquisition rate resulted in a significant reduction in the workload on support teams and the ability to scale services without increasing headcount. At the same time, the high quality of responses was the result of using the RAG architecture and a multi-stage response quality testing process, which allowed us to verify the accuracy of responses and their compliance with documentation.

An additional business benefit was the reduction in operating costs and the ability to roll out new products more quickly, since product information was automatically available to agents once the source documents were updated.

Artificial intelligence now accounts for as much as 68 percent of the traffic related to general insurance terms and conditions, with around 90% of conversations with the bot having no reported issues. However, what pleases us most is that our insurance agents and brokers have received tangible support, which has significantly improved our customer service process.

Paweł Czaja

Deputy Director of Customer Service, ERGO Hestia

Summary

The assistant handles 68% of the traffic regarding the General Insurance Terms and Conditions, with around 90% of conversations having no reported issues.

A production AI assistant that answers from verified insurance terms, runs in the client’s own Azure tenant and is available to 20,000 users.