Relationship of Half-life to the Experience Curve^©

Arthur M. Schneiderman

I'm often asked how the half-life method differs from the experience curve popularized in the 1960's by Bruce Henderson, founder of The Boston Consulting Group. Henderson promoted, under a new name, the Learning Curve that was discovered by T. P. Wright and published by him in 1936. The Learning Curve, developed in the aircraft industry, was based on the observation that unit direct labor usage, expressed in total man-months, declined with increasing experience. Henderson noted that the same was true for unit cost. Based on this observation, he went on to develop a strategy model with the experience curve at its foundation.

The experience curve, like the half-life, is also an empirical observation. It states that for each doubling of cumulative experience (total units produced from the very beginning, not just this year), real unit cost drops by a constant percent, for example 20%. If your first million units cost $10 each, then your next million units should cost $8 each, the next two million units, $6.40, the next four million units, $5.12, etc. Because cost is driven by cumulative units produced (1+1+2+4 million in our example), the rate of decline of cost drops over time unless unit volume grows at a sufficiently high exponential rate.

The half-live method, on the other hand, predicts that the rate of decline of defect level is constant over time. Why the difference? We can't say for sure, since the experience curve is a purely empirical observation and is not based on any underlying theory. We can list the things that probably effect the slope of the experience curve, but we can't write an equation in which they are the independent variables. On the other hand, there is a theoretical basis for the half-life model.

The pareto chart is a graphical tool (bar chart) for displaying the rank ordered causes for a particular defect. With very rare exceptions, these charts follow the 80/20 rule: 20% of the causes account for 80% of the defects. This leads to exponentially declining pareto diagrams. In fact, the biggest cause, that is the "root cause," usually accounts for 20% to 40% of the total defects. Let's take 30% as a typical number. For a process of average complexity, an experienced improvement team takes about four months for each cycle of improvement. 30% improvement in four months corresponds to a half life of 8 months, consistent with the half-life/complexity matrix. For less complex processes, the root cause is often larger and the improvement cycle time shorter, accounting for the shorter half-lives. For more complex processes, there tend to be many more causes so that the root cause is smaller while the improvement cycle time is longer.

Note that I said an "experienced" improvement team. Proficiency in using the improvement process takes several improvement cycles, so that initially, half-lives can be significantly longer. For this reason, I encourage organizations to attack less complex processes at the beginning as the organization masters their new improvement paradigm.

Also keep in mind that the half-life deals with defects, not cost. Of course defects, defined as any gap between current and potential performance, are the principal driver of unit costs, so the two are clearly related. Twenty years ago, experience curve practitioners were baffled by the unexplainable leapfrogging done by Japanese industry. How could they possibly achieve experience curve slopes two or more times steeper than their Western counterparts? Initial reactions was that they were using predatory pricing, selling below cost. The furor even made its way to the US Congress. But careful study, the object of my first visit to Japan, showed that their hyper-fast cost reduction was real and the result of a new approach to process improvement.

The experience curve became the very basis of competitive strategy after its discovery, leading to a focus on the positive feedback loop between market share, cumulative experience and unit cost. I believe that the half-life model can provide a similar strategic nucleus (parental pride acknowledged!).

Last modified: August 13, 2006

Relationship of Half-life to the Experience Curve©

Relationship of Half-life to the Experience Curve^©