Refining global search using Amazon Nova with Siemens

Siemens uses its global website as a central hub for all its business and product lines. Customers from all over the world use the site’s search engine for support with various Siemens products. The search draws results from 15–20 different Siemens sites, including several business lines and ecommerce portals.

Given the company’s massive scale, its product and service data is complex and extensive. Finding the right information proved challenging, with customers frequently struggling to navigate pages and leaving the site when they could not find the technical documentation that they needed. “Users told us that they often have to sift through pages of marketing content just to find technical specifications, which slows them down,” says Melanie Mrochen, Senior Product Manager at Siemens. “We knew that there must be something better.”

After using semantic search functions to better understand its customers’ needs, Siemens wanted to streamline and enrich the search experience with the latest generative AI offerings. For this, the company turned to AWS, its established cloud services provider.

Because Siemens had worked closely alongside AWS for years, it knew that AWS could supply the cost-effective compute capacity that the company needed to solve its search engine scalability issues. “The AWS team always provides us with access to the latest technology as soon as it’s available,” says Fischer, IT architect and service integrator at Siemens. “Our little game is to see who is first to contact the other—Siemens or AWS—about trying a new service when it’s released.”

After experimenting with various AI tools, the company determined that Amazon Bedrock—a comprehensive, secure, and flexible service for building generative AI applications and agents—provided the strongest foundation for maximum value. Though it first used another AI model for the search project, Siemens chose to reduce costs by switching to Amazon Nova 2 Lite, a fast, cost-effective reasoning model. It also provided multilanguage support and integrates naturally with the existing AWS stack of Siemens.

Siemens customers can now query the search bar at the top of the company’s homepage using natural language rather than having to use restrictive keywords. Each query initiates a function in AWS Lambda, which Siemens uses to run code without provisioning or managing servers. An AWS Lambda function for AI search handling orchestrates the entire search process, coordinating the flow of data between different agents and services.

First, the validation agent processes the query to determine whether it’s valid and appropriate and whether it can be answered directly from a specific knowledge base. Next, the classification agent determines the user’s intent and classifies the query into one of several categories to direct the search to the most relevant resources. The query is then processed by the appropriate search backend, which performs a targeted search in a corresponding knowledge base. Finally, when results are obtained, a summarizer agent processes and condenses the information.

The search also sends queries through guardrail agents to filter out disallow-listed topics, questions about non-Siemens products, and prompts that might result in legal issues, such as questions about company stock. 

Less than 1 year after the initial concept, the new AI-powered global search engine of Siemens went live. “Without using AWS Lambda, which we used to build out the agents, and Amazon Bedrock, where we chose the best-fitting model for each agent, this project wouldn’t have been possible in that time frame,” says Christoph Lumme, enterprise IT architect for digital experience at Siemens.

Customers can now find the information that they need with speed and ease. The solution has eliminated no-results searches and improved user satisfaction by 11 percent. Even if a user asks for information that isn’t housed on the company’s websites, the search will offer guidance for the best resource. “It might say, ‘I recognize that you’re looking for a fridge. We don’t produce them in-house anymore, but please visit the company that’s now in charge of those products,’” says Lumme.

The impact on business of Siemens has been considerable. Using Amazon Nova Foundation Models, the company has improved its search speed by 300 percent and reduced costs by 70 percent compared to the models that Siemens was using before. Meanwhile, the project’s guardrail implementation also led to the creation of a company-wide legal framework for generative AI risk governance.

Siemens plans to improve its global search engine even further, transforming it into a more interactive experience, and looks forward to developing more solutions using AWS. “Using generative AI on AWS has helped our development teams to be more agile and productive,” says Fischer. “By being that flexible and that fast, we’re able to deliver more things to our stakeholders.”