2 ms·

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

by gacba 16y ago

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. The tricky part is defining what "events" you want to detect, then getting a set of data with those events for training on the network.

patrickas 16y ago
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!
- techbio 16y ago
  What data do you have? If a predictive system performs with that data, QED.
- matwood 16y ago
  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.

by gacba 16y ago

patrickas 16y ago

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!

techbio 16y ago
What data do you have? If a predictive system performs with that data, QED.
matwood 16y ago
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.