Announcing Early Access of Voice Agent with Watson Analysis Dashboard
5 min read
By: Thomas Banks
IBM Voice Gateway 18.104.22.168 release
Understanding how end users are interacting with AI systems is important for measuring success, identifying areas in which you need to make improvements, and getting an understanding of what the end users are doing. Watson Assistant provides an excellent view into its usage, but when used over the phone with the Voice Agent with Watson service or on-prem with the IBM Voice Gateway, there are additional factors you need to analyse to truly understand the service’s performance.
In the 22.214.171.124 release of the Voice Gateway that is arriving late December 2018, we are releasing an Alpha version of our new UI dashboard which provides insights into how end users are interacting with your Watson powered Voice Agents. The dashboard is designed to query the same Cloudant databases to which the Voice Gateway and Voice Agent with Watson services can be configured to publish reporting events to today. If you are using the existing call data record (CDR) reporting capabilities, you will be able to analyse and understand that existing data. Through this new dashboard, you can quickly gain insights about your BOTs across a large number of calls, enabling you to identify problem areas that need attention more quickly and measure the success of the solution. It can also be used to identify whether changes to the Voice BOT are actually resulting in improvements or not.
With this dashboard, you can select a date range over which to view various details related to your Voice Agents, including the following:
Number of call minutes processed
Number of calls processed
Average call length
Opt-out (transfer) percentages
Average number of barge-ins (where an end user interrupts Watson)
Average latencies related to speech and Watson Assistant
In addition to these high-level statistics, the dashboard can assist you with Voice Agent development by enabling you to filter down to specific groups of calls, on which detailed call transcripts can be viewed. This includes metadata such as STT confidence scores for each transcription event and what Watson Assistant responded with. This enables you to understand how users are interacting with Watson, alongside how and why it is responding to them.
For example, with the new dashboard, you can filter down a list of calls that ended in a transfer to a live agent. You can then inspect each of these calls to get a better understanding of why the caller decided to opt out. Was it a poor response from Watson? Was it an intentional transfer as it was on a specific request type? With the dashboard, you can understand the trigger for the transfer on a per-call basis. Once you know the cause for a transfer, you can work on preventing that in future calls. The dashboard also supports searching on Watson Assistant intents as well as custom values which you can insert into the CDR directly from Watson Assistant during a call.
To get started, please see the instructions on how to pull the Docker container and set it up.
In addition to providing you with access to an early version of the dashboard, we are asking for your feedback. You can join our Slack team at ibm-cloud-tech.slack.com and join the #ibmvoicegateway channel for general discussions. For feedback specific to alpha functionality, we also have the #ibmvg-feat-feedback channel. This is an early view into what we will be providing you with the dashboard—your feedback will help us improve its capabilities over time.