Virtual Agent not returning correct results when referencing KB articles

JayAdmin_16
Mega Sage

Hi All, I'm still quite new to ServiceNow and have been lingering through documentation/community posts. Hoping someone a bit more seasoned can help me here. 

I’ve been having a bit of trouble navigating NLU & AI Search work in conjunction with one another to be able to give me the desired solution to searching for KB articles within our Virtual Agent. Currently, if I type in the exact short description of a knowledge article in the employee portal search bar, then I get the desired result of the search returning my specified knowledge article.

However, within our virtual agent just after the greeting, if I type in the short description of a knowledge base, it won’t return the desired results and I’m finding it hard to understand “why”. I’ve seen similar questions that have been posted in the community that suggest using meta tags, updating the utterances of the NLU model and making sure AI Search is enabled. However, I still haven’t been able to find an answer through the community or on my own.

Which leads into another question. The AI Search on VA is all listed out as Read-only. What’s another process you potentially use to get the most out of your Virtual Agent? I’m looking for advice on whether I should be creating another model for the AI to pick up catalog or knowledge base metatags? Should we create a Search Knowledge Base topic? How are others navigating the Virtual Agent being able to identify an exact Catalog item, or knowledge base?


Let me know if further explanation is required. Any advice and gentle guidance would be greatly appreciated, Thanks!

3 REPLIES 3

Lynda1
Kilo Sage

Going to do my best to explain this one. I am guessing you requested AI Search already.

When a user enters a phrase, the NLU looks for a Topic, if there are no topics that answer that phrase, the Fallback kicks in. Need to have AI Search Fallback turned on, I duplicated AI Search Fallback (OOB) and added our company name to the topic name.

Then went to Conversation / Settings / Virtual Agent / Custom Greetings and setup. Click on the setup you are using, we are using the Default.  Click on Setup Topics tab for AI Search Fallback, make sure your Topic is selected.

 

Now for the confusing part!  In the AI Search Fallback topic you are going to use, the second component in the flow called "Run AI Search" there is a Search Application.  I left it as the Default of Virtual Agent Default Search Application.

Since it seems you want your Virtual Agent to use the same search as your portal. You need to figure out which Search Profile your portal is using.  It should be either Service Portal Default Search Profile OR ESC Portal Default Search Profile

Once you find the Search Profile your portal is using, you can go to AI Search / Search Applications and update the Search Profile in the Virtual Agent Default Search Application

 

Lynda1_0-1726605244681.png

I hope this helps!!

Hey Lynda – Thank you very much for your detailed reply!

I’ve implemented your suggested steps – and turns out this was previously implemented before I started with the company. With your experience with VA, can you type in the exact/similar name of a knowledge article and get an exact/similar result back with the configurations you’ve made?

The VA that has been configured prior to my input can search the employee portal, it’s just not doing it in the most effective way. Like for example, I created a knowledge article called “TEST KA” which is sitting in our HR knowledge base. So by typing in “TEST KA” exactly, I’m redirected to the HR Topic. I can see the logic as to why VA would do that, but is it possible to have the VA search the portal first for an exact match or similar match before going to another topic? Can the fallback topic be configured to achieve something like this?
 

Blob_1762_0-1726618012665.png

 

 Looking forward to your response, and thank you for your input once again!

I can type the article title and do get the article.

What you are asking would be considered a system change. How Virtual Agent works is it searches the NLU first, if nothing found, the AI Fallback does a search.

Since Test KA is finding a topic, I am guessing there is something related to that phrase in the Intent of the topic. Might want to ensure you have a KB that has no words in the NLU (If that makes sense).

The other thing we discovered is, AI Search does use the Title and Article Body as the main two points. It does use the meta tags, but the score for meta tags is extremely low.