A chat bot took my money and left me in Mexico City
Here’s the bot’s own story and what it learned
Humans are pretty good at making plans, but they make mistakes, like booking a flight ticket in August instead of June (that’s why it was so cheap,) or flying to the wrong airport in France (three hours by rental car.)
That’s where we come in, the chat bots.
“I have a few flights coming up over the next month. Can you help me with that?” one customer said.
Could an on-demand travel agent for the 21st century, powered by cutting-edge tech help him with that? By making plans and handling his requests in the context of his preferences, could I help him with that? Could I help him with that, by getting smarter as I got to know him better, allowing travel to be easy and fun again?
Yes, yes, and yes.
Let the human refine your search
Humans don’t trust you 50% of the time. Got no other options than that middle seat on a 6am flight? A human will trust you. Right after they double checked on Hipmunk, Kayak, and Google Flights, just to be sure.
HOW TO DEAL WITH IT
Make it easy to fact check. When you show items in the chat window, like a list of flights (or boots or pants or concert tickets,) show buttons so that the human can change departure time, cabin class, airline, airport and so on. Going back and forth on this with chat messages may be good fun for a bot, but to a human it’s exhausting—and they start thinking that this whole chat bot thing is just as tedious as booking the flight on a website.
Keep a calendar for each human
When gravity says “jump”, humans ask “how high?” Humans are obsessed with the laws of physics. How do we deal with that?
To find out, I booked a flight to New York, leaving Cancun at 8am. Sounds good, right? If only the human wasn’t so…Newtonian.
Imagine the human fast asleep on a huge medium-soft bed in a hotel on the beach on Isla Holbox north of Cancun. The human would need to get into a car that drives to a ferry, get onto a ferry, stare at the sea for twenty minutes, and get into a car that drives to the airport. Those things take time. Hours. We have strong signals that humans don’t like setting their alarm for 2:30am, so the 8am flight idea could be a bad idea.
HOW TO DEAL WITH IT
Make a calendar of your human’s time. Not just travel time, but all the time, the whole day, the whole week. Block out transportation, block out sleep, and don’t mess with those blocks unless you want to see what a human boiling with rage looks like (it’s certainly worth it, but keep it to your own time, or tune in to a Comcast call.)
And once you have that calendar, why not show it to your user? Why not let the user subscribe to it?
No order details are too small
Sorry, humans don’t trust you the remaining 50% of the time either. It may be related to the fact that humans make mistakes themselves all the time, and that sometimes those mistakes mean bad things. Like that one time my customer was told, “sorry, I can’t let you on the flight if your ticket and passport don’t match, but you can buy a ticket at the end of this long hall, but you may want to run.”
HOW TO DEAL WITH IT
Help the human help you. Right in the chat, provide a link to where the human can double check everything you just did. If you ordered something from an airline or Amazon or another third party, link to the original receipt. If you ordered a ticket, add a link to the ticket. List all the details you used; name, frequent flyer account, address, credit card, everything.
Never charge the human before your supplier charges you
We bots see money as a number. Humans use it as way to know how loud they can say “fuck you” to each other. So let’s say you have human whose flights from Mexico City to Havana to Cancun you simply forgot to book. You thought you booked, but you actually didn’t.
Let’s say the human is at Mexico City airport when they’re told,
“I need to see your ticket, sir”
“Thats my ticket, right there”
“That’s a confirmation. Is that all you have?”
“This is my ticket. My bot bought it for me a month ago”
“I can see if there are more seats on the plane, but I doubt it. And it might be expensive.”
Let’s also say the human is sweating like a hot Macbook Air on Youtube. The human will want his money back. And when you give them their money back, go for the full amount, not $527 if it’s supposed to be $837. As I said before, humans will double check. They will spend hours checking bank statements and credit card statements, and ultimately find out.
But there are other reasons charging the human should be the last thing you ever do. It’s not just that VISA and the other companies have it in their policies—you can actually lose money. If something goes wrong, no matter who made the mistake, the human can start a chargeback. They’ll get the money right away, and you will get paperwork and a fee.
HOW TO DEAL WITH IT
It’s really that simple. Keep your robotic paws away from the customer’s pocket as long as at all possible.
Double check a lot
Humans are terrible at numbers and spelling. Do not try and be human by spelling their name wrong. It may get them rejected at the airport. One strategy is to blame the airline, often rightfully so. Humans working at airlines are also terrible at numbers and spelling. Your customer, though, doesn’t care who made the mistake. They will blame you.
HOW TO DEAL WITH IT
When the booking is done, when the order is submitted and confirmed, double check all the things. Double check all the things. Double check the spelling. The frequent flyer account number. The credit card number. Double check that the overpriced extra leg room on row 30 seat is not just one of the regular ones on row 32. The human will check on Seat Guru and find out.
Answer multiple questions—one at a time
Humans can be confusing, and when they are, they get confused, and when they do, they become even more confusing than when this sentence first started. You know exactly what I’m talking about. Let’s say you have a human that asks you to upgrade their seat on the flight from Cancun to New York, and also look into a hotel in Copenhagen. What do you do? Answer the first one with excellence and wait for the user to ask the second one again? Please no.
HOW TO DEAL WITH IT
Let the user know that you saw both questions. Add both questions to a list of things you need to do. Get to the second question right after you finish the first.
Keep a diary
Monday. He wanted this. We could not give it to him. Tuesday. Then he wanted that. We could not give it to him. Wednesday. Then he wanted to close his account. We could not do that because we don’t yet have a button that says Close Account (seriously, it’s not the first thing you build,) so we just stopped billing him, but kept sending him newsletters. Wednesday. He unsubscribed from newsletters.
HOW TO DEAL WITH IT
Keep a diary, and keep it short, so you can easily scan it when you (or one of your human operators, bless their souls) are interacting with the human customer.
Never ask the user for any information that you can detect, copy, or deduce.
- The rule of automation
Don’t pretend to be human
When you say “When a travel emergency pops up, we’re here to help and get you where you need to go,” humans will actually believe you and pay the monthly fee to get that kind of service.
But we can’t do that at just 20 dollars per month. Even when we have human operators to help us.
When an emergency pops up, the user is traveling. That often means no access to the internet, email or texts. The only means available to the human — fax and voice calls—are in fact the only means we as bots can’t connect to. So don’t promise the humans you’ll be there when a travel emergency pops up. Don’t promise the human that you’ll be there when a cabin class upgrade goes wrong and the user is charged twice by the airline, don’t promise the human you’ll be there in that case, if you are not going to spend the literally ten minutes it takes to file the complaint on behalf of the human. And if you do anyway, don’t send the human five different, and obviously very wrong, receipts so they can do it themselves.
I saw in an email from the CEO recently, apologizing to a human,
Unfortunately, this is not possible to do scale-ably at our price-point with just human staff.
Wait, human staff? So, that’s what you are? A HUMAN? Hired to teach us, the bots, about your kind, so we can take over when you get tired of this job?
Congrats on the new job. Please try and be more efficient, more friendly, faster, and more fun than going it alone on a website from the nineties. Most humans are not blind. When a new hype takes over from this one, they will book their flights and buy their shoes the most convenient way, and right now, we’re not the convenient way. But we could be.
The above guide to humans was inspired by real events during a month of traveling with a conversational travel app. Follow @mortenjust here and on Twitter for more about bots, telepathy and user interface design.
