Created and originally published by Solita / solita.fi

To streamline the after-sales team’s troubleshooting work, leading Finnish waste management machinery manufacturer Tana partnered with Solita to develop and integrate a customized Generative AI assistant tool. The AI assistant empowers after-sales team members by swiftly extracting and presenting useful information from user manuals and incident reports, enhancing, speeding up and improving the quality of their technical support service.

Results

1. A proof-of-concept Generative AI (GenAI) assistant to streamline and improve Tana’s troubleshooting process
2. The AI assistant helps the after-sales team figure out what’s wrong and how to fix it, making the process more efficient, effective, and consistent.
3. AI proved to be useful in solving complex problems and shortening resolution times.
4. A fully developed system will greatly help and speed up onboarding new people on the after-sales team.
5. A fully developed system will enable Tana to provide consistently good support service and better customer satisfaction.

Large amount of complex documentation makes the troubleshooting process unnecessarily inefficient – causing potential delays, downtimes, and frustration

Tana is a market-leading manufacturer of waste treatment machinery, selling and servicing their products in over 50 countries across five continents. “We oversee a global value chain and fleet of 1,000 units using a uniquely asset-light business model,” says Teemu Lintula, Vice President of Services & Digital Solutions at Tana.

Although Tana’s waste compactors and shredders are impressive technological marvels, they will occasionally experience hiccups that require servicing. And since downtime is costly and the machines are large investments, it’s important to get them up and running as soon as possible.

Tana prioritises customer satisfaction through exceptional customer service. Unfortunately, many factors contribute to making the after-sales team’s job unnecessarily complicated.

The usual troubleshooting process is to look for solutions in extensive user manuals and old incident reports.

The large volume and complexity of the documentation make it difficult to find the right solution.

As a result, this potentially slow, complicated, and inefficient procedure can cause delays, downtime, and frustration.

The varying experiences of team members lead to inconsistent service. Newcomers must often familiarise themselves with the nuts and bolts of the machines. In doing so, they will sometimes come up with the wrong solution by mistake, contributing to delays and downtime. This may alienate customers who might lose their faith in and satisfaction with Tana’s service support.

Generative AI-powered assistant, reducing downtime and increasing customer satisfaction

Inspired by the promise of AI-powered chatbots like ChatGPT, Tana was determined to bring its customer service to a new level. To realise this vision, Tana needed a partner who understood their challenges and the possibilities of Generative AI technologies, plus who could also deliver results.

Tana decided to collaborate with Solita to develop a proof-of-concept project exploring the possibilities of a custom Generative AI-powered assistant designed to help technical specialists resolve complicated customer questions and issues related to Tana’s machines.

To accomplish this, we used Azure OpenAI Service’s Large Language Model (LLM), which is a form of artificial intelligence designed to achieve general-purpose language understanding and generation. Together with Tana, we set out to create an LLM-based assistant, which generates fact-based answers and recommends appropriate solutions.

Like ChatGPT, the after-sales team can use the assistant to ask questions using natural language, and the assistant will then generate answers.

But unlike ChatGPT, Tana’s AI assistant will base its answers on Tana’s documents. The assistant also cites source details for its answers, so Tana’s after-sales team can check the actual source documents for themselves.

We also focused on minimising or preventing the Gen AI assistant from making up or hallucinating things. There are different tactics to do this. For example, we could “fine-tune” the LLM with case-specific documentation. Fine-tuning is, however, a costly and time-consuming process, and therefore we decided to instead “ground” the LLM in a case-specific context by using a technique called Retrieval Augmented Generation (RAG). RAG is by far the more cost-effective approach than fine-tuning.