IBM watsonx Assistant’s new routing features enhance customer interactions
Efficient routing capabilities are vital for creating personalized and effective customer interactions in conversational AI systems
*Watson Assistant has been rebranded to watsonx Assistant. Any reference to it in this article refers to its new name.
In this post, we will dive into three key routing features that have been added to Watson Assistant:
- Improvements to topic switching
- Action conditions
- Custom validation
Armed with these features conversation designers can now more effectively route the end user to the appropriate action/task, while also needing fewer steps to handle error cases.
Improvements to topic switching — enhancing flexibility and customization
Watson Assistant has improved topic switching (also known as digression), providing users with increased flexibility and customization options. Let’s explore the three enhancements:
Confirmation for returning to switched topics
Previously, when customers changed topics, they would automatically return to the original topic without confirmation. Now the assistant confirms whether customers intend to return to the topic they switched from, preventing unintentional switches. Conversation designers can configure this feature inside their assistant’s Global settings page:
The conversation designers can write a custom message that will display when your customers return from a topic switch action to the original action. For use cases such as assistants integrated with phone systems, the designer can disable the “yes”/”no” response prompt. Once applied these settings allow the customers to better decide how the assistant should route their requests:
Preventing return to topics after switching
Watson Assistant can now prevent customers from returning to an action after switching topics. This is particularly useful when certain tasks are ‘final’ in nature. For instance, in situations where a customer cancels their account, or expresses intent to end the conversation, it does not make sense to return to the original task, or action.
In these scenarios the conversational designer can select the Never return to original action after completing this action option, and instruct the system to always “end the topic” once the topic switch has completed:
In other situations such as those where the user has asked for help, it is of course desirable to return to the original topic after the user has received the requested information.
Switching topics on Free text and RegEx steps
Historically, actions runtime has favored “slot filling over topic switching”, that is to say, at a given step in actions, when the user enters an input, the runtime will check that input to see if it can satisfy what is being prompted at the current step, if there is a match then the step is “filled” and focus moves to the next step.
However, for response types like RegEx, and in particular “free text” the runtime does not discriminate whether the user provided the information requested, or the user is attempting to switch topics. Prior to these recent updates, free text and regex response types would always cause the user entered input to be stored verbatim by the runtime. The following use case neatly illustrates the crux of the problem.
As can be seen, at the second step in the transfer money action, the system is configured to collected literally what the customer entered (verbatim). While this approach works great when the customer does as asked, and responds with the expected type of response, in situations where the customer asks a followup question, the system treats the follow-up question as the answer to the original question.
Watson Assistant has added a new capability whereby the conversation designer can tell the system whether or not to allow topic switches at a given free text/regex step. The conversation designer simply needs to configure the Allow customer to change topics during a free text response, after which point the system will favor switching topic instead of “filling the step”.
The following screen shot shows the behavior of the bot after the “Allow customer..” toggle:
To learn more about the improvements to topic switching in Watson Assistant, this docs topic has you covered.
Action conditions — customizing routing criteria
Action conditions in Watson Assistant offer authors the ability to customize the criteria for routing customers to specific actions at the topic level. This feature is especially valuable for managing customer access based on their current context. For instance, if a customer is not authenticated, certain experiences or functionalities may be restricted to them.
By defining action conditions, authors can tailor the routing logic to match specific customer scenarios. This flexibility enables the creation of dynamic and personalized conversational flows, ensuring that customers are directed to the most appropriate Actions based on their unique attributes and permissions.
The conversation designer can select the “Conditions” button in the action trigger page to enable the conditions to set on the current action.
Once the condition is set, the action can only be entered once the action has been triggered, and the conditions in the action condition have been met:
In the above action, only employees who are categorized as “current” can access their salary details:
Having the same conversation with the bot, where the logged in user has “employment_status” as “current”, will yield different results:
To explore the capabilities of action conditions in Watson Assistant, refer to the topic in our documentation.
Custom validation — streamlining user response evaluation
Watson Assistant recently introduced custom step validation preferences. Users can now add range (constraint) settings for steps whose type is date, time, number, currency and percent. Range validation is particularly useful in situations where the conversation designer wishes to collect information, but wishes to ensure that the data collected is within a desired range.
For instance, when scheduling an appointment, it will be necessary to schedule the appointment in the future, but no longer than 6 months in the future. Or when collecting an amount to transfer (in a banking app), limiting the amount to being a positive currency is likely necessary.
Custom validation allows the conversation designer to handle business rules neatly inside the definition of the step, previously the conversation designer needed to build separate steps to handle error cases where the user enters a value outside of the desired range.
With the recent update, the conversation designer simply edits the validation info on a given step where they can add minimum/maximum values, along with a validation message that let’s the user know the acceptable range of values:
Later, when the user enters an invalid value, they will be routed to enter a valid value:
As can be seen from the above example, the conversation designer’s workload is significantly reduced as now, what took 2 or 3 steps to complete, can be handled within a single step.
To learn more about custom validation and how it enhances user response evaluation, consult this docs topic.
Routing features are instrumental in creating personalized and efficient customer interactions within conversational AI systems.
By leveraging these features, authors can optimize customer experiences, enhance conversational flows, and deliver more efficient and personalized interactions. To learn more about these routing features and access additional resources, visit the Watson Assistant documentation.
Be sure to check out Watson Assistant’s new collaboration and productivity capabilities, as well as Watson Assistant learning.
Note: These features are available on Watson Assistant on the IBM public cloud, and also in the upcoming (July 2023) CP4D 4.7 release.
