The main mission of chatbots was to revolutionize customer service, but they were also expected to make their way into many other areas and even enter directly into sales. As recently as 2016, for instance, Chris Messina declared the “Year of Conversational Commerce.”
Things have calmed down considerably around the digital helpers since then. The reason for this is simple: They don’t work well enough. There are two reasons for this: For one, they continue to run up against technical limitations; and for another, they are not always put to wise use.
The original enthusiasm occurred against the backdrop of the triumph of the messengers. They were viewed as a major new platform, alongside the mobile web and apps. Now, though, no company can afford to speak with all of its customers and prospects around the clock. That’s where the chatbots come in.
Naturally, the idea behind them is not new. Automatic conversation machines like these are some of the oldest means of demonstrating something like artificial intelligence. But they are far from being very successful in this endeavor. They managed to celebrate a comeback two years ago, but this owed not just to the growth of WhatsApp et al. but also to the great optimism in AI development. Technologies such as deep learning had made impressive progress. Computers were expected to do simply everything.
Let us fast-forward two years. The consumer testing organization Stiftung Warentest came to a devastating conclusion in a comparison: “The telecommunications companies’ chat robots are not intelligent; in fact, they’re rather dumb, overwhelmed and moody.” The foundation had reviewed the quality of advice provided by eleven providers via telephone, chat and e-mail. “Often, the chatbots’ responses miss the point entirely, and they frequently either fail to understand anything at all or react irritably.”
So, to sum it up: To date, chatbots have adhered pretty closely to Gartner’s famous Hype Cycle: Following the “peak of excessive expectations,” we have now entered the “valley of disappointments.” The good news: Next up is the “path of enlightenment,” which will lead us to the “plateau of productivity.” The only question is how we intend to get there. This will happen in two steps.
The power of exponential progress
Part of the problems with today’s chatbots is that they still aren’t smart enough. General technological progress will help them in this respect – almost automatically. Modern AI systems are growing better and better as they are fed with more and more input.
Google, Facebook, Apple, Samsung and others are investing heavily in digital language assistants. They face lots of additional challenges compared to chatbots and have to cope with a wide variety of dialects and speech styles, for example. The environment in which chatbots work, on the other hand, is more clearly defined. They stand to benefit from the developments in this field. Because even though they still often react only to certain keywords, these new developments are likely to make them much more flexible – and hopefully smarter and more useful.
The great advantage of the “valley of disappointments” is that expectations are reset to zero. In terms of chatbots, for instance, people are beginning to realize that human conversations are amazingly complex. As Dave Feldman, formerly a member of the bot team for Facebook Messenger, writes here: They are not linear. Themes can revolve around one another, take random turns, re-start or end with a surprise. All of this is difficult for today’s computers to grasp.
The power of improved applications
Another problem with chatbots is that they are used in the wrong applications, entrusted to accomplish too much, or simply not implemented well. People should relinquish the notion that they currently have the capability of replacing a conversation between two people. Nor are chatbots an alternative to websites or apps. They are another way to convey information and get in contact with one another.
DMEXCO exhibitor LoyJoy has created a chart on the five most important rules for chatbot dialogues. You can download it here for free.
They work well hybrid applications. The tool HubSpot Conversations, for instance, combines communication from a wide variety of sources under a single roof. No matter whether the customer walks in the door via Facebook Messenger, social media or e-mail, everything occurs in a single place. And chatbots are touted as an extension – not a substitute.
Multimodal apps are a further approach: they use the medium that works best at the moment. A user might ask a spoken question but then be provided with a list or a map as a result, if that is the best way to get the desired information across.
Facebook has also just unveiled a new tool in which automatic and human interlocutors replace one another, depending on the situation. Generally speaking, it turns out that the best policy is to be honest with the users. Bots should be recognizable as bots. In California, this will even be required as a matter of law beginning in 2019.
As indicated above: If chatbots adhere to Gartner’s hype cycle, a few years from now they’ll be a natural part of the toolbox for customer service, marketing, and perhaps even sales. After all, messengers will not vanish so easily, and more and more users are already speaking with their devices via Alexa, Siri et al. Technical progress will accomplish the rest.