Short search terms and deflection data

[drosenst2]

Hi all. We are exploring case deflection data and looking at Now Assist results. One of the issues I ran into is that people are putting in 1 or 2 words in the search bar which makes it really hard to measure deflection. It will often give the searcher all the options of that word and go off in an tangent with that word or two as well.

How can we encourage better searches/prompts from our end users? Also, how can we better equip Now Assist to handle these types of searches when they do occur?

How are you all handling this currently? Thanks. 

[MaryG]

I think there are two things worth exploring here: improving search relevance and understanding how search behavior influences deflection metrics.

One- or two-word searches are fairly common. Users are accustomed to searching with keywords, so I think it's important that we design the search experience to accommodate those behaviors while also encouraging more descriptive searches when appropriate.

A few areas I'd consider:

• Search configuration: Review AI Search profiles, synonyms, result improvement rules, and autocomplete suggestions to help improve relevance for common search terms.

• Knowledge content: Evaluate whether article titles and content reflect the language users naturally use when describing their issues.

• User experience: Consider adding examples or contextual guidance that demonstrate how a more descriptive search can produce better results.

• Now Assist: Test common short queries against both AI Search and Now Assist results to understand how each handles limited context. This may help identify opportunities to improve retrieval, content, or configuration.

For deflection measurement, I'd also look beyond the search terms themselves. A search or article view doesn't necessarily indicate successful deflection. Understanding what happens afterward, such as whether the user refines the search, views an article, finds the information helpful, or proceeds to create a case, provides a more complete picture.

ServiceNow's Self-Service Analytics capabilities may also be worth exploring, particularly the distinction between confirmed deflection, potential deflection, and no deflection.
https://www.servicenow.com/docs/r/servicenow-platform/knowledge-management/ssa-concepts.html

Ultimately, I'd focus on improving the search experience based on how users actually behave, while using the available analytics to better understand whether they're finding the information they need and resolving their issues successfully.