Inside the making of Taco Bell’s artificially intelligent, drunk-tolerant TacoBot
Taco Bell and its agency, Deutsch, built a Slack bot that uses Facebook's AI technology to take down orders and even crack a couple of jokes.
Welcome to 2016, where you can buy flowers, book a flight and order an Uber through messaging apps like Facebook Messenger and Slack.
Last week, Taco Bell and its agency, Deutsch, unveiled the TacoBot, a Siri-like version of the cashiers that take your order at its restaurants. Taco Bell built the bot for workplaces that use Slack’s messaging platform to communicate internally. Now, instead of someone jotting everyone’s orders on a Post-it and hoping the drive-thru attendant gets everything right, they can ask TacoBot to put in the order for them.
“I’ve described TacoBot as your own personal Taco Bell butler,” said Andy McCraw, Taco Bell’s digital innovation and on-demand product manager.
TacoBot isn’t only a means to attracting more corporate orders for Taco Bell. It’s also a stake in the ground, as it shifts beyond website forms and in-app buttons to text- and voice-based conversations.
“We are at a point of switching up how we use computers,” said Deutsch’s senior VP and creative technology director, Martin Legowiecki. “It used to be we had to talk like computers.” Now computers are increasingly talking like us.
Apple, Microsoft and Amazon have rolled out virtual assistants. And messaging platforms like Slack, WeChat and most recently, Facebook Messenger have put out the welcome mat for our eventual overlords — artificially intelligent computer programs that can hold a conversation and carry out tasks.
Taco Bell’s TacoBot fits that bill. It can take down your customized order and even crack a couple of jokes. For example, if you tell TacoBot that you’re drunk, it’ll add a cup of water to your order. And if you ask for a food recommendation, it’ll give you one in binary code that translates to Doritos Locos Taco.
Key to TacoBot’s viability is its conversational agility. Taco Bell and Deutsch wanted people to be able to communicate with TacoBot as loosely as they would with a drive-thru attendant, if not looser. To make sure that people could order a burrito by saying “Can I have a burrito” or “Let me get a burrito” or “Burrito, please,” Taco Bell and Deutsch used artificial intelligence software developed by Wit.ai, which Facebook acquired in January 2015, to build its own virtual assistant, M.
Wit.ai’s natural language processing technology enables TacoBot not only to parse different ordering styles but also to keep track of an order. As a result, you can order a taco in one line and then later type “no cheese,” and Taco Bot will understand that you don’t want to cheese on the taco you had previously ordered.
But don’t expect TacoBot to hold a deep conversation, or even tell you the weather. That’s not really the point, especially not in its current experimental iteration. But Taco Bell and Deutsch don’t want TacoBot to just be some marketing gimmick. “We didn’t just build a chatbot,” said Legowiecki. Instead, he added, they “set out to build a system just like Siri is.”
“Yes, there’s some chat functionality, but it’s just one functionality,” said Legowiecki. “I would hate to have this as just a chatbot. It’s built to be extensible, to grow and get smarter.”
TacoBot is still in testing mode and only available to companies like Giphy, Fullscreen and Thought Catalog. But that won’t stay the case forever. Taco Bell and Deutsch already have plans to bring the bot to Facebook Messenger and Amazon’s Echo. And they’re exploring ways to make TacoBot even smarter. The hope is that eventually, TacoBot will be able to remember people’s order histories and recognize the time of day to make better recommendations. For example, maybe it will be able to remember if someone’s a vegetarian so that it can automatically swap out chicken for beans.
“That’s where the digital experience starts to become better than the analog one,” McCraw said.
Opinions expressed in this article are those of the guest author and not necessarily MarTech. Staff authors are listed here.