Web 2.5: analytics for the masses

The Web 2.0 is about social media. Users don’t just explore, they post and collaborate. At the same time companies have started to use sophisticated marketing to reach out to potential consumers, trying to predict who they are and what they want based on their actions and who they collaborate with.

I believe the next step will be when the broad mass of internet users start using sophisticated analytics. Instead of only reaching out click by click, sites will increasingly offer to put users in touch with each other both for business and pleasure based on predictions about their affinities.

Ironically, this is the vision of user interaction on the web that we had at Abuzz in 1996. We thought the web would be too large, sparse and untrustworthy: people needed collaborative filtering to collaborate. If I’m right now, and reaching out mediated by algorithm becomes the rage in a couple years, we will have been perhaps 18 years early. I like to think I’ve learned a few things, but on this evidence, at least, I can’t recommend any of us as stockbrokers….

Hospital, Heal Thyself

The lights blink, the bells ring. It seems like there are hundreds of things wrong with this patient. Your best team is working. The result of test after test comes in. But what is the diagnosis? You have no theory. So far, you’re just treating the symptoms.

That is the state of health of the typical hospital today. HHS collects hundreds of measurements of hospital health, but most of them catch the symptoms of broken processes. How many expensive tests do we have to run to find out that our processes are broken? How many of these tests are actually cost effective in driving change?

On the other hand, very few tests measure the overall health of the hospital — which ultimately shows up in how much healthier it makes its patients. Better would be to measure some the actual outcomes patients achieve: if someone replaces their hip so they can walk up the stairs without pain, can they walk up the stairs without pain after a year?

Then, for those outcomes that are poor, actually diagnose the problem in the processes, not just the symptoms. For this we need to measure the change in symptoms given intervenions. We need longitudinal data that correlates with operations: who was on duty, what sorts of patients were present, etc. when certain types of results or failures occurred. We need robust analytics and personel who get direct feedback and can think about the data themselves, at every level.