Hard to imagine now, but in 2020, 85% of customer contact will be via chatbots. Today, however, talking to most chatbots feels robotic. It’s efficient at best and a terrible impersonation of a human giving all the wrong answers, at worst.

The interesting thing about computers is that they perform extremely well on tasks that humans find very difficult, but are rather poor at tasks that humans can do without thinking at all. For example, computers can multiply very large numbers but can’t have a fluent conversation about the weather.

For a computer, understanding all the subtle and little things in human conversation is complex work, such as the meaning of minor inflections in someone’s voice, emphasis behind annunciating one word and not the other or the hint of a laugh, which indicates the speaker isn’t all that serious.

>See also: The chatbots are here: are you prepared?

While written language is a little less difficult for a computer to process, it’s still missing that human interpretation of conversations. We know how to interpret the ‘uhm’s, mmm’s’ and sarcasm that someone else might type and can react to it appropriately. We know that a profanity means someone is getting very upset, but the same profanity could mean someone’s really excited.

Most chatbots used nowadays are unable to have actual conversations. They are positioned on a company’s landing page and are able to recognise some key words, for example, ‘order status’ or ‘payment’. Once recognised, they will respond by giving the customer a number of items the question might be about. In the example of ‘payment’, the chatbot can ask, “is your question about making a payment or about receiving a payment from the company?”

Once an item is selected by a customer, the chatbot will drill down some more or give a generic answer satisfying 80% of the visitors; for example, it could tell you that you’ll be receiving money back from the company in between 2-4 weeks and it will be on the bank account from which the original transaction was made.

How will chatbot technology evolve in the future?

Slowly, however, we are starting to see a trend towards chatbots (virtual assistants) that can function inbound as well as outbound and are able to hold actual conversations.

The trick described above to pre-program a set of responses will not work in this case. It will be a life-long, if not impossible, task to teach a computer one response to all these variations by giving it specific rules.

It is here where recent improvements in artificial intelligence and natural language processing can cause a handful of chatbots to perform quite well in a natural human interaction. For a while, you might actually be fooled to think you are talking to an actual person.

>See also: Chatbots can do more than chit-chat

The machine learning approach to letting a computer engage in a conversation is not to try to give it all the potential questions and answers, but for the computer to learn from actual conversations that have occurred in the past.

For a part, this learning is ‘supervised’. This means that we let the computer experiment by trying to do a task itself. Take the example of a company’s chatbot deciding whether or not a customer’s remark requires the attention of a human agent. We give the computer lots of data from actual conversations, including the ‘right’ answer. The computer is then taught to identify patterns in communications: What words in a sentence indicate someone is getting upset and needs human attention and which words indicate someone is pleased and the conversation can be closed.

The computer will cluster the conversations into ‘needs human attention’ and ‘does not need human attention’. For every answer, humans will reinforce its learning and indicate whether the computer did it right or not. This will eventually enable the machine to make the right decision about 95-99% of the time.

Opportunities chatbots, AI and machine learning present to companies

The rise of artificial intelligence and machine learning in chatbots also gives companies great opportunities to deliver better and faster customer service. Simple questions can be answered instantaneously by a chatbot without a customer having to wait in a queue of callers first.

Conversely, the humans operating the contact centre will have more time and energy to help customers solve the more complex issues without having to deal with boring and simple questions any more.

>See also: Chatbots for customer service in industrial IT

And why not stretch it even further? With enough data and learning, computers can quite accurately predict the probability of problems or question arising in the future. Knowing this ahead of time makes it possible for companies to be increasingly pro-active and solve a number of issues before they actually arise.

Think of an invoice deviating a lot from the usual amount due. Explaining why this is so will leave the customer feeling attended to and respected ahead of time, rather than potentially calling in fury after receiving the invoice.

From a customer experience point of view, another interesting application of using the insights from big data analytics is to support the contact centre employee by making decisions.

For every customer that calls, the virtual assistant can help the real life assistant by using all the data and context of the moment to suggest a number of relevant ways on how to help the customer.

Sourced by Laura Van Beers, head of AI at automated customer communication technology leader, ContactEngine