How To Successfully Deploy AI in Customer Service

Artificial Intelligence has been grabbing the headlines recently, from Google's Go victory, the use of IBM's Watson to provide insight in medicine, to the surge of interest in customer service bots partly fuelled by Facebook's Messenger platform. 

But not all the headlines have been positive. If you're looking for examples of the risks of badly deploying AI there's probably none more powerful than Microsoft's Tay. Designed to be the pinnacle of natural language and machine learning interaction, it instead transformed in to  a "Hitler loving sex robot".

How? Because it's responses were created based on learning from conversations it had with real people - and in the words of The Telegraph, "real humans like to say weird stuff online and enjoy hijacking corporate attempts at PR"

So how do we leverage AI to provide the instant, always-on service support that customers want - with zero risk of innapropriate content?

The answer is really simple. You've already deployed autonomous intelligence in your service environments - more powerful than any machine intelligence has ever achieved - in the form of your human contact centre advisors. We have hundreds, even thousands of them working on individual brands. But instead of encouraging these highly intelligent brains to create NEW answers and NEW processes to the majority of questions, we in fact wisely do the opposite: we spend large amounts of resources, time and effort training these agents to provide the single correct answer (or guide the customer through the single correct process) once they've established the customer intent. 

The role of intelligence in automated customer service and sales is to understand intent, to provide the single correct answer, and to measure and optimise that process. It is NOT to create new responses. 

Just the same as a human advisor, our customer service bots should understand what the customer is saying in natural language to derive their intent. Just the same as a human advisor, they may need to clarify that intent by asking questions or providing options back to the customer. And just like a human advisor, wherever possible they should be leveraging contextual data (eg. has this customer recently booked a flight with my airline?) to short cut clarification and make the process fast and easy for the customer. 

And it's as simple as that. The line is drawn at understanding intent and never crosses to creating new content. Even though there are millions of ways that customers can ask a questions or state their problem, there are limited number of underlying intents that they will have. Each and every intent has a response approved by the business. It works for our human operations, and it's worked for 150 Virtual Agent deployments by [24]7. 

So what if there's a new intent? Or the issue is sensitive, perhaps a complaint?

Well sometimes, we do need a little human creativity and sensitivity, and the answer here is simple - escalation. Let's not risk asking our bots to try to create something new, or try to emulate the empathy and soft skills of a human. Instead let's use them for what they're best at - determining whether or not they should assist - and if not - providing escalation to human chat or connecting the customer to a human advisor with a phone call.

So does deploying a Virtual Agent or Chat Bot sound a little less risky and a little more achievable?

I hope so. If you're thinking about whether or not it's right for your business then I'd recommend this great article by PV Kannan as further reading. 

Or ping me a note with a question (as I think about it, there's probably a limited number of queries that you're likely to have, and they might just be perfect candidates for automation...)

_______________________________________________________

If you liked this post, you might also enjoy reading: Can We Trust Intelligent Machines To Serve Us?

Want to transform your customer experience, self service and channel mix? Contact me on: 

+44 (0) 7534 058 945 / [email protected] / @tweetjonwebb