fb-pixel
Back to Case studies

Terveystalo: A customer service AI solution to improve information retrieval and boost employee experience

Together, Terveystalo and Futurice developed an AI solution, Nero, that consolidates information from various sources into a single, user-friendly platform, enabling efficient and intelligent search capabilities. We closely collaborated with customer service agents to ensure the solution met their needs. As a result, agents quickly adopted the new tool, significantly improving their efficiency and overall speed of customer service.

Terveystalo ASKE GenAI case study header image of a health practitioner on tele consultation

Technologies used

  • Natural Language Processing (NLP)
  • Large Language Models (OpenAI)
  • Advanced Prompt Engineering
  • Entity Recognition
  • Structured Information Extraction
  • Azure AI Search
  • Azure OpenAI
  • Unified Data Model/Schema
  • Python

The challenge

Terveystalo’s customer service agents handle approximately 2.5 million customer calls annually, often requiring access to dozens of systems and tools to find contract information, locations, and the right experts. The primary requirement for this project was to achieve fast search speeds in customer service cases. It was essential that the available information be reliable and up-to-date and that the system function efficiently under high user volumes, as more than 400 agents might be searching for information simultaneously.

Daily challenges in customer service include verifying contract and payment details, which require using multiple tools and sources to access accurate and up-to-date information. Agents frequently handle information related to locations and experts, such as doctor profiles, office locations, hours of operation, and accessibility details. All this information needs to be provided quickly and reliably.

Customer service relies heavily on experienced employees familiar with complex systems and information retrieval environments. Much of this expertise is implicit, acquired through years of experience. Training new employees can take months, impacting efficiency, adding costs, and lowering employee and customer satisfaction.

Impact and outcomes

  • Rapid adoption by customer service agents

  • Significantly reduced call durations

  • Increased customer and employee satisfaction

What we did

The project was carried out in two phases. In the first phase, a four-week proof of concept (PoC) trial, we validated that the planned solution would ease the work of customer service agents. In the second phase, we expanded the solution to include additional data systems and created a version suitable for production.

We collaborated closely with customer service agents using an agile development model, collecting feedback at each project stage. We used this feedback to continually refine and improve the application in partnership with Terveystalo's agents.

During the first phase, we focused on understanding the scope of the problem and selecting the most critical use cases. We initially identified the tools and information sources most frequently used by customer service agents. Within two weeks, we developed the first version of an AI solution called Nero, which agents could immediately begin testing. An active group of agents used the tool in real work situations and provided feedback to the development team. Simultaneously, we conducted a business case analysis to estimate the potential benefits and costs, supporting Terveystalo’s investment decision.

Nero expanded Terveystalo's search capabilities, allowing agents to retrieve information from systems previously entirely lacking search functionality. We designed the new search tool to be easy for personnel to adopt by making it operate familiarly. We also developed more sophisticated AI-powered tools that agents could use concurrently with the traditional search methods. The AI search understood context and enabled queries in natural language. It organized unstructured source data to improve search functions. For example, when an agent searches for an 'active contract,' the system can immediately differentiate it from inactive contracts. Previously, this would have required multiple searches or manual checks.

Throughout the project, we focused heavily on helping agents transition to the new information retrieval method, bringing them into the development process and providing thorough training on the tool.

In the second phase, we expanded the services by adding new data sources in collaboration with Terveystalo's data team. The updated application needed to provide broader and more precise searches quickly and accurately, even with hundreds of concurrent searches. This required a careful consideration of how large language models (LLMs) could be utilized to maintain the application's speed, accuracy, and cost-effectiveness. Given the high volume of searches, using LLMs to process search results wasn't practical — instead, we employed pre-processing to structure natural language information, thereby simplifying the search process.

Throughout the project, we collaborated with Terveystalo's architects to design a scalable AI architecture and built the production environment in Microsoft Azure. This architecture offers a cost-effective, centralized method to manage AI resources within Terveystalo's cloud environment. It facilitates the easy addition of new data sources and the development of applications for new use cases using existing resources.

The illustration on the work by Futurice with Terveystalo GenAI POC process flow.

Terveystalo has a highly developed cloud environment and substantial internal expertise in cloud technology. This allowed the Nero project to be executed swiftly and efficiently by leveraging existing practices and integrating the new AI solutions into Terveystalo's overall architecture.

The final solution is a browser-based Nero search tool that allows agents to retrieve information from multiple sources simultaneously. Previously, agents had to use several different systems and search for information manually during calls. Now, a single search retrieves information from multiple places, with all results displayed in one place.

Our contact center employees received a new AI-assisted tool this year, which has made customer interactions smoother and more enjoyable since information can now be quickly found in one place. Good and rapid collaboration with Futurice was crucial– the project team ensured agents were involved from the start in testing and developing solutions that genuinely facilitate their work.
Titta-Liisa Luoma
Customer Service Director, Terveystalo

Why it matters

The Nero search tool was launched in the fall of 2024, and agent feedback has been positive. The solution has significantly streamlined and improved customer service work. Agents especially appreciate being included in the development process from the beginning.

Preliminary measurements show that accelerated information retrieval has noticeably shortened call times. Monitoring indicates that the time spent on information retrieval has decreased from one minute to just 20-40 seconds in many cases. Given that agents handle about 2.5 million customer calls annually, the optimized search saves significant work time. Preliminary calculations suggest the project investment will pay for itself within the first year of operation.

About the client

Terveystalo is Finland’s largest private healthcare provider in terms of revenue and network, as well as the leading player in occupational health services in the Nordics. In 2023, Terveystalo served 1.2 million individual customers in Finland, with approximately 7.6 million visits. Terveystalo employs over 15,500 health and wellness professionals, and its revenue in 2023 was €1.286 billion.

Contact person

Get in touch

Wherever you are on your digital journey, our services and experts can help you along the way. Let's get talking!