AI Search: The secret to better customer self-service

Hari Vats, senior knowledge manager of global technical support at ServiceNow, co-authored this blog.

Who wouldn’t want a faster, more relevant search on their knowledge base? Improving search results is tricky. Organizations can spend a lot of time and money making small improvements to very complex, dynamic environments of information.

When ServiceNow acquired an AI search technology company in 2020, we were thrilled to see an immediate return. Since we adopted AI Search, we’ve seen our search results delivered 3.5 times faster than with our previous search engine—from an average of 10.3 seconds down to 2.3 seconds. Perhaps even more important, relevancy improved, with results now 2.9 times more relevant.

How did we achieve these results? AI plays a big role. But AI is just a technology in service of a business goal. Our goal was simple: Deliver the right content from our knowledge bases to our customers at the time they request it. We knew we had a lot of useful content on our Now Support portal. But if our customers couldn’t find the right content at the right time, all the AI in the world wouldn’t help.

Improving search relevancy and results

AI Search connects employees and customers to the answers they need in a simple search experience with personalized, relevant results. It uses machine learning to identify user intent and deliver results users can act on from within the search results window.

In the past, our legacy search engine used keyword searches to match words or phrases within a query. AI Search delivers a very different user experience. It can interpret contextual cues to identify user intent and produce highly relevant information.

To see how AI search worked, we conducted extensive user acceptance testing in global technical support for more than six weeks. We engaged subject matter experts from 13 different teams who tested more than 50 functionalities in only two weeks. Using these suggestions, our development team improved relevancy and performance prior to the go-live date.

Our favorite AI Search features

Here are a few of the AI Search features we found especially useful:

• Auto suggest: We refine the list of popular keywords by analyzing what users click and search. AI Search doesn’t yet support an index-based auto-suggest, but we look forward to that in a future release.
• Spellcheck: The spellcheck function can find multiple spelling mistakes in each query and use AI to correct them and deliver more relevant results.
• Search as a service: We can now index once and create relevancy models across applications. Once implemented, this solution can help make operations more agile and maintenance more streamlined. Powering multiple channels with a single search solution means users can access quick, relevant information regardless of whether they use a desktop or mobile device.
• Attachment search: Our engine can search attachments as well as articles, a feature that was unavailable in the past.
• Promoted result: One major challenge of any search engine is to surface relevant content around generic queries, which make up about a third of all queries. Subject matter experts decide which answers are the best match for these queries. This has significantly improved our relevancy, as well as how people engage with the results.
• Personalized query rules: We can now personalize search results for different user roles and user groups. Gone is the black box relevancy algorithm that didn’t recognize different needs for different user types.
• Content security: Following a content security enhancement in the Rome release, users can now search external content securely using a single search application across all data sources, providing a more cohesive experience.

Other new features included integrated support for five languages, fast setup and integration, and easily configurable synonyms and stop words.

Impact of AI Search

The power of AI lies in its ability to constantly learn and to improve relevancy through machine learning. For example, if users consistently click on the third entry in the list, the system recognizes it to be the most relevant and changes rankings. AI learns from users’ behavior to predict the best answers. This is called automatic tuning.

The impact of AI Search is proven by our metrics:

• A 3.6 average click rank indicates the user clicked one of the top five results.
• 80% of users click on one of the top five results, compared to 52% previously.
• 92% of the time a search automation request does not lead to a case submission.

Although we’ve seen dramatic improvements in our search functionality, we’re working on further improvements to both time and relevancy. It’s all about delivering the right content to the right person at the right time. AI is just one way we can give customers the tools they need to self-serve answers quickly and move on to their next task.

Learn more in our Now at Work 2021 session, Migrating to AI Search—a customer zero story. Registration is free. Take a deeper dive into how AI is transforming work.