Automating customer service is the holy grail. Wouldn’t it be great if you could generate automatic answers for most common service questions? There are many solutions in the market that promise automated customer service in a personalised way, hassle free and with friendly pricing. But to get a grasp of how effective these solutions are – from chatbots to Intelligent Virtual Assistants – it’s important to understand how they’re being developed and how they operate. You need to know how to make a successful chatbot for the enterprise.
Most solutions are good at computations. Machine learning allows them to look up date, calculate numbers and recognise patterns. They consider millions of fact-based options and find the right one in seconds. Show them 500 million pictures and ask them to select the ones with bicycles in them and they can do it. That’s providing, of course, that you have taught them exactly what a bicycle is, how varied they can be and how a bicycle is different than a unicycle, motorbike or whatever…
In short, they only know what their human engineers have already told them. Once they have that knowledge they can apply it with astonishing speed. But their accuracy will depend entirely on how well they’ve been taught.
Deep learning is advocated by many as the next big step forward because it allows a computer programme to ‘acquire knowledge’ independently. In short, it allows chatbots and IVAs to be programmed to ‘learn’ based on the patterns they repeatedly see in the data – primarily historical customer interactions – they’re asked to process. When a certain question has been asked and answered thousands of times, it will pick up on it and learn how to answer that question automatically.
This obviously isn’t fool proof. Brilliant as deep learning is, its conclusions are based on a fabulously complex evaluation of statistical probabilities – not on known certainties – so miscalculations, false assumptions and mistakes must be accepted as a common fact of life. What? When you automate customer service solely depending on deep learning, you are going to make mistakes. No exceptions.
So what does this mean for customer management?
It means that although machine learning and deep learning are a gian leap forward, they can’t, on their own, deliver automated customer service that your customers will love and set you apart as a business. Think of the wheel – a transformative technology in its own time just as these learning technologies are in ours – it only achieved its full potential when we attached it to a cart, then a car, a train, an aeroplane…
The short answer is that we need to build solutions that utilize the potential of deep learning, but are embedded in a platform that allows you to control the operation to maximize its value. Add a summarisation feature, use business rules to control the output, add compliance and quality rules to protect your brand and reputation, focus on security to become enterprise ready.