Cost of Delay, NPV and Taguchi

Cost of Delay is a very different animal to many of the concepts that people try to shoe-horn it into. Sometimes those are useful analogies. Other times, it’s not helpful, and leads to confusion, missing the actual power of Cost of Delay. A

Is Cost of Delay like a really high NPV discount rate?

For instance, Cost of Delay is not like the discount rate applied in a typical Net Present Value analysis. I’ve written about this previously. In short, discounting is rarely worth more than a rounding error on the Cost of Delay. If you need to apply discounting to your software investment decisions you’re doing it all wrong. It’s also not a similar concept, so using it as a launch pad to understanding Cost of Delay would be a category error. \

Cost of Delay is not like NPV discounting.

Is Cost of Delay like a Taguchi Loss Curve?

Another case of miscategorising Cost of Delay is thinking about Cost of Delay as if it were like a Taguchi Loss Function. Taguchi’s work has been a useful development in Manufacturing, closely aligned with Deming’s work and the Lean Manufacturing world.

The primary contribution of Taguchi when compared to a traditional analysis of loss due to variance in manufacturing is that the impact of variance is not a step function (i.e. good vs bad) but that the losses accelerate along a curve as you deviate from the ideal. Taguchi also argued that quality should be considered more broadly. That we should include the loss to society from poor quality, and that those losses eventually find their way back to the organisation responsible.

The payoff function for manufacturing pre and post Taguchi looks like these two:

taguchi loss function specification tolerance limits satisfaction
Manufacturing Payoff Functions: Minimise variability, FTW!

The problem with applying Taguchi in a Product Development as opposed to Manufacturing context is that variance plays a very different role in Product Development when compared got manufacturing. We are farming positive “Black Swans”, so eliminating variance is actually bad. Yes, we want to limit our downside (which is why the focus on batch size is so important). But we actually want to expose ourselves to the positive potential returns that are unknown and unknowable in advance. This produces a very different payoff function to the Taguchi Loss Functions that you see above: an asymmetric one. More about Asymmetric Payoff Functions here, if you are interested.

Product Owner Training. Leading Product Development, Software Development,
Product Development Payoff Function: Asymmetric, FTW.

In short, Taguchi Curves have made important contributions to our understanding of the importance of impact of variance, especially in the manufacturing context. When bringing those ideas to Product Development though, the game changes, and how we tilt the playing field also changes. Not only is Cost of Delay is very different to Taguchi Loss Curves, but the applicability of various methods and means to analyse the impact of variance needs to be applied very carefully in the presence of asymmetric payoff functions. In Product Development, variance is sometimes good. Necessary, even.