Three years ago, people thought that chatbots would automate most tasks in our lives. They expected that chatbots would automate customer service, system support, and purchasing, and even act as sales agents. Moreover, VentureBeat expected that chatbots would replace most mobile apps. Today, almost all chatbots are not doing well. In addition, many tech giants have scaled down their investment in this technology. So, what is the future of chatbots?
The rise of chatbots
Back in 2016, many tech experts expected that chatbots would be the next big thing. Satya Nadella, Microsoft’s CEO, expected that bots would be the new apps. In this year, software giants like Microsoft, Google, and Facebook invested millions of dollars in chatbots. In addition, many startup accelerators incubated dozens of startups developing chatbots. All this caused developers to consider building their own bots.
Unfortunately, most chatbots that have been developed so far are pretty useless. These bots were designed to automate things that were already automated through other technologies. For example, reserving event tickets can be done through the organizers’ websites. Why, then, would someone prefer to chat with a bot to reserve these tickets? Furthermore, using these bots takes much more time than using other channels.
According to Facebook, the failure rate for these bots hit 70%. Apparently, this rate can never be acceptable to any user, especially if alternatives exist. In my opinion, no one will bother using a technology with such a high failure rate. The problem with the existing chatbots lies in Natural Language Processing (NLP) technology. Current bots cannot conduct complex conversations in which any human can engage. Accordingly, only 30% of interactions with bots were successfully achieved without human intervention.
Besides the failure rate, most developers have lost interest in building apps with conversational interface. The traffic that most chatbot-building frameworks receive is shrinking over time. For example, the global rank of wit.ai (owned by Facebook) dropped from around 70k to 170k during the past 12 months. This indicates that most developers are no longer willing to invest in building new chatbots. In addition, they are no longer interested in improving or maintaining their existing bots.
Due to the high failure rate, as well as the narrow scope, many firms decided to dump their bots. In January 2018, Facebook killed its flagship messenger M, as most of its tasks were done manually. In addition, many businesses that relied on chatbots to reduce customer service costs ended up dumping these bots. According to PointSource, 80% of people prefer to chat with humans over bots. We also noticed that most startups that were developing bots in 2016 didn’t survive for long.
What is the future of chatbots?
Answering this question isn’t easy. Every technology has its own hype cycle through which it grows and declines. You can read more about the technology hype cycle in our previous article. In July 2017, Gartner placed virtual assistants, to which chatbots belong, in the “peak of inflated expectations” phase. Accordingly, I believe that chatbots and NLP are now in the “trough of disillusionment” phase. If we’re correct, chatbots will probably rise again.
If we compared chatbots with other mature assistants (like Siri, Alexa, and Google Assistant), we can guess why these bots failed. The main problem with the current bots is their wide scope. On the other hand, mature assistants, like Alexa, have a very narrow scope. That’s why these assistants didn’t disappear like bots. In other words, to succeed, every bot must be built to tackle a very niche problem.
Besides the scope, developers must invest a lot of time in training their bots. NLP needs a lot of training so that it can understand user requests. In addition, building chatbots through platforms that require no coding proved to be a waste of time. Almost all bots built through these platforms failed to gain any traction. For the NLP part, developers can either rely on open source solutions (like RASA.ai) or use online NLP services (like DialogFlow, Wit.ai, etc.).
According to MIT Technology Review, the use of conversational bots isn’t always desirable. In addition, relying on interactive controls (like images, cards, buttons, etc.) proved to be much more efficient. In other words, because NLP technology isn’t mature enough today, developers can rely on interactive controls to better understand user requests.
Besides chatbots, some platforms can now build assistants that are more robust alternatives. For example, Advicenode can build websites and native mobile apps that act as assistants. These websites and apps can understand user requests through many interactive controls. Unlike bots, they can have very complex logic, solve decision-making problems, and integrate with any online service. To build apps through Advicenode, people must first submit an application proposal.
Is building a chatbot viable?
If it can automate a niche area and will be well-trained, then yes, it will be viable. We discussed before how to know whether an idea is viable. Once the expectations in any technology fall, people will start using it in the proper way. Personally, I believe that chatbots can successfully automate many areas in online retailing, like suggesting products or guiding users. They can also be used to automate some areas in product support. In addition, many firms started using a hybrid model between bots and human support. When the bot fail to understand the user’s requests, it will refer him/her to a human agent. This way, companies could reap the benefits of both.
Have any questions? Feel free to post a comment below!