Computers have drastically changed the world over the last couple of decades, and they have changed how we humans communicate and interact with each other. As the powers of Artificial Intelligence and Machine Learning continue to improve, we have helped computers communicate with humans as well. Whereas once upon a time you needed to be a computer programmer to understand the weird languages and error codes of computing, we are now getting close to the point where computers can speak to us in everyday languages like English. When computers start to talk like people, it’s no surprise that businesses will find ways to use chatbots (or virtual assistants) to provide a new interface for customers to interact with the company.
The role of the customer service agent in many cases is one of interpreting questions from customers, finding the right information within the business or pushing the right buttons, and then responding to customers. For example, you might call up your bank to open a new transaction account. The customer service representative at the other end of the call is listening to what you’re saying, sustaining a conversation with you and asking you questions, filling out a form on their computer, and then informing you of the outcome. When the interaction is broken down like this, it becomes easier to see how we might use a computer to achieve many of the same goals - after all, the human operator is often just an interface for the computer because they are just entering information into the computer based on what the customer says. The computer already has the body of knowledge related to the company’s policies, shipping details, customer relationship management, marketing, and so on. So what’s stopping computers from taking over the world?
The tricky part of the problem has always been language. Computers natively speak in binary - sequences of 0s and 1s. The translation of English sentences into 0s and 1s is not so simple - in fact, anybody who has learnt English as a second language can tell you that the whole language is not so simple. It’s full of rules and exceptions to those rules, and then people add on metaphors and idioms, and that’s before we add colloquialisms into the mix. Natural Language Processing (NLP) is the field of computer science that focuses on interpreting language using computers. The two main problems are Natural Language Understanding (NLU) and Natural Language Generation (NLG) - in other words, being able to listen or read, and then being able to speak. There are a bunch of sub-fields related to NLP that you may have heard of: text processing, sentiment analysis, language translation, conversational intelligence, text filtering, information retrieval, and response generation. These all rely on computers being able to understand words and sentences from humans, even if we aren’t using the Queen’s English. Engineers have been working on these problems for decades.
So chatbots are all of this implemented in the real world by software developers. In a general sense, they aim to provide an easy-to-use, natural language interface for humans to communicate with computers. In a more specific sense, chatbots allow customers to interact with a company and its knowledge base via its computers. Now, let’s take a look at some of the use cases and applications of state-of-the-art chatbot technology.
The most common application of the chatbot is customer service portals. Many companies and websites already have questionbanks or FAQs, but the customer still has to spend a long time scrolling through many questions, or cleverly using the right key terms in a search bar to bring up the right questions. With the power of NLP and a chatbot, customers can strike up a conversation instead, and ask questions in a natural way that don’t necessarily have to match the question/answer pair as written in the question bank. For example, the question in the database might be “How do I book a flight?” but the user might type “I need to get a flight ticket” - the NLP engine needs to be able to resolve these as being the same question. The chatbot can try to infer intention in the question, remember previous parts of the conversation to provide more context, and detect when the customer is getting agitated and a human needs to take over. There are advantages for both the company and the customer here - the company can have fewer human support agents, and customers get 24/7 support and don’t have to wait in queues. All this results in a better customer experience and ultimately increases the net promoter score for the company.
Amtrak, the North American train and railroad operator, has a great example of successfully using chatbots to augment their customer service operations. Julie, their virtual assistant, understands queries from users such as requests for train schedules or rules around luggage, and points them in the right direction. The chatbot doesn’t need a sophisticated or fancy design, just a way for messages to get to and from Julie. Within a year of putting Julie on their website in 2012, Amtrak answered over 5 million questions, saved over $1 million in customer service expenses, and generated more revenue by making the sales process simpler for customers and adding on upsells to the interactions. Something that’s important is that the limitations of Julie are well understood, and the system redirects conversations to human support agents when the questions are too difficult to understand.
Chatbots can also be used as a sales funnel to help companies understand what products specific customers need, and provide information to help them make the purchasing decision. For example, we at Spark 64 developed an insurance chatbot for Cove Insurance that integrates into Facebook Messenger, which interacts with the customer to ask them the questions that a salesperson or insurance agent would normally ask. Based on these interactions, the chatbot can produce quotes for different insurance policies, or help submit a claim with the customer directly with photos on their phone. The key point here is that the chatbot isn’t just answering questions - it’s also collecting the necessary information from users in an automated way.
Chatbots are also great in situations that don’t have any sales involved. Some interesting chatbot applications include helping users find recipes for dinner, learning new languages, or planning travel itineraries and finding directions. A particularly cool one is National Geographic’s chatbot that was used to promote their TV show Genius, which allowed users to interact with Albert Einstein or Pablo Picasso. This demonstrated how chatbots can be given some personality in the way that they respond to humans, and they don’t have to be stilted and robotic. At the same time, it helped increase customer engagement between users and the brand, with the average conversation lasting 6-8 minutes!
It’s important to note that using a chatbot isn’t necessarily about getting rid of humans and pursuing automation at all costs - in many cases, chatbots help increase engagement and sales by adding another channel for customers to interact with the company. It can help with customers who are nervous about talking to people, or customers who have embarrassing questions. It can support a more natural form of communication for a new generation of digital natives who prefer to text or write messages rather than call people on the phone. It can also free up staff time from routine queries and allow them to focus on delivering better results for complex enquiries, and it can help protect staff from particularly aggressive customers. Chatbots are also helpful for maintaining a global presence, and allowing customers to interact with your company at any time, even when most of your workforce is asleep. But ultimately, chatbot technology should be targeted towards augmenting human effort, not replacing it.
In the next article, we explain a little bit about how to tell when you’re talking to a genuinely automated chatbot or a human, and what tricks are used to cheat a little bit to ensure that customers get good experiences.
Want to incorporate a chatbot onto your company's platform?
Or just want to find out a little more?
Contact us and we'll get you sorted.
This Lunch and Learn webinar provides a detailed overview of reinforcement learning, its applications and use cases, and a selection of resources you can explore to learn more about reinforcement learning.
What is the edge? What are some of the key considerations to explore when you're planning an edge AI implementation? How can you best leverage the edge to give you the best business outcomes? Read on to learn more about how the edge brings AI to your fingertips.