AI can’t solve your customer service problem. Here’s why.
Everyone’s talking about AI. Like “the cloud” numerous years ago, if your product isn’t powered by AI, forget it. If you believe everything you read, you’d think AI is the answer to everything from driverless cars (which it is) to natural and engaging chatbots, intelligent assistants, and IVR systems (which, unfortunately, it isn’t).
The problem is in the terminology. Intelligence is a broad term, covering everything from low-level human abilities like hearing and seeing to higher-order abilities like planning, strategy, and innovation. Now, if our artificial intelligence could do all of those things, then we could fill our contact centers with them, and we humans could pack our bags and head to the beach, but the reality is that, usually, when people talk about artificial intelligence, they’re talking about machine learning, which is a powerful but limited technology that’s been around for several decades.
AI and machine learning
Machine learning software takes an input, let’s call it A, and maps it to an output, let’s call it B. It “learns” how to “map A to B” by processing lots of examples, and that’s where the recent advances have come from. We have a lot more examples – data – to work with, and we have much more powerful computers with which to process the data. But the algorithms themselves haven’t changed much, and the types of problems they’re good for are at the hearing and seeing end of the intelligence scale. They are certainly not good for planning, strategy, and innovation.
What does this mean for conversational chatbots, intelligent assistants, and IVR?
Well, there are some important “map A to B” tasks in this area. Voice-driven intelligent assistants and IVRs need to map a customer’s speech patterns (A) to words (B) – that’s speech recognition. IVRs and chatbots need to map words (A) to what a customer wants to do – their intent (B) – that’s natural language understanding or NLU. You can even map intents that are questions or search terms (A) to answers in a knowledge base (B) – that’s a kind of search. All of these things can, and are, learned from data. But that’s about as far as ‘map A to B’ can get you.
To create natural and engaging conversations that help customers get stuff done, like checking the status of an order or making a payment, you need a much more complex algorithm than “map A to B.” As a bare minimum, you need to understand the structure of the task, the order in which customers find it most convenient to complete the sub-tasks, and strategies to recover from errors. For a branded, engaging interaction, you also need a conversation design that feels natural and portrays a personal – the character of the system – that matches your brand and your customer’s expectations of it.
In theory, this could be reduced to map intent (A) to system response (B) if you had enough data to work with, but we’re talking orders of magnitude more data than the lower-level speech and intent recognition tasks where machine learning excels today. Several years in the future — maybe 5 or 10, although it could be sooner than that — it will be possible to learn these tasks, but today you need a designer to create them. Real intelligence, if you like, not artificial intelligence.
Great Conversation Design needs Great Conversation Designers
Great conversational AI designs and designers draw upon a host of data – including call center statistics and call or chat transcripts as well as a deep understanding of the way human beings use language. They understand how to combine the machine learning components like speech recognition and natural language understanding with carefully crafted dialogues and prompts that are natural, engage customers, and communicate your brand.
The next time someone tells you their new cloud, chatbot, or conversational AI technology is the answer to all of your problems, ask yourself, is your problem as simple as “map A to B?” If not, then take a step back, and as Steve Jobs said: “Start with the customer experience and work back to the technology.” AI or machine learning might be part of the answer, but more often than not, you need a conversation design and, therefore, a conversation designer who can craft an experience that your customers will love.
Check out “Bot Analytics: Keep Tabs on Your Bots, So They Don’t Run Amok.” Get the framework and action plan needed to properly measure and optimize the performance of your conversational AI. Whatever the platform.