I love The Hipmunk. And who wouldn’t? Their agony score, ranking of flight by level of inconvenience, have saved me from many crazy red-eyes and long layovers — I was always a fan.
So when I heard of the Hipmunk bot: i) one of the first bots on Facebook Messenger; ii) who can both tell a joke and book a flight, I swooned all over again — I’ve got to try it out.
Unfortunately, this experience was a little bit rough.
My mistake? Choose a difficult destination. I blamed this on my friend, N. — she wanted to go to New Orleans out of all places!
Hipmunk was not a fan of “New Orleans.” It immediately thought I meant “Orleans, MA” and didn’t allow for any correction.
I desperately tried to persuade it otherwise.
- Adding the state name (“Louisiana”)
- Protesting with “No” and “Stop”
No avail. Hipmunk was already too busy finding a flight, and unfortunately there was none. Our conversation went into a loop (full script here).
01:22 Tam Le Flights from San Francisco to New Orleans, early September
01:22 Hipmunk You’re looking for the best time to go from San Francisco, CA to Orleans, MA between September 01 and September 30.
Hold on, I’m digging through the data now. 🙂 Hang tight… it’ll take just a couple more minutes
Whoa! This has got to be a first! I can’t find any results that match your travel plans. Try changing your locations or dates and asking me something like “flights from San Francisco to Miami, December 5–10”.
01:23 Tam Le Ah no flights from San Francisco to New Orleans, Louisiana
01:23 Hipmunk You’re looking for the best time to go from San Francisco, CA to Orleans, MA between September 01 and September 30.
Hold on, I’m digging through the data now. 🙂
Whoa! This has got to be a first! I can’t find any results that match your travel plans. Try changing your locations or dates and asking me something like “flights from San Francisco to Miami, December 5–10”. Information Not Found
I don’t love Hipmunk less. It is perfect for unambiguous destinations (“San Francisco” for example). Hipmunk also does a great job at understanding dates (I only said September, it searched for Sep 1–30), and remembering context (I didn’t have to type the time for the 2nd time).
But the Hipmunk experience reinforces of three rules that I often use in bot design, the “false-safe” options to prevent your bot from an epic fail in dicier situations:
Rule No 1: Exit option: Remember the monotone voice greeting you customer support center? The voice tells you something like “Press 0 to go back to the main menu,” or “Press # to talk to a customer representative.” Your bot also needs one of these, in case it is not reading the input correctly, or when the customer wants to go back to “Reset.” Provide them with an option to “Pause” or “Exit.”
Rule No 2: User-input Word Disambiguation: I get it, “New Orleans” is a confusing word. Because I love exotic locations, I also tried “Cuba,” and got “Cuba, Portugal” instead of our neighboring country, Cuba. Of course, my ignorance again: there are so many cities in Cuba, one would not go search for flight to Canada, because that is just too broad.
Is there a way to design for disambiguation? Floating prompt as the user types? Or perhaps validation with buttons when some word is ambiguous? Instead of going forward, the bot should plan for these cases and asks for confirmation. It can even offer a few interpretations that the user can click on.
Rule No 3: Frustration Detection: If the entry from the user doesn’t change, it might be an indicator of a confused or frustrated customer. Or, if you detect swear words, it might be a good idea to stop and check with the user, too. This is easy and programmable, but will save your user from the dreaded “F*!! I’m never going to use your bot again!”
In short, plan for the worst cases where your bot fail and make the experience a little less painful for your users. They will love you for it.
Hipmunk in a Loop and the Three Rules of ChatBot Design was originally published in Chatbots Magazine on Medium, where people are continuing the conversation by highlighting and responding to this story.
Source: Chatbots Magazine