Chatbots - the future of customer experience

Sanja Umicevic, Desa Marinkovic & Nikola Vukobrat, Serbia

Endava Belgrade

How are chatbots improving the customer experience

Companies constantly strive to provide better customer support as it greatly improves customer satisfaction and retention rate. In often results in customer support department being one of the most resource and cost intensive departments in a company. On the other hand, it is hard to link customer support to a ROI and a cost of handling customer support often outweighs the value which it brings to the table.

Al and chatbot technologies are dramatically changing the face of customer support. Companies are getting an opportunity to provide better experience in customer support and save money, all at the same time. From businesses perspective, automating a portion of customer support with chatbot technologies will result in considerable savings and increased customer satisfaction. It’s expected that by 2022, banks will automate 90% of their interactions using chatbots, and it will save 0.70$ per interaction.

Customer statistics is favorable to chatbots as well: over 50% customers believe that a business should respond to their inquiries 24/7 and over 60% thinks customer support should be available via messaging applications as well. This makes chatbot a perfect customer support agent that never sleeps and provides convenient and approachable support.

Second aspect that is absolutely worth visiting is the fact that chatbot can handle endless number of request simultaneously. Recall the times when you waited for 10-15 mins listening to on-hold music and how immensely frustrating this was. Chatbot will not make you wait and will respond to all requests immediately.

Third and very important aspect in which chatbots are exceeding their human co-workers is personalization. Same as humans, chatbots are able to recognize human emotions such as anger, confusion, joy and fear and act accordingly, or eventually delegate to human agent. The fact that chatbots are collecting data from previous interactions and actively learning from it, also goes in favor of technology. This will result in overall improvement and more personalized interaction over the time.

To summarize, chatbots are providing highly personalized customer experience available 24 hours a day, 7 days a week on customers preferred channel which holds a great value in satisfactory customer experience.

So we move onto the next point…

What makes a great chatbot
So what it is that makes a great chat bot?

Before you start implementing, place yourself in the user’s shoes: what experience are you expecting?

More than being able to believe that they are speaking with an actual human, users need to feel that they are being heard. Often conversation with chatbot will lack flow, feel clunky and robotic and end up in frustration. A lot of this has to do with chatbot lacking personality and empathy, thus we come to our first point: give your chatbot a distinct personality. This will require a good knowledge of your target audience, but the goal is to make interaction with chatbot easy and interesting. For example, if your target market involves teenagers, using colloquial language or emojis in communication with users may be extremely beneficial. If you are creating chatbot for banking market, you would want your bot to come across as being helpful and more polite.

Give your chatbot a name. Give it a back story, so when the user asks a personal question, like “What is your favorite song?” or “Where are you from?”, your chatbot can gloriously rise to the occasion. Inject humor where possible - being able to make user smile will make the interaction so much more pleasurable. Keep the language simple and slow down - you don’t want to overwhelm the user with an avalanche of information; you want to keep a steady pace, sending out content bit by bit, like you would do when messaging a friend. All of this will make your chatbot more relatable, friendly, and last, but certainly not the least, more fun.

Once you created an engaging persona for your chatbot, next important thing is bot’s ability to quickly identify and solve the issue. You don’t want to be stuck in a conversation with a chatbot that is asking you an endless series of questions, no matter how friendly or funny it is. You want your problem solved, as quickly and as efficiently as possible. Efficient use of Natural Language Processing (NLP) and Machine learning to the rescue! For bots to get better, they need to be able to learn from every conversations they have with users. Your initial bot may be pretty limited, but this way, its ability to understand, ask questions and maintain conversation flow will only increase over time.

Now that you made your chatbot both fun AND more or less efficient when it comes to understanding user input, you need to make it proactive. A large percent of bots existing today is purely focused on answering user questions in reactive manner, leaving it up to the customer to drive the conversation. Our bot needs to be able to facilitate conversation. It should be able to understand user’s intentions and predict their needs. This is incredibly hard, but learning their behavior and knowing their data, like locations, preferences, etc., may be incredibly powerful and useful. For example, let’s say you have a chatbot that provides personal assistant service. So you say to your bot: “Meeting with a boss tomorrow at 8 am”. Your chatbot asking you if you want it to schedule a meeting and set a reminder for you is definitely example of what chatbot “usefulness” is all about and clear example of value that chatbot can bring to more satisfying user experience.

So, with all of the new skills you gave it, it still happens that your chatbot can’t understand or solve the issue or perform a request. It happens to the best of them, but how you choose to react to this situation is what might set your bot apart from competition. Leaving some channels that will let your bot delegate the experience to the human agent once it’s exhausted all the possibilities is a good way to go and one that may be greatly appreciated by your users.

Limitations and challenges

Now that we have a plan, what are the limitations and challenges we will inevitably face while trying to build a worthy chatbot? Right now, most chatbots require us to master their syntax. Depending on how you phrase the question, you may or may not get the answer. To get around this, we need to make bots able to understand all the ways we may phrase something, as well as distinguishing different meanings depending on context. It needs to be able to understand multiple languages, colloquialisms, short forms or get around grammatical or spelling mistakes in user’s input. This is where natural language processing comes into play. But current state of NLP is not yet advanced enough to tackle everything, and for now, this makes for a pretty big limitation. When it comes to ML, learning from previous conversations implies collecting huge amounts of data. In order to utilize Machine learning to improve your chatbot, you will need to sift through the data and determine what parts are relevant and will bring the most value to your learning process.
One common mistake that companies make is thinking that in order to succeed, chatbot needs to be infinitely vast and complex, able to do all but make you a coffee. As it turns out, simple is the key of success. Keep it as simple and focused as possible, from the conversation to the interface. Don't try to address problems that go beyond your scope. Bots that do one thing well are more helpful that bots that do many things poorly.
Finally, the last big obstacle is our understanding of human interaction. After all, the technologies can be only as good as programmers that build them. This inevitably implies employing cognitive science and social psychology, which have a great deal to offer to those seeking a degree of interpretation, especially when it comes to “understanding” and “meaning”, as well as social mechanisms that need to be translated into human-computer interactions. Rather than teaching humans to understand machines, the tides have turned, and we’re now teaching machines to understand us.