PZU tests AI in practice: 94% accurate answers in a pilot project

Smiling family of four outdoors under a clear blue sky, with two adults carrying their children on their backs while everyone stretches their arms out playfully as if flying.
pzu-logo

The Challenge

 

Between documentation and customer service

Insurance is a document-driven business. Policy terms and conditions, clause matrices, sales presentations, and internal procedures represent thousands of pages of dispersed knowledge that continuously evolve as new product versions are introduced.

This created several challenges for PZU employees:

  • Excessive time spent searching for answers: Manually reviewing documents during customer interactions was inefficient and extended service times.

  • Lack of consistency: Different sales channels, including agents, call centers, and bancassurance teams, provided answers in varying ways.

  • Scalability challenges: As new products were introduced, managing organizational knowledge became increasingly complex.

To address these challenges, PZU decided to test SUNZINET’s AI Knowledge Assistant.

Strategy: Testing in a real business environment

  • Two colleagues are standing in a modern office corridor, looking together at a smartphone held between them. The woman is wearing a light pink blazer paired with dark trousers, whilst the man is wearing a white shirt and dark trousers. In the background is a glass wall covered with colourful sticky notes, creating the atmosphere of a collaborative workshop or a work environment involved in project planning.

    As part of the Proof of Concept, SUNZINET and PZU conducted 2 parallel testing streams. The first focused on validating the quality of answers generated by the assistant. Approximately 50 insurance and product-related documents were uploaded to the knowledge base, including policy terms and conditions, clauses, and sales support materials covering selected motor, home, and travel insurance products.

  • A man and a woman are sitting close together on a grey sofa in a modern office lounge area, looking at the open laptop on the man’s lap. The man is wearing a light beige overshirt, blue jeans and white trainers, whilst the woman is wearing a black blazer, black trousers and white trainers. The dark cushions, wooden floor and neutral colour scheme of the walls create a relaxed, collaborative working atmosphere.

    Together with PZU product experts and trainers, test question sets were developed to systematically evaluate the solution's effectiveness. Across 3 pilot iterations, answer quality was analyzed, areas for improvement were identified, and the system was continuously refined.

  • Three people are gathered in a friendly, modern office space. A woman with short blonde hair is holding a smartphone and showing something to the man sitting next to her, who is looking at the screen with a smile. The woman on the right, with dark hair, is also smiling and appears to be engaged in the conversation. In the background, you can see plants and a stylish interior with a reflection of the kitchenette. The whole scene conveys an atmosphere of collaboration, open communication and positive relationships within the team.

    The second testing stream involved future users of the solution, including sales employees and pilot program participants. Several dozen users worked with the assistant in realistic business scenarios, evaluating its usefullness, ease of use, and potential to support everyday tasks.

Two colleagues are standing in a modern office corridor, looking together at a smartphone held between them. The woman is wearing a light pink blazer paired with dark trousers, whilst the man is wearing a white shirt and dark trousers. In the background is a glass wall covered with colourful sticky notes, creating the atmosphere of a collaborative workshop or a work environment involved in project planning.
A man and a woman are sitting close together on a grey sofa in a modern office lounge area, looking at the open laptop on the man’s lap. The man is wearing a light beige overshirt, blue jeans and white trainers, whilst the woman is wearing a black blazer, black trousers and white trainers. The dark cushions, wooden floor and neutral colour scheme of the walls create a relaxed, collaborative working atmosphere.
Three people are gathered in a friendly, modern office space. A woman with short blonde hair is holding a smartphone and showing something to the man sitting next to her, who is looking at the screen with a smile. The woman on the right, with dark hair, is also smiling and appears to be engaged in the conversation. In the background, you can see plants and a stylish interior with a reflection of the kitchenette. The whole scene conveys an atmosphere of collaboration, open communication and positive relationships within the team.

Key features

Icon of a browser window with a green magnifying glass, symbolizing website search, inspection or analysis.

Intelligent search

Answers are generated exclusively from uploaded documents and include source links and supporting excerpts.

