Edit

Use sentiment prediction-based model in work classification (preview)

  • Article

Omnichannel for Customer Service offers a suite of capabilities that extend the power of Dynamics 365 Customer Service Enterprise to enable organizations to instantly connect and engage with their customers across digital messaging channels. An additional license is required to access Omnichannel for Customer Service. For more information, see the Dynamics 365 Customer Service pricing overview and Dynamics 365 Customer Service pricing plan pages.

Important

To use this feature, it must be turned on and the license terms must be accepted. What are Preview features and how do I enable them?

A preview feature is a feature that is not complete, but is made available before it’s officially in a release so customers can get early access and provide feedback. Preview features aren’t meant for production use and may have limited or restricted functionality.

We expect changes to this feature, so you shouldn’t use it in production. Use it only in test and development environments.

Microsoft doesn't provide support for this preview feature. Microsoft Dynamics 365 Technical Support won’t be able to help you with issues or questions. Preview features aren't meant for production use and are subject to a separate supplemental terms of use.

Introduction

Sentiment prediction for routing enables organizations to use the end customer sentiment scores as a factor in classifying work items to route to agents.

By using sentiment prediction for routing, you can write rules to classify work items and assign them to agents who are best equipped to handle them.

The sentiment prediction uses the following seven-point scale that allows organizations flexibility to build nuanced rules while maintaining simplicity.

  • Very positive
  • Positive
  • Slightly positive
  • Neutral
  • Slightly negative
  • Negative
  • Very negative

The sentiment model is configured to look for and use the messages provided by the customer and doesn't consider agent chat strings.

Enable sentiment prediction

To enable sentiment prediction

  1. In Dynamics 365, go to one of the apps, and perform the following steps.

    • In the site map, select Routing. The Routing page appears.

  2. Select Manage for Sentiment prediction for routing.

  3. On the Sentiment Prediction for routing (preview) page, set the Enable sentiment prediction for routing toggle to Yes.

Dry run of sentiment prediction

You can optionally test the sentiment model by using the dry run option. You can test using real phrases that you expect to see when customers contact support. It helps you understand what the sentiment prediction on those might look like. You can see key information, such as the sentiment category on the seven-point scale, the sentiment score that is scaled from zero to 100, and keywords from the phrase that influenced the sentiment score.

To test the sentiment model

  1. On the Sentiment Prediction for routing (preview) page, select Dry run.

  2. On the Dry run panel, enter a word or phrase, such as "I am frustrated", and select Test. The model predicts the sentiment score, category, and the keyword that helped determine the score.

Create classification rules based on sentiment prediction model

After you enable sentiment prediction, you can create a classification rule based on the machine learning model, and select the type as Sentiment prediction. You can use the rule with other rules to help categorize the work items to be routed to the right agents who will help with the customer issue.

To create a rule based on sentiment prediction

  1. Create or edit a workstream. More information: Create workstreams

  2. Go to the Work classification (optional) section to create a classification rule. More information: Configure work classification rulesets

  3. In the Create work classification ruleset dialog, select the rule type as Machine learning model, and then select type as Sentiment prediction.

    Sentiment prediction as machine learning model.

  4. Select the input attribute in which the customer sentiment is most likely to be found. For example, for a case workstream, you can select the related entity Routed record (Case), and the attribute as Description that might typically contain keywords indicating the sentiment.

    Input attributes for sentiment model.

    For a messaging channel, you'll typically set the related entity as Context item value (Conversation), and then select the attribute. The bot context variables should be set or pre-conversation survey must be set up to be able to use the input attribute. Frame the pre-conversation survey questions such that they elicit the customer sentiment. For example, you could use a question such as "How has been your experience with the product?" or "How can we help you?".

    You can define up to 10 attributes, whose values can be used to build the prediction model. At least one attribute, the first one, is mandatory and can't be deleted.

  5. Save and close.

While the application doesn't prevent you from creating more than one rule that's based on the sentiment prediction model, a warning message is displayed when you try to create another rule based on the same model.

Additional work classification rules can then be created using the output from the sentiment prediction rule. The output is stored in the Sentiment Category attribute. You can use any of the values on the seven-point sentiment output to build the rule.

Create route-to-queue rules

Create route-to-queue rules to route the work items based on the sentiment prediction.

To create route-to-queue rules

  1. For the workstream in which you created the rule based on sentiment prediction, in Routing rules, select Create ruleset or See more for Route to queues to create a rule. More information: Configure route-to-queues rulesets and rules

  2. Create a rule to define conditions. The Sentiment Category output attribute contains the sentiment prediction for the work item, and should be used in the rule builder to define the rule.

    Note

    If you've defined a condition on a routed record, make sure that you use the Add row option that's independent of the Routed record group. Choose add row option for sentiment category.

  3. Select the queue to which the work items need to be assigned when the conditions are met.

    Rules to categorize on sentiment and route to queue.

Use diagnostics to analyze the efficacy of sentiment prediction

Routing diagnostics help you see how the work items have been classified and routed after you've configured the rules.

You can view how the sentiment prediction model was used to route a work item.

Diagnostics for sentiment prediction model.

Language support for sentiment prediction models

All Azure Cognitive Services supported languages are supported in sentiment prediction-based routing. More information: Language support - Translator - Azure Cognitive Services

See also

Overview of unified routing
How to set up unified routing