/* ---- Google Analytics Code Below */

Tuesday, February 17, 2009

Thinking About Metering

About a week ago the NYTimes had an article on Google's Power Meter Project. Jim Stogdill has a good followup article on O'Reilly Radar where he looks into more of the underlying details. He makes the point that setting up a metering system can then make you think differently about utilizing resources. He also points out some of the problems with having a Google-centric design. A good detailed piece.

When I first saw the NYT article I started thinking about other things that could be metered and what kinds of models could be built on top of the data that would be acquired. Inspired, in part, by such apparently failed projects like PRISM. In retail, capture of a number of basic metrics and the inclusion of models makes sense. It has been done, but capturing the rate of sales and predicting an out-of-stock event is a good example. This is less about meters, but more about combining the input of multiple sensors. Also called sensor fusion. We can install new sensors, which are cheaper every day, but it would best to first harvest sensors in place today.

Power, Point-of-Sale, Employee allocation, Sensors at key store points, etc. and temporary and disposable sensors that track promotions in specific store contexts. Then add a floor grid to track traffic. While some of these examples are simple, there are opportunities to link data, visualize the results, and produce real feedback from the retailer.

You have to admire Google for taking the first steps in seeing how existing sensors could be leveraged to construct a value proposition for the home. They will get value by starting to build models and experiment with the resulting data. The store is a yet richer environment, with more data being collected today, and sensors becoming cheaper.

The inclusion of business intelligence methods and specifically visualization capabilities that can combine the results from multiple sensors also makes sense. Models, starting with very simple ones, can blend data to feed other metrics. Then we can start answering questions like: Should we change the product mix in this aisle on this day of week? Do we need labor coverage in this aisle at this time? And many more. Yes, I know there are large revenue management models out there that already try to do this. But they are not sensor centric. Much opportunity pointed to here. Does anyone want to play with me on this?

No comments: