Integrate user insights

In addition to normal research activities, develop an understanding of how active users are interacting with your models.

Gather implicit feedback from your users by tracking their activities

In addition to regular discovery research activities, there’s a lot you can learn about users’ behavior by understanding how they use your product. This can best be done by tracking events, or if possible, watching user sessions via tools like Full Story or Highlight.

How often do users ask for recommendations?
What do they do after a recommendation?
Do users drop off after initiating an AI interaction?
Are there trends in the AI interactions they initiate?

Bring this back to your ML teammates

Make sure to share your insights with your ML teammates. They’ll often have useful insights on why a particular user interaction may be occurring, as well as their own list of difficult or risky outcomes that should be tackled. User insights can help prioritise the most important development areas to work on.

Keep in mind

Model developments aren’t the only ways you can improve undesired outcomes. Optimistic loading, analyzing user patterns to constrain and suggest inputs, and other automations can all help to guide users to a better result.

Next guideline

Explain the AI's capabilities