Customer Feedback and Machine Learning
I’m an Airbnb host and while reviewing my most recent Airbnb guest, James from Australia, I saw the section again that said “Anything you would like to tell Airbnb? It’ll be private…” and it got me thinking about what happens behind the scenes at any company that receives customer feedback.
For me, I’ve only used that section for extremes. The > 90 and the < 10. “BEST GUESTS EVER” and “He made me feel icky”. I mean why bother the great folks at Airbnb to say “They were great. I’d recommend them.” The stars are there for that.
In the time that I’ve been at indico (we do machine learning — where we teach computers to translate text and images like a person can), it turns out that people don’t care too much about the middle ground. Why don’t they care? Middle ground is the last group of people that can be swayed and the last group to target.
So why do you care about the extremes at your company? Say someone is desperately trying to ruin your reputation. Now I know it’s just a small sample group, but ew. Some people just can’t be happy. Unless you’re Spirit Airlines… and well…. with Spirit, the customer is actually always right.
I then got to wondering how Airbnb is perusing through all of our host/guest responses and then… well, how is Uber doing the same? I know for Uber that if I rate a driver anything under a 4 stars, I’ll get an email. So that’s based solely on human ratings. But Airbnb? Do you guys really read everything we write? So far I’ve only received an email about cases that I have opened or a guest has opened (one about the guest who broke my washing machine and the other about one that needed to leave early). What about the other data we’re giving them? To be honest, I never heard back about the guy I called “icky”.
indico can flag all of those messages received based on sentiment (yes! or oh hell no…) and we’re now in beta for emotional writing. Here are some scenarios we’re targeting:
“I felt uncomfortable” indico score: 1% positive
“Best guests ever” indico score: 99% positive
“They were dirty” indico score: 3% positive
“He made me feel icky” indico score: 4% positive.
Now what about a review? Check out my latest review run through the indico playground:
Now Airbnb has 2400 employees and a TON of reviews every single day. My thought is — is the “complaint line” STILL being manned by a fishbowl of a customer service team. Do companies even do that anymore? I remember walking into my first out of college job in Boston in 2002 and being told “To the right is our fishbowl… they handle technical difficulties.” It was a transparent sphere that reminded me of a PBS telethon minus the self-esteem.
Imagine ALSO being able to take your chat calls through LivePerson, Intercom, etc. and “world clouding” by sentiment customized exactly for your business. Lets say people keep telling Doug he’s not being helpful but, honestly. Who has the time to manually sift between ALL the conversations going on?
We can tell you if it’s time to fire Doug.
You have the data. indico has APIs that replaces a Data Science team.
Holla at me: julie@indico.io.
PS: We’re also multilingual.
