In the world of customer service, we're experiencing the rise of chatbots, virtual digital assistants, and artificial intelligence (AI) agents, answering basic queries which allow humans to tackle more complex problems and improve the speed and efficiency of decisions.
Emerging technology such as machine-learning applications, chatbots, and mobile messaging will play a much more significant role in customer interactions in the next five years.
The data backs that up. According to Juniper Research, chatbots will create a cost savings of more than $8 Billion annually by 2022, up from $20 Million in 2017. In the enterprise, 31% of business executives believe virtual personal assistants will have a substantial impact on their business, more than any other AI-powered solution according to a PwC report in 2017. And 34% of business executives say that the time saved that was generated from using digital assistants allows them to focus on deep thinking and creating.
And this is the big reveal, in that same report by PwC, 27% of consumers said they weren’t sure if their last customer service interaction was a human or a chatbot.
Delivering an exceptional customer experience remains one of the most important differentiators for businesses today. Because we now live in a 24/7, 365 day a year on-demand world, companies are turning to AI and machine learning (ML) to step in and automate repetitive tasks and alleviate mundane functions from customer service agents which allows them to build better relationships with customers and focus on tackling more complex customer conversations. With AI and ML, customer service centers will have deeper foresight into future customers outcomes, more recommendations on the team’s next best course of action, and can be more efficient due to the automation of simple support tasks.
According to Adrian McDermott, President of Products, Zendesk, there are three areas where AI will significantly improve the customer experience and augment the role of the agent. These include automating simple interactions with customers (using a bot to send a customer a knowledge base article such as a returns policy); augmenting an agent’s abilities (predicting whether a customer is struggling and at risk of having a bad experience); and, automating internal tasks like routing a customer request to the correct agent.
"Over time, we’ll see agents become more experienced, as their roles become more strategic and productive," said McDermott. "We’ll also see more roles dedicated to designing customer experiences and training algorithms to automate specific tasks within those new customer journeys."
"We know that many businesses don’t have the resources to staff their chat team operations 24x7. That said, based on research from more than 50,000 Zendesk customers, a top driver of customer satisfaction is the time it takes for a business to resolve a customer’s issue. AI can automate many of these simple interactions, allowing customers to get the answers they need in real-time and significantly improve customer satisfaction and retention. Additionally, research has shown that customers increasingly prefer resolving their issue without another human-in-the-loop," said McDermott.
Chitra Dorai, Ph.D., and IBM Fellow, Master Inventor says that AI is simply reimagining the way professionals work by scaling our knowledge and elevating our expertise with AI systems in everything we do.
"Things humans do and excel at are to do with decision-making. If the work is routine, we have automated," said Dorai. "Even in decision-making, in today's world of the digital explosion of data - the volume, variety, and velocity - all make it impossible for humans to comb through them and extract the relevant information to aid in their decisions. We have begun to rely on AI systems to collect and combine, and analyze diverse data sources to extract new insights that can augment our intelligence and expand our knowledge."
Chitra adds, "Imagine, giving one employee the insights of 1000 employees with the help of AI. A great case in point is Woodside Energy in Australia where the company is using AI to create a sort of tribal knowledge that's available to all of their employees."
"Jobs will change, as they often do with the emergence of disruptive technology, but we cannot mistake disruption for replacement — humans in every field will start to work smarter and faster alongside automation," said McDermott. "Businesses have a large role in smoothing this transition. The most progressive and successful companies will work hard to support the development and growth of their employees while redeploying time saved to improve their customer experience."
But, at the same time, these algorithms driving AI and ML in the enterprise and customer service applications are driven by data generated in the real world, reflecting real-world biases which can have unintended consequences on our society. How do businesses avoid algorithmic bias as these chatbots and AI customer service agents become more sophisticated?
Jason Maynard, Vice President and General Manager, Guide & Data Products, Zendesk says that for many customer service applications, there aren’t the same risks of algorithmic bias replicating human bigotry as there are in other fields, such as approvals for financial services or job applications.
"That said, there are many areas where algorithmic bias can be introduced that lead to suboptimal experiences," said Maynard. "Customer service is a tough job, and many teams are overworked and undertrained. This means that often times their interactions with customers can be suboptimal. If these interactions are used to train an AI, then it can result in the same suboptimal interactions being automated."
Maynard says that can result in an irrelevant message being sent to a customer or the incorrect routing or tagging of a customer request.
"For most machine learning algorithms data is the bias - or the absence of it. If the outcome of the algorithm is primarily based on examples used to train it, the only current way to prevent bias will be to audit the data and test it against fairness, said Mars Cyrillo, Vice President, Machine Learning, CI&T, New York. "This may be tricky because you still need people - and their own biases in the loop, it's still the best way to go, and biases may also occur consciously or unconsciously when the data scientists working with the data-set do what is called feature engineering."
"This process will slice and dice the data-set to feed the algorithm with what the scientist believes is relevant. It's one of the most time-consuming tasks when training a typical ML and still very dependent on manual work," adds Cyrillo.
Zendesk's Jason Maynard agrees that data is critical.
"Except for some very recent AlphaGo research, almost every applied application of machine learning today is supervised — this means that an algorithm infers the best course of action based on a labeled set of training data, and usually labeling is done by a human. When building AI, it is imperative to find ways to create the most relevant set of data to train AI," added Maynard.
Zendesk says they developed a technique to train their Answer Bot algorithm to only replicate positive customer interactions. They have millions and millions of interactions where a human agent has sent a knowledge base article to a customer, but dozens of times that agent didn't get it right and submitted an article that didn’t resolve a customer’s issue.
"We only wanted train our AI on positive customer interactions, so we developed a two-stage learning approach where Answer Bot first learns to identify interactions where a customer-validated that they received a helpful article, and had a great experience. The algorithm only learns from interactions that fit that profile. This effectively created an algorithm that “learned” to remove the bias introduced by human error," said Maynard.
Dollar Shave Club used Zendesk's Answer Bot to resolve around 10% of their customer interactions which saved their customer experience teams a lot of time. With the extra time, they created a task force that generates additional knowledge articles for Answer Bot to use and also review Answer Bot tickets which provide feedback on where the model has learned incorrect patterns.
"All of this feedback has been reincorporated into the model, much like a human agent would learn from feedback, and over time these mistakes are resolved," added Maynard.
Cyrillo says that the most important contribution of AI in general terms, it's automation.
"Any company regardless of their field will have hundreds, if not thousands, of processes that could be partially or fully automated by AI. This is the reason there are so many people worried about mass job losses. But this will quickly become the status quo like the use of computers once became necessary to run any company. I believe that in the future, the biggest value-add will be a deep understanding of the DNA of the company and needs from the market that AIs will be able to grasp," said Cyrillo.
