Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Scott Adams: FaceBook Killer

dilbert.com

Post reply on HN
Print view View original

31–40 of 56 posts

Re: Scott Adams: FaceBook Killer

#31

Post by patrickas »

gacba wrote: post #27
icefox wrote: post #24

"Likewise, your credit card company and the phone company have records of what you did, as opposed to what you plan to do next." Sorry to bust your bubble, but your CC company has already given talks about how it knows when you are about to move, when you plan on getting a divorce, when you are trying to have a baby, when you got actually going to have a baby and plenty of other personal thing... I can only imagine h…

I don't work for a CC company, but I did work with predictive analytics regarding churn in mobile phone and cable companies a decade ago. It's not hard to back-mine the data to establish certain patterns that happen before major events. That's how we predicted churn at a major provider (rhymes with "Blecks-dell") with 90% accuracy. There's no reason that same logic could apply to a rich data set like CC transactions.…

That sounds interesting! I can't think of any types of things to check for to predict a user is about to churn! Care to give more examples?

The closest I could think of that might work is a neural network that would be fed data for people who churned and people who didn't but even then I cannot think of useful types of data to be fed!

Re: Scott Adams: FaceBook Killer

#32

Post by techbio »

icefox wrote: post #24

"Likewise, your credit card company and the phone company have records of what you did, as opposed to what you plan to do next." Sorry to bust your bubble, but your CC company has already given talks about how it knows when you are about to move, when you plan on getting a divorce, when you are trying to have a baby, when you got actually going to have a baby and plenty of other personal thing... I can only imagine h…

As appealing as the opportunities appear, my naive guess is that "neat" is superseded by "regulated" but this makes me wonder... while finances are about as graphically personal as information can be, the optimization potential for billions of people, each with hundreds of choices, is astounding.

How could this computation be performed locally, in a directly parallel device, and in a way that protects the consumer? How about smart (real: assistive AI) cards as a disruptive technology?

Re: Scott Adams: FaceBook Killer

#33

Post by techbio »

patrickas wrote: post #31
gacba wrote: post #27

Earlier quoted context omitted.

I don't work for a CC company, but I did work with predictive analytics regarding churn in mobile phone and cable companies a decade ago. It's not hard to back-mine the data to establish certain patterns that happen before major events. That's how we predicted churn at a major provider (rhymes with "Blecks-dell") with 90% accuracy. There's no reason that same logic could apply to a rich data set like CC transactions.…

That sounds interesting! I can't think of any types of things to check for to predict a user is about to churn! Care to give more examples? The closest I could think of that might work is a neural network that would be fed data for people who churned and people who didn't but even then I cannot think of useful types of data to be fed!

What data do you have? If a predictive system performs with that data, QED.

Re: Scott Adams: FaceBook Killer

#34

Post by techbio »

tel wrote: post #29
GavinB wrote: post #11

This is what Google does. What you search is becoming a better and better predictor of what you will do. Most searches are based on future plans, purchase, or interests. They even know your current and future locations if you use google maps for directions. If you use gmail, Google can tell that you just broke up with your girlfriend, or are graduating college. Sidenote: we don't "give away" this information. We exch…

Ironically, the "French Crush" commercial directly advertised this transaction. Play cute music against a cute story and highlighting the fact that you can be very accurately understood and predicted based on Google searches becomes heartwarming.

Oui--heartwarming, and not a post-apocalyptic nightmare scenario. I'd like a low-powered computing device that (a) observes my habits and emulates their neighboring variations in a sandbox and (b) I can unplug if I choose, so as to carry the quantitative overload of life, but help ONLY ME, recurse for all others.

Re: Scott Adams: FaceBook Killer

#35

Post by m_myers »

portman wrote: post #5

Is there a reason why he intentionally mis-capitalizes "Facebook" to "FaceBook"? Scott Adams strikes me as a careful and fastidious writer; I'm wondering if there is an inside joke that I'm not getting.

