In 2016, Microsoft released an online chatbot called Tay, built with a mix of artificial intelligence and content crafted by writers including improv comedians. The bot was designed to engage with the 18-to-24-year-old audience that already connects with friends through online chat platforms. But within a day, the company had pulled poor Tay, which had begun to regurgitate racist comments and questionable material it picked up from online trolls.
"We take full responsibility for not seeing this possibility ahead of time," Microsoft VP Peter Lee wrote in a blog post.
News of Tay's demise came during a well-publicized push by Microsoft and its social media rivals into automated chatting assistants. Robotic conversation partners from major brands were brought into the world—integrated into chat platforms such as Facebook Messenger, Skype and Slack—to offer customer service, provide shopping assistance and take pizza orders. One UBS analyst even warned the robots could pose an "existential threat" to Apple's smartphone dominance: If bots proved popular, smartphone users could leave the App Store behind when it came to things like ordering food, and simply interact with online services through text.
But as Tay and other prominent failures suggested, building bots interesting enough to draw users and media attention yet reliable enough to represent a prominent brand online is not easy, which is why the early excitement around chatbots has dimmed. Facebook notably shut down M, a limited-release digital "concierge," earlier this year. Other companies have scaled back their chatbot operations too, including fashion retailer Everlane, which announced last year it's reverting to old-fashioned email for customer notifications. Meanwhile, voice-powered assistants like Apple's Siri and Amazon's Alexa have elbowed their way past texting bots to the center of the cultural limelight.
Yet many industry observers say the chatbot dream is far from dead. In fact, research firm Gartner recently predicted that by 2021, 50 percent of enterprises would spend more on bots than mobile apps. It's more accurate to say brands and bot builders have spent the past couple of years learning what consumers actually want in a conversational interface.
"I think, in general, the user experiences that we try to get away from are entertaining use cases or trying to mimic human behavior," says Jonathan Shriftman, director of business development at Snaps, which helps brands build bots. In other words, bots with overly cutesy personalities are mostly out. And ones that try to have open-ended conversations about anything and everything turn out not to be particularly effective; with the current generation of artificial intelligence, they're more likely to get confused or slip into not-quite-English colloquialisms. Either way, users leave frustrated.
Instead, the bots that seem to work best (i.e., drive up sales and customer satisfaction numbers) guide humans through simple tasks: asking a question or placing an order. Sephora reported in 2017 that in-store makeover bookings rose 11 percent after it let customers book them through a bot. And according to research Gartner unveiled in February, companies see a decline of up to 70 percent in call, human-to-human chat and email inquiries after implementing bots, as well as higher customer satisfaction scores.
Companies that already connect consumers with human agents through messaging platforms can have an advantage, says Lauren Golembiewski, co-founder and CEO of Voxable, another bot-making agency. That's because they know what types of questions people tend to ask through chat services, and for a minimal additional tech investment they can design bots to work alongside the humans to field some of the simple, more common requests—think password resets and shipping inquiries. That helps cut costs and often boosts critical customer satisfaction measures like Net Promoter Score. After all, consumers, especially younger ones, are already using messaging tools to talk to friends and family. More than 1.3 billion people use Messenger every month, according to Facebook.
"Businesses that take people's messaging behaviors and preferences into account are the ones that tend to see the most success," Itai Leibowitz, Facebook's product manager for the Messenger platform, writes in an email.
It turns out that while clever language tricks and conversational fluency aren't so important, understanding customer needs is. For instance, a financial app asked to transfer funds should probably warn users they could overdraw their accounts, even if they don't ask for a balance check, says Susan Etlinger, an industry analyst at research firm Altimeter. A bot from 1-800-Flowers lets customers begin a transaction when they have a moment, then seamlessly pick it up again to finish when they have more time, something any online shopper knows is often hard to do on the web and usually impossible over the phone.
Some of the best bots use platform-specific tools to offer more complex options. Mode.ai, which works with fashion companies like Levi's and Louis Vuitton, has enabled Messenger users to send snapshots of interesting outfits to branded bots; the bots then use image-recognition engines to find matching outfits in the brand's catalog. Delivering as smooth an experience through a traditional website would be a challenge, says Mode.ai Chief Business Officer Karen Ouk.
Chat platforms allow companies to mix the personal assistance customers would find in a brick-and-mortar store with the anywhere, anytime availability they expect from e-commerce. And each company's bot needs to adapt to its surroundings. Work chat app Slack has its own style of bots. Bots that post to broadcast-centric Twitter have a different flavor than those on chat sites. And brands from Sephora to CNN have built bots on Kik, a platform with a younger user base, that deliver teen-friendly content like GIFs, image stickers and quizzes.
"You need to know your audience and build for your audience and connect to that audience," says Sergio Silva, Kik's director of partner success. "You're not going to build the same bot on Kik that you are on Facebook."
Exactly what kind of bots companies will build in the future is still uncertain. Messaging platforms themselves are still evolving and adding features from new emojis to in-app payments, and consumers and brands alike are still learning how to use them to transact business. The jury is still out on exactly when or if consumers will choose to use text-based tools like Messenger versus voice platforms like Alexa. And it's likely we'll see more brands kill off underperforming bots as they continue to get the hang of the various platforms.
"I think right now we're in a huge experimentation phase," Altimeter's Etlinger says. "That's why I wouldn't be too turned off by early failures. It's just inevitable."