Improve your Power Virtual Agent bots, with Power BI & Custom Analytics

If you are not familiar with Power Virtual Agents (PVA), you can have a look at my previous article where I explain how to overcome Power Virtual Agents limitations by using Bot Framework Composer.

I enjoy working with Power Virtual Agents since it is very simple to get started and to build a production ready bot. But recently, while working with an actual customer, we faced the limits of the built-in analytics during the bot's life cycle where you want to continuously improve your bot based on previous interactions.

Currently the built-in analytics are great for signaling which topics have a high escalation or abandon rate and which topics effect customer satisfaction negatively. However, it is not enough to understand which topics don't perform. If you want to improve your bot, you will need insights on the 'why'. The product group is rapidly adding new AI features that will help you in this process, however in the mean-time we will need another solution.

Learn from customer interaction

It would be great to learn more from how customers are interacting with your bot. Power Virtual Agents allows you to download a maximum of 7 days of sessions from the past 28 days in the CSV format. The downside of this export is that the CSV format is hard to work with, thus you would need to write your own data parser and even then it will lack context like buttons, adaptive cards and rating controls...

Power Virtual Agents stores your data in Dataverse, which gives you access to all the raw transcript data and other data that can be used for analytics purposes. Matthew Farmer, Sr. Program Manager for the Power Platform CAT team, created a Custom Analytics solution in Power BI that leverages this data in a report that you can use as the base.

This is a great start to get better insights from your bots, but visualizing transcript data in Power BI is still hard since the 'table' visual is the only visual that is capable of showing this data. And next to that, this visual is still hard to read and you will lose the metadata like buttons and adaptive cards.

In order to overcome this once and for all, I developed a custom visual for Power BI based on the open source Bot Framework WebChat. This visual is fully capable of showing a transcript as a chat, just like how the end user would see it. The first version of this custom visual has been published open source on GitHub and can be used in your reports from now on!

How to get started?

The transcripts functionality has already been added to the Custom Analytics solution for Power Virtual Agents, where you can follow the installation steps. You will need a Power BI Pro license in order to create a dataflow and Power BI report.

Feel free to give it a try and to deliver you feedback in the comments or via the GitHub issues. We are looking forward to learn more about your use-cases, to improve the custom reporting and custom visual.

The future

Did I spark your interest about the new artificial intelligence features in Power Virtual Agents, so that you don't have to read through transcripts yourselves? Have a look at the video below or read more about how to get topic suggestions from bot sessions with AI-assisted authoring (preview).

GitHub: Custom analytics solution for Power Virtual Agents

GitHub: Chat Transcripts for Bot Framework - Custom Visual