Advanced Conversational AI Concepts
Keeping up with the volume and scope of customer interactions can be challenging for even the most efficient business. AI assistants are fast becoming the go-to tool for many businesses as they automate customer support sessions to provide a high quality experience at lower cost.
There are three advanced conversational concepts within Watson Assistant that will make your solution successful: Natural Conversations, Integrations and Analytics.
Ensuring your conversational AI system is able to gather why a customer is there in the first place, that is the user’s intent, is key to producing a better customer experience. In Watson Assistant we have built several advanced features to make sure your virtual assistant can have a more natural conversation by better understanding the user’s intent and taking the appropriate action, be it to give a one-off response, collect more information, or digress. Let’s discuss some of these features:
In Watson Assistant, you can create an entity model to detect entities based on context. This is important, because we can’t always cover every type of entity that an end-user might mention, but we can help the system understand in what context any type of entity might appear. For example, let’s say we create an entity called ‘@type_of_cusine’. In this list of entities we have Chinese, French, Indian, Italian, and Thai. This means it will be easy for our virtual assistant to identify these five types of cuisine as the right entity. But what when someone says “I want to find a Scandinavian restaurant”? It is not in our list. This is where contextual entitles comes in handy. It allows us to annotate intent examples to train it on the context a entity might appear. For example, in the examples below, we would highlight the places “@type_of_cusine” appear:
What is the best French restaurant
Restaurant with best reviews
Is there an Italian restaurant nearby?
Cheap Chinese restaurant
Place good for kids
So even though we don’t have “Scandinavian” in our list, based on the context the entity appears, it will be able to understand what the end user is looking for a restaurant that serves Scandinavian food.
Many dialogues involve gathering several specific pieces of information from a user, such as booking an appointment. Business can use another technique, called ‘slots’ in Watson Assistant, to manage these complex interactions too.
Different users may deliver these pieces of information in varied ways. One person may be happy to provide a day and a time separately when prompted with different questions, while another may leap straight in and provide all that information in one question. It is important for an AI assistant to handle the information however it arrives.
The slots feature creates separate spaces for each piece of information that the system needs to collect during a discussion. Watson Assistant has the ability to recover the necessary information and store it in those slots regardless of how the user delivers it. This enables the user to interact with the system in complex ways at their own pace.
What if a customer needs to break the conversational flow by asking another question? He or she may want to know what times are available before suggesting an appointment. This question might not fit the expected narrative and could throw some AI systems off track. However, thanks to Watson Assistant Digressions, the solution has the ability to handle conversations in which the user branches the discussion to fulfill a different goal, and then come back to where the previous flow left off.
It is important that any conversational system can integrate with a variety of front-end channels and back-end systems.
Any conversation starts with a communication channel. That channel was traditionally voice-based, usually via the phone. Today, businesses must interact with customers in any way that suits them, including phone calls, websites, social media conversations, instant messenger chat sessions and SMS text messages.
As the list of channels grows, the complexity and cost for businesses can increase. By integrating AI systems with communication channels ranging from Facebook to Skype and even with IoT devices, businesses can interact on their customers’ terms.
These integrations feed customer interactions into pre-processing functions that can help prepare customer communications for Watson to respond to. This preprocessing includes converting language from speech to text or deriving sentiment from a customer’s words by analyzing the tone of their speech. By the time it reaches your AI assistant, a customer’s question is ready to be processed as part of a natural conversation.
Just as it uses integrations at the front end for omni-channel communication, it is also important the system integrates with back-end enterprise data and services including data stores, file repositories and even third-party applications. These integrations help find the information that Watson Assistant needs to surface relevant information or perform an action, like make an appointment or change information in an insurance policy.
Understanding how your customers and virtual assistants are interacting with each other is the third main advanced technique in conversational AI. As it conducts thousands of conversations each day, Watson collects analytics on what type of questions are being asked and how they are answered.
This analytics information can help a business to improve performance by measuring the system through customer utterances. Using Watson’s analytics dashboard, a business can evaluate valuable metrics such as how long and complex the average user session is, how many times users come back for more conversations, and how well Watson understands what they are asking.
Using analytics in conversational AI also provides more qualitative information, including what the top intents are and what people are talking about. Businesses can use this to gather valuable customer intelligence for everything from product development to marketing. This data can also produce better recommendations during customer interactions.
Years ago, advanced AI concepts like these would have taken hundreds of thousands of lines of expensive bespoke code, if they were possible at all. Today, Watson Assistant has dramatically lowered the barrier to entry by making these features accessible via simple graphical user interfaces that make it far easier to capture business experts’ knowledge and harness it for automated conversations.
View original content: Here
Related IBM News: