Optimize containment rates best practice

If your containment rate decreases, you must take a number of steps to understand why the changes occurred.

Prerequisites

You have a Bot Analytics license.

Page location

Bot Analytics

Containment rates

If your containment rate decreases, you can take the following actions:

  • Check the releases that were completed in the last week to see if any of the changes might have contributed to the drop in containment. Check to see if you provided more agent escalation opportunities or if there were any transactional use cases that might incur failures which automatically hand users off to an agent.

  • Search for relevant news about your company. For example, check if there was an external impact that might cause the escalations.

  • View the performance comparison page using the Daily filter. You can monitor the containment numbers day-by-day to see if you can narrow down the decrease to a specific day.

  • Check the response performance. You can view all of the active bot responses you have to see how they are performing. For example, you can view the number of escalations on a response, negative feedback, or repetition rates for each bot response.

  • Determine if the escalation is due to a business rule (as a business you decided that customer questions about a certain topic should be immediately escalated to an agent) or a user escalation (the user specifically requested to escalate).

  • Review the previous trends of this conversation topic. The containment for topics can fluctuate based on the season. Retail companies experience this significantly during high volume shopping times such as Black Friday or Christmas. It is possible that the intent volumes may have increased during these periods, as the company struggles to keep up with orders and maintain carrier delivery timeframes, causing more frustrated customers who want to escalate.

  • Create recommendations that focus on the conversation topics with the highest volume.

  • If the natural language understanding (NLU) percentage decreases due to too many unknown user messages, you must update the training using the unclassified message section.

  • Go to the Transcript Search tab and apply the conversation topic filter based on the topic that you know this experience primarily falls under.

  • You can apply further filters, such as conversations marked as handed off and conversations that had negative sentiments, as you can see in the image above. These filters give you a clear set of transcripts to review, which might help you quickly identify a trend in what is causing the escalations.