Don't trust the data

A nice post from the always interesting Venkatesh Rao on Big Data, and how it will eat us. (A small and geeky aside: reading Rao is seeing someone who has always lead with strong Ti suddenly discover Ni. His lack of fatigue remains extraordinary to me). You should read the post, because it's interesting and fun, and note this:

Data is also a world begging for tight vertical integration of the supply chain from raw, unrefined wild data all the way to AI programs whispering insights into CEO ears. If you are a Republican, the holy grail vision is a dashboard-driven company that will allow a CEO to run it as if he/she were driving a car, with no additional human involvement above minimum-wage levels. If you are a Democrat, the competing vision is the similarly empowered citizen (at the moment, the Republican vision is winning). I expect a keynote at the next Strata titled “Big Data and Little People” (hashtag #OccupyData).

Well, maybe. In my experience though, data without context is meaningless and can often lead one astray. I once designed a massive call center where, through Erlang C modeling, we thought we could run a very efficient operation using very high occupancy rates for the agents. In practise, the load, response time, and queuing looked nothing like our models. Calls weren't being answered, people were on hold forever, we had trunking problems, etc. The solution came from spending time on the floor. Even though an Erlang C model says you will have 99% occupancy, the truth is that humans need a break between calls before they are ready psychologically to task switch to the next one. In practice, you get 90% occupancy at the agent level, max. Because we modeled higher occupancy, we understaffed, so wait time increased, which caused abandonments to increase, which caused in-period re-dials to increase, which increased load, which trunked, which increase (apparant) load more etc. The solution was to acknowledge reality and put a 90% cap on occupancy. We staffed a little more, but suddenly everything started to behave just as the models predicted. I don't think any quantity of Big Data whispering into a CEO's ears would have produced this insight.