How Can We Improve the Accuracy of Now Assist in Virtual Agent?

[Kohei Tominaga1]

Hi Community,

We are evaluating Now Assist in Virtual Agent and have a couple of questions:

• When Now Assist generates an incorrect response, does providing the correct answer (by an agent or user) help improve future responses automatically?
• If not, what are the recommended ways to improve answer accuracy? For example, should we configure synonyms, stop words, search relevance, or other tuning settings?

I'd appreciate it if anyone could share best practices or real-world experiences.

Thanks!

[Vamsi_Krishna07 · Accepted solution]

Hello @Kohei Tominaga1 ,

Now Assist in Virtual Agent doesn't automatically learn from an agent or user correcting an incorrect response. Those corrections are helpful for identifying where the knowledge base or search experience needs improvement, but they don't become part of the model's knowledge or influence future responses on their own.

The best way to improve answer accuracy is to focus on the quality of your knowledge content and AI Search configuration. Keeping knowledge articles up to date, filling content gaps, and tuning search relevance with features like synonyms, stop words, and other AI Search settings can make a noticeable difference. It's also a good practice to review user feedback and search analytics regularly to understand where users aren't getting the right answers and refine the content accordingly.

You can find Analytics in the Now Assist Admin Console's performance tab.

Thanks!