Securing troubleshooting for 40,000 technicians using AWS with KONE

KONE builds and installs elevators and escalators worldwide. It maintains thousands of models, including ones from other manufacturers. No single technician can be familiar with all those models, so KONE has regional call centers—technical help desks (THDs)—that provide additional expertise. But as the numbers of elevators grow, so do call volumes—sometimes leaving technicians with long waits to connect with support.

The company identified an opportunity to use generative AI to enhance its technicians’ efficiency. AI that’s trained on KONE’s documentation could accelerate maintenance by providing technicians with quick responses to technical requests. Because this documentation is highly confidential, however, the solution needed to comply with security standards and best practices. So, KONE turned to AWS, which it was already using for its Internet of Things (IoT) solutions. “We use AWS because it can deliver the security services that we need to quickly build, run, and monitor AI,” says Tero Hottinen, vice president of strategic partnerships at KONE.

In addition, accuracy was paramount. Because safety is a critical concern in the industry, the solution needed to have steps and checks to verify that the assistant would provide relevant and correct information. To create an assistant that meets these requirements, KONE used Amazon Bedrock—a comprehensive, secure, and flexible service for building generative AI applications and agents.

KONE accelerated the solution development process by working closely with the Generative AI Innovation Center, which has AWS science and strategy experts with comprehensive expertise spanning the generative AI journey. The company also collaborated with the AWS prototyping team to bring the solution to life. “Using AWS is how we can move quickly to bring this new innovation to market while delivering the security and reliability that our customers rely on,” says Hottinen.

Technicians access the assistant through a mobile app called Technician Assistant to ask technical questions as a first step before reaching out to the THD. The assistant augments the THD capabilities with access to the same documentation—including user manuals, historical maintenance reports, and IoT data for connected devices.

The first pilot of the assistant had 100 users for 3 months. During that phase, technicians could opt out of the app, but none did. In fact, one of them expressed his fear that the app would be removed after the pilot period. “The biggest impact is happy field service technicians,” says Hottinen. “It’s difficult to attract skilled people in this field, but when they enjoy their job and can do it efficiently, that’s the best way to keep talent.” The assistant is now live in 11 countries with around 1,500 active users and growing.

KONE implemented a comprehensive pipeline for the app, covering the preprocessing, retrieval, evaluation, and text-generation stages. The solution architecture prioritizes data protection and privacy, and least-privilege access and data-minimization techniques limit access to data. To protect sensitive information through data isolation and encryption, KONE uses AWS Key Management Service (AWS KMS), a service for creating and controlling keys that are used to encrypt or digitally sign data. Additionally, data anonymization removes personal information from chats and messages after a certain period. “As one of the world’s leading cloud service providers, AWS offers robust solutions with enhanced security features,” says Hottinen. “We rely extensively on the expertise of AWS to support our key initiatives in this area. The security fundamentals are already in place.”

Using the assistant, technicians troubleshoot issues and get elevators back up and running more quickly. Now, instead of waiting on the phone for the THD, technicians get conversational responses with highly accurate, documentation-based guidance from the assistant on how to fix issues. “It’s saving a lot of time onsite,” says Hottinen. “An elevator that isn’t functioning is a real pain point for the customer, so this decrease in downtime is a significant improvement for our customer service.”

With the solution in place for over 1 year, KONE has been improving it, and the hard work has paid off. Using the assistant, technicians can interact more effectively and access accurate information more quickly to find solutions. This has led to an increase in assistance requests compared to what the THD would experience. This means that not only is the solution removing the need to escalate some questions to the THD, but it’s also answering ones that technicians didn’t ask before. “Now, technicians have a solution to solve our small daily issues,” says Hottinen.

The company is rolling out the assistant progressively across several countries. KONE expects to reach around 6,000 users in the coming months and make the solution available to all its 40,000 technicians going forward. As it continues the rollout, KONE is testing recent large language models and looking to incorporate multimodal input and output into the solution so that it can handle not only text but also images.

“Using generative AI on AWS, we’re providing our technicians with tools that they are enthusiastic about—modern tools that are solving actual issues,” says Hottinen.