Where conversational AI shines is in adding proprietary language understanding algorithms on top of commonly used ones.
Being able to interpret and understand what a customer is saying is the foundation of any good virtual agent or chatbot. Without the ability to understand, any virtual agent is liable to frustrate your customers and increase the likelihood of them going elsewhere. That may sound obvious to some, but it is not a trivial matter, nor is building an agent with the required levels of understanding.
Customer service is complex, particularly when it comes to chatting online. For example, many customers will use informal language, including slang, or colloquialisms to communicate with your customer service teams or virtual agents, which can make it incredibly difficult to understand what they are saying.
See also: Imparting Chatbots with Multiple Conversational Styles
Deep learning and natural language processing (NLP) are commonly used methods for building the understanding capabilities of many virtual agents. Those technologies cover the basics, such as training data and predicting the right intent, and can also clean up customer requests to make them easier for any AI to understand. Including verbal quirks, like stemming and removing misspellings. However, most solutions tend to stop there and, while being competent, can be limited.
Where conversational AI shines is in adding proprietary language understanding algorithms on top of commonly used ones. These additional capabilities can give a virtual agent the boost it needs to go from being ‘ok’ at answering customer queries, to reducing false positives by up to 90%. Natural language understanding (NLU) is essentially the brain of a conversational AI-powered virtual agent. NLU consists of three key components working together to decode the nuances of human language:
- Deep learning
- Natural language processing (NLP)
- Additional proprietary algorithms
Combined with other technologies like self-learning AI, NLU can be used to do a lot of the manual work that is required to build and maintain a virtual agent. Conversation data can be analyzed, and new topics suggested on-the-fly. Building a virtual agent from scratch also becomes easier. Leveraging existing chat logs, an organization can reduce the time it takes to have a working prototype agent that can answer thousands of questions in a matter of days, not months. The bottom line is that with better language understanding, you get higher automation rates and, importantly, a better overall customer experience.
How NLU can improve business operations
The main advantage of employing a powerful language understanding technology in a virtual agent is that it doesn’t confine an agent to only being able to assist your customers in a specific situation or domain. NLU allows for a better understanding of all human-machine interactions and can be applied in a number of different ways, across channels, product lines, and topics. Here are some examples:
Sharing knowledge: Gaining access to company-wide resources in an enterprise organization can be difficult. Often, the information that your teams need can be scattered across various silos or legacy systems, with no central repository in place to access it. A virtual agent can be used as a ‘front-end’ to tie these disparate systems together through a familiar chat interface.
Whether its questions about HR or IT, the NLU powering a virtual agent can easily handle employee requests and point them to the answers they need. Similarly, an internal support virtual agent can be used to streamline employee onboarding, or even greet employees when they log in at their desks each morning, distributing daily info updates on new products or changes in company policies.
Offering support: The most common virtual agent use case is in customer-facing support and service. NLU makes this kind of virtual agent extremely powerful. Enabling it to answer thousands of questions with incredible accuracy. It also means that thanks to NLU, your team can be confident that your customers are getting the answers they need. Opening up the possibility of expanding your CX function or virtual agent into far more interesting territories.
With a solid language understanding foundation, a virtual agent can be used in a more transactional capacity, combining with APIs and third-party integrations to not just provide useful information; but actually perform personalized, core business functions directly in chat, on behalf of logged-in customers. This can include tasks like updating a mobile data package, to offering certified financial advice. When a virtual agent truly understands what your customers are looking for, the sky is the limit for what it can do for them.
Sales: Using a virtual agent to drive sales is a use case that we are beginning to see more often as a direct result of the advances in NLU. This capability can either be standalone or, more commonly, can be built on top of a support virtual agent, in order to offer a full-service customer experience.
NLU is crucial to driving sales with a virtual agent because these kinds of interactions can be complex. With multiple questions, transactions, and procedures taking place over the course of one conversation. NLU makes it possible for a virtual agent to understand precisely what a customer is looking for, and this information can be combined with features, such as context actions or conversation goals, that can help both parties reach an actionable outcome.
By understanding which words are important in a given context, NLU or ASU can also identify any mistakes that have been made by deep learning models (if any) and can correct them (as long as the training data quality is sufficient)—offering an extra layer of understanding which reduces false positives to a minimum.
In short, a combination of machine learning, deep learning, ASU, and NLP forms the most robust NLU imaginable and can easily decode and handle complex nuances to human language illustrated in the use cases above.
