Some conversations are easy to plan for. Take the cup of coffee you ordered this morning for example. Chances are it started with a reciprocal greeting and moved onto a request. Maybe there was a short back-and-forth hammering out small particulars (what size, do you need room for cream), followed by payment, and a quick exchange of thank yous and pleasantries.
Chatbots do especially well with normal interactions like that. But, what’s important to note here is that just because thousands of conversations might be about the same thing generally, the details, phrasing, and particulars will essentially never be identical. You might ask for the “biggest coffee you’ve got” while someone else will say “24 ounce” while the next guy in line will just say “extra large.” You might all want the same humongous coffee, but you’ve got your own way of asking for it.
For more nuanced conversations, the variations are manifold. So while you might know exactly the sorts of issues you need a chatbot to handle, the variations of those interactions are incredibly vast. How do you train a bot to handle discrete issues you can predict but language that you simply can’t?
Enter Yalo. Yalo is an innovative company who leverages partnerships with everyone from Facebook and Whatsapp to AWS and CrowdFlower to help companies build 1-on-1 personal relationships with their clients. They partnered with Aeromexico to build the first airline chatbot in the Americas to help Aeromexico better handle the massive amount of traffic and engagement they get on their Facebook page. As we mentioned, two of the central premises of chatbots are that they interact with customers where they already are (Facebook, in this instance) and they’re available 24/7. And when you’re dealing with a business that never sleeps like air travel, having accessible, accurate, always-available support goes a long way.
Now, back to that point about nuance and variance: air travel is not a simple issue. Just booking a flight is a lot more complex than ordering a cup of coffee and, when you factor in how much more expensive and important air travel is, getting it right is incredibly important. There are thousands of variations here that require a bot to understand context, what airports are where, and what layovers are unreasonable.
And that’s just for bookings. Rescheduling, frequent flyer rewards, travel documentation, promotions, upgrades, amenities, seat changes, you name it: the list is long. Training a bot to handle all of this takes significant effort but doing it well can be a massive benefit for both the company and their customers. The company can farm out more subtle, difficult conversations to live reps who can best handle those requests while keeping their reps from spending time on rote requests a bot can handle in less time for less money. Customers, meanwhile, get faster service whenever they need it.
Yalo tapped us here at CrowdFlower to help train this bot. They’re using our platform to pair customer questions (sometimes called “utterances”) with both broad categories and smart, distinct answers. These judgments are used to train and tune a chatbot so it can provide instant, intelligent replies to some of the most pressing issues travelers have.
Bots will change service organizations in the same way chat did, allowing for more efficient solutions to real world problems and consumers should expect to see more and more of these kind of implementations in the future. And the smart folks at Yalo understand both the legwork it takes to create a smart chatbot and the value in doing so, all while injecting those bots with the right sort of attitude and flair.
We’ll be talking a lot about bots in the next couple weeks, complete with a eBook about what we’ve learned about training, tuning, and testing them. So if you’re interested, stay tuned and visit this space in the coming month. The most important bit?