Life insurance: the next frontier for social media analysis?

For the record, I’m a reasonably good health insurance risk. I eat lots of vegetables, I don’t smoke and haven’t been in an ambulance for at least 18 months. I’ve even avoided the obligatory Christmas cold.

Question is, who’s listening? According to the Wall Street Journal, British insurance giant Aviva are.

“Insurers have long used blood and urine tests to assess people’s health – a costly process.Today, however, data-gathering companies have such extensive files on most U.S. consumers – online shopping details, catalog purchases, magazine subscriptions, leisure activities and information from social-networking sites -that some insurers are exploring whether data can reveal nearly as much about a person as a lab analysis of their bodily fluids.”

Interesting – and a potential chance for people to game the system. If I was seriously ill, would I mention it in my Twitter feed? As fellow ‘head Thomas pointed out on his blog, the observer effect is ever-present in online word-of-mouth. Equally though, if I was setting out to deceive a life insurance company, it wouldn’t take much to be economical with the truth on a health questionnaire.

So what are Aviva testing out with people’s social data?

Firstly, they assume that their current medical-based model is correct. This may be their first error as described by Kaiser Fung. Then they map all of the marketing variables against these medical records and derive correlations. Just like a credit-rating agency, they apply these known correlations to decide how to classify new applicants.

How does this approach differ from best practise in social media analysis for brands looking to engage with their consumers?

Let’s compare two scenarios. Based on the data:

1) Who should insurance company X reject for a health insurance policy?

2) 1000heads are inviting 100 people to take part in a trial programme. Who should we invite?

For question 1, let’s say that for every 1000 applicants insurance company X identify 3 bad insurance risks and falsely identify 7 more as bad risks.

Why might insurance company X be prepared to accept these falsely assigned bad risks? It’s all a question of economics – if the overall risk of the group of insured people is reduced by enough, then the falsely rejected are negligible. After all, there are no perfect methods for these things if we don’t want our insurance to cost a lot.

But for scenario 2, we are looking for a small number of people to include in a small activity, rather than a small number to people to exclude from a large activity.

This means that any systematic errors completely overwhelm the overall picture. If we engage people, we aim for 100% of them to respond positively. Clearly if 7 out of 10 of them have been falsely identified as having the right interests, we will have a problem. On the other hand, it may well be easier for a community exec to choose people objectively from such a shortlist than by avoiding robots entirely.

What do you think about the ethical implications of what Aviva are doing? Let us know in the comments below.

Brought into the world on Friday, December 10th, 2010 and filed under Measurement, Stats & Research.