Article
Unlocking AI’s potential: how we transformed Hartmann’s B2C chatbot
Thursday 26th of September 2024
The Evolution of Hartmann Group’s Customer Interaction
In the fast-paced world of healthcare and hygiene, staying competitive isn’t just about having the latest technology—it’s about building real connections with your customers. Hartmann Group, a key player in healthcare solutions since 1818, understood this well. With a presence in over 100 countries, they knew they had to take their customer engagement to the next level. Enter addData, a key partner under the Customer Collective umbrella. Together, we took on the challenge of overhauling Hartmann’s B2C chatbot, delivering an intelligent, speedy, and personalized customer service experience for their global audience.
The Challenge: Getting it Right
Hartmann had a clear vision: their B2C chatbot needed to provide accurate and useful information to customers in a highly regulated environment. However, their initial efforts hit a major roadblock—AI hallucinations. The chatbot often generated incorrect or irrelevant responses, frustrating users with a default message: “I cannot help you.” Unfortunately, this response was triggered in one out of every two legitimate queries. This was a huge issue, especially in a field where precise information is critical.
Hartmann needed a reliable solution—one that would ensure accurate and trustworthy communication. That’s where we stepped in. Our team worked closely with Hartmann to fine-tune their chatbot, making sure that every customer interaction was not just accurate, but also built on trust, thereby removing barriers to effective engagement.
Our Approach: Precision at Every Step
Recognizing the need for expertise, Hartmann decided to partner with us. Our approach was hands-on from the start. We started by diving deep into the data to pinpoint the chatbot’s issues. From there, we built custom data pipelines tailored specifically for Hartmann, streamlining and transforming their product information to ensure smooth integration into the Large Language Models. The result? A chatbot powered by precision and ready to deliver results.
How We Did It: Step by Step
- Data Audit: We began with a thorough audit of Hartmann’s content data, identifying gaps and inconsistencies that were holding the chatbot back.
- Custom Data Pipelines: Next, we built customized data pipelines from scratch, organizing and streamlining Hartmann’s product information so it was ready for chatbot integration.
- Addressing AI Hallucinations: Tackling AI hallucinations was our top priority. We fine-tuned the data inputs and enhanced the AI’s understanding, significantly improving the chatbot’s accuracy and reliability.
- Correcting Product Allocation: We pushed precision even further by using advanced chunking techniques, ensuring that responses were accurate across Hartmann’s entire product range.
The Results: A Game-Changer for Customer Experience
The results of our approach were nothing short of remarkable. Hartmann’s chatbot success rate skyrocketed from 50% to an impressive 98%.
- Phase 1: We redefined the scope of LoyJoy, a German software company, to align with the addData parameters, achieving a hit rate of 80%. We focused on the most relevant PDFs and webpages to achieve this improvement.
- Phase 2: We refined the chatbot further, introducing document-level context for PDFs and section-level context for webpages, which brought the hit rate up to 85%.
- Phase 3: Next, we enhanced the chatbot with section-level context for both PDFs and webpages, pushing the hit rate to 88%.
- Phase 4: During the final phase, we focused on reducing noise and enriching context, leading to a significant increase in the hit rate to 98%.
These improvements meant that customers received highly accurate responses, reinforcing Hartmann’s commitment to excellence and reliability in their customer interactions.
Challenges and Takeaways: Navigating the Complexities
Working within the healthcare industry’s strict regulations, we had to ensure Hartmann’s chatbot complied with the standards set by the Department of Regulatory Affairs. Throughout the process, it became clear that while AI is powerful, success depends on meticulous attention to detail and deep knowledge. We guided Hartmann through the complexities of effective data management, essential to preventing misunderstandings and hallucinations in AI—particularly with LLMs.
Without proper guidance, LLMs often make sense of only 50% to 80% of the data, leaving the remaining 20% to 50% prone to errors. This highlighted the critical need for precise data management to ensure reliable AI performance.
Conclusion: Setting a New Standard
In the end, our work with Hartmann Group set a new standard for what’s possible with AI in customer service. As Tamas Balog, Global Digital Marketing and Media Hartmann Group, aptly put it: “Unlocking the true potential of AI isn’t just about hopping on the trend—it’s about expertise and precision. At Hartmann, we shared the excitement about the ‘GenAI’ buzz. To transform that hype into concrete results, we needed more than just enthusiasm. Enter addData: the true experts behind our AI success story. With their guidance, we transformed our chatbot into a fully operational agent, seamlessly integrating Hartmann data and LLM comprehension, resulting in a B2C GenAI chatbot that answers absolutely accurate responses in a highly regulated environment.”
Through this collaboration, Hartmann has not only enhanced its technological capabilities but also set a benchmark for future AI-driven projects in the healthcare industry.