Skip to main content
M
Maciek Lekowski
New Participant
May 15, 2026
Question

Analytics: topics classification

  • May 15, 2026
  • 2 replies
  • 125 views

Despite clear user utterances (e.g. payslip, salary, tax form, leave balance), the  Moveworks AI often fails to classify topics correctly, pushing high‑volume demand into “uncategorized” buckets. Improving intent‑to‑topic mapping here would increase accuracy and , insights quality. How can it be done better? Or is there anything on our side that we should consider improving? We also noticed that Moveworks Analytics struggles to categorize a  Domain when citing mainly (only?) pdf files when generating response and it states Domain as “N/A”. Is it known limitation?

    2 replies

    H
    hershd
    Participating Frequently
    May 15, 2026

    Hey Maciek

    In Moveworks setup, you can add positive/negative utterances to improve domain mapping:

    • MW Setup > Core Platform > Domains
    • Edit a domain and add/remove positive & negative utterances as needed.
    • Wait a few mins
    • Retry a prompt in Moveworks AI Assistant that wasn’t mapped correctly before and check 

    Related MW doc.
     

    For your second observation about PDF misclassification, I think that might warrant a deeper look via MW support. Typically domain classification is done based on the user’s prompt/query rather than files/KB that MW surfaces in the response.

    Hope this helps!

    89 Solutions - The World's Trusted Moveworks/EmployeeWorks Partner
    B
    Bartosz Fijol
    New Participant
    May 15, 2026

    Hi,

    Replying on behalf of Maciek.

    We are going to review the Domains and update the list of positive examples to assess whether this improves classification.

    However, one thing I’ve noticed is that topic classification appears inconsistent. Some employee utterances or keywords are not being classified into topics, even though they seem quite straightforward (at least from my perspective), such as tax, leave balance, or payslip. Is there a way for us to influence or improve this?

    Regarding the PDF issue — I wouldn’t necessarily call it misclassification, but rather a lack of classification altogether. When reviewing the Analytics platform under the N/A Domain, I can see that the content cited for this domain consists only of PDF and DOCX files. Please see the screenshot below and note that is a view filtered based on Resource Domain rather than Conversation Domain.

     

    Terms & ConditionsAccessibility statement