What conversational agents can learn from driverless vehicles
“By 2020, customers will manage 85% of their relationship with the enterprise without interacting with a human”, Gartner predicted in 2011.
Many misinterpret this claim. The statement does not mean: “chatbots will handle 85%”. ATMs, the first of them already operational more than fifty years ago, didn’t require human interaction. Online shops gave another boost to humanless interaction. And these are but two examples of what is included in that percentage.
Conversational agents, supported by NLP, cannot realize such a business value yet. But hold on, there is absolutely no reason to throw away the baby with the bathwater.
Natural language processing (NLP) is to modern customer service what computer vision is to self-driving cars: a key enabler.
Driverless vehicles are rarely seen on the road. Too many technological, social and legal problems remain unsolved. Many more test miles are needed. Contrary to that, driver-assist technology is widely adopted and ubiquitous. Tire pressure control, GPS and other smart options make car driving safer and easier.
In parallel, augmentation—NLP and Machine Learning technology supporting human agents instead of replacing them—is the customer service equivalent of driver-assist technology. It is the valuable and realistic middle ground between the unsupported human agent and the autonomous question-answering bot. Indeed, autonomous conversational agents are very promising, in the same way autonomous cars are. But in real life, interactive agents still face a large number of issues. Hence, they are applicable in only a few well-identified and ultra-specific areas.
As a whole, the aspirations of conversational bots are still too difficult. Nevertheless, NLP can and do support human agents in a range of sub-activities already. Content-based routing and prioritization of incoming messages, language and sentiment detection of complaints, structured information extraction from customer requests, and case-based reasoning on reported problems are but a few successful examples where technology assists, without fully taking over from the human.
Awareness about a technology’s maturity level is crucial in generating beneficial and reliable use cases. Gartner’s technology life cycle illustrates that overconfidence and exaggerated expectations in an immature technology leads to a hype, followed by a disillusionment. Unfortunately, overoptimistic, and failing, chatbot initiatives may cause such a deception.
Experience leads us to conclude that augmentation is a realistic and reliable alternative for most sub-activities of customer service; full automation only for a few.