decodeDynamics

Conversational BOT Design: Although this topic is thoroughly explained in Nuance Mix, it remains relevant for any organization aiming to create effective conversational AI experience

When crafting a conversational bot, one of the critical design aspects revolves around managing conversation deviations. In the context of bots, this phenomenon is referred to as “intent switching.” In this article, we delve into the essential considerations for handling such instances within Nuance Mix.

If you want to know more about overall design consideration for Nuance Mix, you can read here.

What is Intent : An intent is something a user might want to do in your application. You might think of intents as actions. For example an Intent could be “Order Pizza”. If you want to know more about Intent and Basics of Nuance Mix you can refer here.

Okay, in this session we are going to learn about Intent switching and one step correction.

In real life scenarios , it is very difficult to know the mind of the user who is interacting with the Bot for example, a user is interacting with the Bot regarding bill payment and bot has set of questions defined for that, and in-between the conversation user asks a different question for example “what is my account Balance”, so this is called intent switching, means Bot is expecting and trying to collect the response for a specific question, but user asks another question rather than replying to the question asked by BOT.

If we don’t design the Bot to handle these situation, it will impact user experience with the application

So let’s assume the situation, in our example we will use RichiPizaa an online Pizza centre, which takes order for the Pizza. While user is requesting for the Pizza and bot is in-between the conversation and collected most of the inputs but yet not completed the request, now user asks a different question back to the Bot, for example he asks what is the discount for members available, since user is asking a genuine question , bot must respond, also once bot responds to the question, it should-not start from the very first question around Pizza , rather it should be able to continue from the user deviation point.

Okay, so how to handle such scenarios. So while designing the NLU model, we have to see, which is our core entity or entity in focus we are capturing or building the bot, and then the outer layer can be other related entities in task flow and then the top layer is the entire NLU model.

To illustrate, consider RichiPizaa, an online pizza centre. Suppose a user is ordering a pizza, and the bot has collected most of the necessary inputs. However, before completing the request, the user asks about member discounts. The bot should seamlessly address this new intent without losing context.

In summary, when designing conversational bots, understanding intent switching is crucial. Intent switching occurs when a user deviates from the expected conversation flow, posing a challenge for the bot. Imagine a scenario where a user interacts with a bot about bill payment, but suddenly asks, “What is my account balance?” The bot must handle this intent switch gracefully.

Properly handling intent switching ensures a smoother user experience with the application.