Icon of a person surrounded by connected network nodes with a green tag, symbolizing user relationships, categorization or profile tagging.

Channel-specific personalization

Responses can be tailored to different user groups, such as insurance agents and call center consultants.

Icon of a magnifying glass focused on a settings gear, symbolizing configuration search, technical analysis or system inspection.

Answer reliability assessment

The system automatically estimates the risk of an incorrect answer and flags potentially inaccurate responses before they reach the user.

Icon of a browser window displaying image thumbnails with a green action symbol, representing image or media content management.

Client-side administration panel

PZU teams can independently manage documents, versions, knowledge bases, and users without relying on a technology partner.

Icon of multiple user profiles surrounding a central settings gear, with one user highlighted in green, symbolizing user management, roles or access configuration.

Feedback loop

Users can rate answers as correct, neutral, or incorrect and report issues, enabling continuous model improvement.

Icon of a search field and list with a magnifying glass, with one result highlighted in green, symbolizing search and content discovery.

Conversation history

The assistant maintains context throughout a user session, making follow-up questions more effective and natural.

Illustration of a man and a friendly robot working on laptops, surrounded by performance metrics showing a 4.5/5 solution rating, 396 questions submitted, 94% correct answers and a response time under 25 seconds.

Results

94% accurate answers and strong user adoption

The Proof of Concept delivered measurable results that clearly demonstrated the solution’s business value.

Quantitative results:

  • Questions submitted to the assistant: 396

  • Average response time: under 25 seconds

  • Correct answers: 94%

  • Incorrect answers: 6%

  • Users who considered the solution valuable for sales work: 94.29%

  • Average solution rating: 4.5/5

Illustration of a person interacting with a large digital screen displaying a friendly robot, surrounded by the words “Speed,” “Consistency,” “Scalability” and “Accuracy” to represent key benefits of AI-powered automation.

Qualitative results

The successful Proof of Concept confirmed that the AI assistant can create a tangible impact across several critical business areas:

  • Speed – Access to information was reduced from several minutes to just a few seconds, resulting in smoother customer interactions.

  • Accuracy – With 94% correct answers, source references, and automatic risk assessment, the risk of providing outdated or inaccurate information was significantly reduced.

  • Consistency – A unified standard of responses across all sales channels, including agents, call centers, and bancassurance teams.

  • Scalability – New products and documents can be added easily by PZU teams without involving a technology partner.

  • Onboarding support – New employees gain immediate access to verified knowledge, accelerating and simplifying onboarding.

  • Future-proof architecture – The underlying LLM can be replaced with newer models without rebuilding the entire solution.

Marcin_Kurczab_PZU
Marcin Kurczab Director of Innovation & AI, PZU

It was the transparency and quality of answers that convinced us

A key objective of the project was to find a solution that not only provides answers quickly, but above all delivers answers that can be trusted. The Proof of Concept results confirmed that the AI assistant developed by SUNZINET effectively supports employees in working with insurance documentation, ensuring a high level of answer quality and full transparency of sources. These qualities proved to be the most important from both the users’ and the organization’s perspective.

About PZU

PZU Group is one of the largest financial and insurance institutions in Central and Eastern Europe, with more than 200 years of market presence. The company serves millions of customers in Poland and internationally, offering a comprehensive portfolio of insurance products, including motor, home, business, travel, life, and investment insurance. PZU employs thousands of professionals across its sales network, including agents, call center consultants, and bancassurance advisors.

Do your employees spend too much time searching for information across documents?

We help organizations turn scattered documents, procedures, and product materials into accessible knowledge resources that support employees in their daily work. Let's discuss how we can help you improve the speed, consistency, and quality of answers across your organization.

Simply fill out the form to schedule an initial, no-obligation consultation, and we'll get back to you within 24 business hours.

Dominik Krusemark - Business Development Manager - SUNZINET

Dominik Krusemark

Business Development Manager

krusemark.dominik@sunzinet.com