How Watson Works

By looking at the intent, entity, and context in combination with other cognitive abilities, Watson analyzes the question and finds potential responses and ranks them with a confidence score. If Watson is confident in a response, it chooses the highest ranking potential response to answer the customer’s query.

If the confidence score is low, Watson looks for other options, including transferring the call to a human representative or searching for a possible answer within documentation.

Confidence Score
How Watson Works

Learn More About Watson’s Techniques


  • Intent

    Watson extracts intent through an understanding of the subtleties of language, including idiom, syntax and ambiguities - regardless of how the question is asked or what words are used.


  • Entity

    When trained appropriately, Watson can extract key entities relevant to the domain in question and establish relationships between them to respond more usefully.


  • Context

    Watson extracts context by looking at previous exchanges in the conversation to ensure it responds with a relevant answer.


  • Tone

    When integrated with the optional Tone Analyzer service, Watson can recognize when a specific tone (joyful, confused, angry, frustrated, sad) is used, and tailor the response to be empathetic of the identified tone.


  • Retrieve and Rank

    If Watson is equipped with the optional Retrieve and Rank Add-On and has low confidence in potential responses, it will search for an answer within the corpus of documents it is trained on before handing off to a live agent.


See how much your company could save with Watson Engagement Advisor.