Unified Routing – Sentiment Predictions

Within Microsoft’s Unified Routing for Dynamics 365 we can route work items using AI based sentiment predictions. Say you detect that a customer is really unhappy when starting a conversation in Omnichannel for Customer Service … with this feature you could route the customer to a specific queue possibly staffed by agents who are the most able to handle difficult conversations!

I believe this feature is currently still in preview (in August 2022) but like most preview features I assume it will be generally available in the fullness of time. Before you enable sentiment predictions in a production environment, I do suggest you confirm its status.

Below you can see that I have opened my “customer service admin center”. Within the routing option you will find several options to intelligently classify work items. Including the option of focus for this blog post “Sentiment-based routing”.

Clicking the manage option allows me to enable sentiment predictions for routing. (As shown below.)

The dry run option is interesting! As that lets you add some phrases and test the responses you can expect in your predictions. For example, it told me that saying “I am unhappy about your response times” was a slightly negative comment. But when I changed “unhappy” to be “really angry” the prediction was upgraded to negative. And a phrase of “I am angry, unhappy and depressed by our response times” finally got a rating of “very negative”.

Next, I wanted to see how this worked in my actual routing. I already had web chat setup so decided to use sentiment predictions with that. I wanted to keep my existing routing logic but if the system detected a negative conversation, I wanted that to always be routed to a specific queue of agents.

I first created a queue called “Negative Message Queue” and then set about altering my work stream classification and routing.

In my workstream, in the work classification area I created a new classification ruleset. I called this “sentiment”. And as you can see below, I set the type to “Sentiment Prediction”.

With my chat I have a pre-chat survey. In that survey I have a question “How can we help today”. That ends up in a context variable called “Query”.

So, in my sentiment prediction I selected the input as this question from my pre-chat survey. The output of this classification will be a sentiment rating.

You can see below that I kept my existing work classification. Meaning I had two classifications, one which assigned a skill requirement. And the second used machine learning to return a sentiment prediction.

Next in my queue routing I created a routing rule which would always route to my “Negative Message Queue” if the sentiment category was returned as negative or very negative.

You can see below that my final routing setup has two rules. The first would “spot” and route any negative conversations whilst the second would handle all other chats. (As it always did!)

Next, I had a cup of coffee! Why??? Because all Omnichannel changes take 15 minutes to apply.

After drinking my fresh ground coffee I decided to test my change!

Below you can see that I start off my entering a negative statement. By saying I was “sad unhappy and angry”. Sounded pretty negative to me!!

As expected, my web chat was routing and showing up in my customer service workspace. By looking at the queue associated with the conversation I was confident that this chat had been routed via my “Negative Message Queue”.

I of course repeated my test with a more positive query. This time I entered the phrase “Happy with the World!”.

The point of this second test was to prove that my conversation had been routed via my normal messaging queue. And as you can see below the queue associated with this conversation was “Neils Messaging Queue”. So, I had proved that positive and negative conversations were routing as expected.

I hope you agree that routing conversations based on predictions around the customers sentiment might be very useful. But with great power comes great responsibility. Please use this feature for the good of your customers …. don’t take all the negative conversations and route them to a virtual agent to get them off your back!!! Enjoy.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s