A cashier is better than self-checkout machine because, unlike the machine, “she isn't on hair-trigger alert for any sign that I might be trying to steal toilet paper.” That is but one of “Humans 1, Robots 0” in the Wall Street Journal – and from my experience it’s true. I still use the self-checkout fairly often, but not because it is easier or better, but because the line to use a real person is usually longer and I don’t want to wait.
The most compelling argument in the article (although the one about the toilet paper is pretty strong) is the reference to “Dancing With Robots”, and the two rules set down by Frank Levy and Richard Murnane to determine the value of automation. They are an “Information Condition – All information necessary to carry out the task can be identified and acquired in a form the computers can process” and a “Processing Condition – The information processing itself can be expressed in rules”.
This view of information technology is simple, but powerful stuff. Think about it in terms of the fundamental weakness of MRP/ERP and, conversely, the power of demand pull or kanban methods.
At the most fundamental level we don’t know the single most important “information necessary to carry out the task” – what customers want us to make - until it actually happens. So we plug in a forecast of dubious accuracy. It goes far beyond that, however. We don’t know how long it will really take for materials and sub-assemblies to flow through various steps and from one pace in the factory to another – so we plug in averages and standard lead times. We don’t know how well people and machines will perform, how long they will actually take to cycle or whether the output will be any good or not, so we plug in average cycle times and yield rates. In fact, there are so many moving parts in a factory we actually know next to nothing about its reality so the whole thing is modeled based on a bunch of gut feels, historical data and approximations.
Of course, none of that gets into the wonderful world of lot sizing and that academic’s dream – the EOQ formula. Somewhere in the middle of the math we get to ‘Order Cost’ which is a purely fictional concept and a made-up number necessary for the math to work. It falls victim to that most basic principle of math: X * BS = BS (no matter how accurate X might be and no matter how bad we want to know an accurate result).
Then the system gets into its logical rules. For MRP/ERP to function everything has to be pretty normal so it assumes ‘rules’ such as prioritizing by due date, and it assumes everything scheduled will actually be made when it is scheduled – in short, it believes its own ability to model things. But that rarely (never?) happens, so the reality is lots of expediting. The reality also includes quite a bit of information reconciliation and data clean-up since the reality is nothing like all the data that was fed into the system.
If we knew everything the customers wanted well in advance, if we knew what yields would be before we made things, if we knew beforehand who would call in sick and when suppliers would be late, if we knew when machines were going to break down before they did so, if, if, if …. manufacturing would be so easy and we could all put our feet up and let computers run the factory. As the old quarterback Don Meredith used to say, if if’s and but’s were candy and nuts what a merry Christmas it would be. Unfortunately we have to shut the computers off and get back to reality.