From the comments:

  Why for heaven's sake are you writing "FaceBook" instead of "Facebook." I will assume you are not an avid user of this so-called "Book."

  [Ha! Busted. Indeed it is not practical for me to be on Facebook. I have corrected the spelling. -- Scott]

Re: Scott Adams: FaceBook Killer

#36

Post by janj »

I'm trying to do a very specific subset of this within my app. The plan is ask users when they would like to take a cruise, where they're leaving from, about how much to spend, how to contact etc. Store this in a database and give cruise vendors access to this information where they can send out what they have available that matches the users criteria through the app.

Obviously this is much less general than what Adams is describing but makes more sense to me. I've never liked all the spammy aspects of social technology and hope I can be profitable sending info only when the user requests it. Of course this is just one of the many ways I'll be trying to monetize my app.

Re: Scott Adams: FaceBook Killer

#37

Post by Judson »

GavinB wrote: post #11

This is what Google does. What you search is becoming a better and better predictor of what you will do. Most searches are based on future plans, purchase, or interests. They even know your current and future locations if you use google maps for directions. If you use gmail, Google can tell that you just broke up with your girlfriend, or are graduating college. Sidenote: we don't "give away" this information. We exch…

It is almost freaky when I go to type something I thought was obscure into the google search bar and it predicts what I am going to type after the first word.

A less obscure example (but the only one I could think of) is when I type is "How to fix " and it correctly filled in "a leaky faucet". No context, but it guessed correctly.

Re: Scott Adams: FaceBook Killer

#38

Post by matwood »

patrickas wrote: post #31
gacba wrote: post #27

Earlier quoted context omitted.

I don't work for a CC company, but I did work with predictive analytics regarding churn in mobile phone and cable companies a decade ago. It's not hard to back-mine the data to establish certain patterns that happen before major events. That's how we predicted churn at a major provider (rhymes with "Blecks-dell") with 90% accuracy. There's no reason that same logic could apply to a rich data set like CC transactions.…

That sounds interesting! I can't think of any types of things to check for to predict a user is about to churn! Care to give more examples? The closest I could think of that might work is a neural network that would be fed data for people who churned and people who didn't but even then I cannot think of useful types of data to be fed!

One of the easiest today is using universal default. If you happen to pay your water bill late then that tells me that you're having money problems so I can expect other bills to start to falter. Some CC companies will actually raise their rates on you proactively assuming you're about to default. I think that's crap, but it's part of having everything so connected.

Other big ones that we found are simply changes in behavior. The person who pays their bill late every month is fine, and if they pay late enough to get the late charge even beneficial. The person who pays on time every month, but is suddenly late, big red flags. A person who suddenly starts using their CC card for groceries, but never did before is another example.

People also tend to live near similar people. If you're a company providing a service to many people you can start to build models of neighborhoods. If both my neighbors foreclose on their house there is a higher likelihood that I'm also going to be running into money problems.

Re: Scott Adams: FaceBook Killer

#39

Post by Tycho »

I think he's got the right idea: most web surfing right now is a waste of time, the wave of the future will be getting practical benefits from the internet. (I kniw online shopping and email and stuff is really useful in theory, but I just mean most of the time normal people actually spend online is frittered away on Facebook etc)

Re: Scott Adams: FaceBook Killer

#40

Post by slowpoison »

GavinB wrote: post #11

This is what Google does. What you search is becoming a better and better predictor of what you will do. Most searches are based on future plans, purchase, or interests. They even know your current and future locations if you use google maps for directions. If you use gmail, Google can tell that you just broke up with your girlfriend, or are graduating college. Sidenote: we don't "give away" this information. We exch…

That's quite perceptive of you. I hope more people look at Google that way. I especially hope Google management looks at it that way. I think this perspective always gets lost on people both inside and outside Google when they talk about Facebook overtaking Google, or when they fret over Google not making big in the social networking space.
Post reply on HN