I pay Verizon about $1,500 a year for family cell-phone service. If
you figure that one-third of that goes into subsidizing the phones they
sell us, it’s about $1,000. And if I switch to another carrier, that
money comes right off their bottom line. For phone companies, churn is a scourge; they use their unparalleled troves of data
to search for customers likely to churn. They they use more data
analysis to see which types of incentives are most likely to keep such people in the fold.
How can they spot a churner? They look for changes in behavior. They
might take a look at me, for example, and see that I’m no longer
commuting obediately in and out of Manhattan. My pattern has changed.
Instead of freezing at the bus stop, peering down Valley Road for a
sign of the 66 bus, I’m at this moment sitting in the living room
wearing a prototypical blogger’s bathrobe. My cell phone is upstairs,
turned off. Verizon’s computers may well have placed me into a new
tribe. And since it includes many of the newly unemployed, I wouldn’t
be surprised if my churn risk score has jumped way up. (I should give
them a call to see if they’ll give me a discount.)
Yesterday I had lunch in New York with Tony Jebara, the head of machine
learning in Columbia’s computer science dept. He’s also the chief
scientist at Sense Networks, a start-up (I wrote about) that studies
(anonymous) user behavior for cell-phone carriers. He told me that in
one study of 8 millions users (he wouldn’t name the company), Sense
focused on subscribers who use pre-paid cards. They tend to have less
money, and since they’re not locked in by subscriptions, they churn
much more. Studying the behavioral data, Sense saw that one subgroup,
people who frequent used car lots, churned at eight times the average
rate.
Why would that be? He had lots of ideas, but none of them backed by
research. You’d think that people with pre-paid cell phones who are
looking for used cars might be short on funds. But who buys or sells a
car and gives up a cell phone? I don’t know. It would help to look at
the historical data, to see whether this pattern picked up with the
economy went south. It would be interesting to look at the geography,
the neighborhoods these people appear to live in. All kinds of
variables. The possibilities for analysis are endless.
But for the cell phone customers, the simple correlation might be enough. Sense, like other data-mining start-ups, is under increasing pressure to shift the focus from really cool stuff (that attracts the attention of people like me) to studies that pay the bills. In June, Intel’s venture arm invested $6 million into Sense. Since then, they’ve pushed out Sense’s founder and CEO, Greg Skibiski, and are said to be looking to replace him with a telecom industry insider. I’m sorry to see Greg go. He introduced me to this whole fascinating arena. I look forward to seeing where he pops up next.
