Slashdot Mirror
Insurance Startup Uses Behavioral Science To Keep Customers Honest (fastcompany.com)
tedlistens quotes a report from Fast Company: Insurance startup Lemonade won itself headlines in January with the boast that it had successfully approved a claim in just three seconds. In that time, Lemonade's software had run 18 anti-fraud algorithms and sent a payment to the lucky customer's bank account -- a process that would have taken a traditional property and casualty insurer days, if not weeks. But it's what happened before Lemonade's artificial intelligence kicked into gear that makes the renegade insurer so potentially disruptive to this trillion-dollar industry, for which premiums alone comprise 7% of U.S. GDP. The customer, Brooklyn educator Brandon Pham, opened Lemonade's mobile app, signed an "honesty pledge" to attest to the truth of his claim, and then recorded a short video explaining that his Canada Goose parka, worth nearly $1,000, had been stolen. That deceptively simple claims process is the byproduct of academic research on psychology and behavioral economics conducted by Dan Arielyblog, one of the field's most prominent voices and Lemonade's chief behavioral officer. "There's a lot of science about when people behave and misbehave that has not been put to use," says Lemonade cofounder and CEO Daniel Schreiber. Lemonade is even applying behavioral science to itself, publishing unusually transparent blog posts that include data on customer growth, bank account balances, and more.
I attended a presentations in the mid 90's sometime by Dr. Hecht-Neilson who had a company that evaluated people for their credit worthiness using neural networks.
That's like saying that Amazon is making money by providing customers with an online shopping cart.
Algorithms are no longer a barrier to entry in analytics; you can get them for free from various Apache projects (Spark, Mahout, etc). The challenge is in acquiring the right data sets and finding features that deliver the kind of indicator you need by constantly evaluating samples and tuning your model. Everyone and their neighbor is using neural nets these days; most fail at achieving something meaningful with them.
lucm, indeed.