When that friendly face pops up in the corner of the screen and asks “Hi! Can I help you with something?”, you never know if you are going to be chatting with a person or a chatbot. But until recently, you usually found out pretty quickly. The rule-based “bots” that predated commercially-available AI were easy to flummox, repeated responses frequently, and were generally about as helpful as a t-shirt in a blizzard.

The next generation of voice- and text-based chatbots are much smarter and more human. But are consumers ready to talk to bots? More importantly, are they always a viable substitute for help with a heartbeat?

Recent Advances in Chatbots

Advances in AI and natural language processing have made conversations with chatbots startlingly natural. Just take a look at the Google Duplex demo from this year’s Google IO.

The world was absolutely floored by how naturally Duplex was able to navigate a conversation with a real person. It handled unexpected twists and turns better than many people would, and it was nearly impossible to tell whether it was Google or Gabriella on the other end of the line.

The very meaning of “Duplex,” to have a true two-way conversation with a bot, was virtually science fiction just a few years ago. While you can ask Siri or Alexa questions and get answers, you aren’t really having a conversation, so to speak.  

A number of factors have made human-to-bot conversations possible, the first being a huge leap in the perfection of speech recognition. In 2017, Google’s machine learning algorithms achieved a 95% word accuracy rate for the English language. We’ll call this “perfect” because that’s about the accuracy level of a typical person.

Duplex in particular benefits from years of Google’s research in speech recognition and its recurrent neural network that’s designed to cope with the challenges of understanding and processing natural speech. When we talk normally to another person, we speak quickly and not particularly clearly, make corrections mid-sentence, interrupt conversations, and frequently throw in speech disfluencies like “hmm”s and “uh”s that make it difficult for machines to understand what we’re getting at. Think about how you talk when you know you are speaking to a machine on the phone — you use as few words as possible, speak loud and clear, and don’t ask any questions. You usually have to speak like a machine to communicate with a machine, but this new technology will soon make that a thing of the past.

The current sticking point is that even advanced chatbots like Duplex can’t carry on just any old conversation. Google Duplex has very specific and deep training to make appointments, which was based on piles of anonymized phone conversation data. But the potential applications for this technology are myriad and they could, for example, turn today’s clunky voice IVRs into a much more pleasant, conversational experience that doesn’t just end in yelling “OPERATOR!” over and over.

How People are Responding to Chatbots

The reaction to Duplex has been a mixture of awe, outrage, and horror. Despite the ethical questions it has raised and the mixed public reaction, some research shows that people are actually ready to chat with bots. According to the 2018 State of Chatbots Report, 15 percent of people have used chatbots to communicate with a business in the last year. While that may not seem like a huge number, it is still significant and will likely continue to grow as the bots become more advanced and useful.

But the world isn’t ready to go all-bot, all the time. According to the report, 43 percent of consumers would still rather talk to a real person, but 34 percent say that they expect to use a chatbot to get connected to the right person. Compared to other business communication channels, chatbots came in second when it comes to getting instant responses, only losing out to online chat. Which, of course, can now be handled by a bot.

When Chatbots Work Well

As more consumers handle a majority of their tasks and transactions online, they are also becoming more impatient. People want to find the information that they are looking for instantly and they don’t want to dig around a website to find it (thanks, Google!) The most common complaints are not being able to find answers to simple questions like business hours, locations, and inventory, and that it takes too long to find services. If the information is more than a couple clicks away, you can plan on customers leaving to see if your competitors can provide the answer faster. This is where chatbots can excel in serving customers — giving people a simple path to instant answers to simple questions.

When compared to phone calls, consumers believe that chatbots can provide quicker answers to simple questions, are more likely to provide 24-hour service, and that they are a bit more convenient. But the margin of victory for the bots are fairly slim, so it does not appear that they are ready to replace phone calls or emails for communicating with businesses right now. And there are some areas where chatbots fall significantly short of meeting consumer expectations.

When Chatbots Fail

One area where chatbots aren’t ready for primetime is high-ticket items and considered purchase categories like home mortgages. When consumers are considering purchasing items that are either costly or could have far-reaching impacts on their lives, they trust people more than a bot. About 30 percent of people think that a bot is likely to make a mistake and accordingly, only 9 percent of consumers find chatbots useful in their attempt to purchase an expensive item. When it comes to getting detailed, expert answers, a phone call is also preferred to chatbots by over 20 points. But this sentiment may again change soon, as the phone beat bots by just 1 percent among millennials in the same survey — well within the margin of error. Another survey we cited in our post on AI in the Insurance industry also showed that preferences for using chatbots to get information from insurance companies are no different from the preferences of engaging with retail companies. It appears there is some trepidation in trusting the decisions of bots when the stakes are high, but people are nearly all-in when it when it comes to increased convenience.

When bots fail, that’s when it’s time to bring in the humans. According to our Rise of Voice report, 76 percent of people said that if their voice assistant could easily connect them to a human who could answer their question, they would rather do that. The same sentiment likely applies to chatbots, so making it easy for customers to connect with a human when your chatbot lets them down should be seriously considered. 

Chatbots are Part of the Customer Experience Strategy

Chatbots aren’t ready to jump in and replace call centers, but they are an excellent way to supplement human assistance. Customers want to get answers to their questions instantly, 24 hours a day and a conversational chatbot can make that a really pleasant experience. Chatbots can communicate more like people, and they can not only take the burden of fielding basic inquiries off of call center staff, they can help direct customers to just the right person or department when the time comes to call. On mobile or with virtual assistants, this can also be turned into a seamless, no-click, single-device experience for customers.

The age of the chatbot is indeed here, and it should be a part of your future customer service strategy.

Get the Invoca Rise of Voice report to see what the proliferation of AI, voice technology, and conversations mean for marketers. 

Owen Ray

Posted by Owen Ray

Owen Ray is the Senior Content Marketing Manager at Invoca. Prior to that, he worked with SaaS companies like Aria Systems, Glassdoor, and Mindjet. Owen sharpened his writing tools at San Francisco State University and Bay Area newspapers before working his way into the Silicon Valley creative services set. He hails from Petaluma, California and definitely does not leave work early on Wednesdays to go drag racing at Sonoma Raceway.

Leave a reply

Your email address will not be published. Required fields are marked *