Making Software

How Much Architecting Is Enough?

If a steep change-of-cost curve makes Agile methods such as XP impossible, how much should you invest in making preparations for a development project? Some decision makers draw on analogies such as, “We pay an architect 10% of the cost of a building, so that’s what we’ll pay for software architecting.” But what percentage is really appropriate for software? Perhaps 10% is way too little or way too much. Many cost-cutting decision makers see that software architecting doesn’t directly produce the product, and as a result try to minimize its cost. But this often leads to an increased amount of late rework and embarrassing overruns in time and costs. Here again, any relevant evidence on “how much architecting is enough?” would be highly valuable for enterprise or project decision makers.

Cost-to-Fix Growth Evidence

Evidence on how the costs of making changes or fixes vary phase by phase for large software projects was reported by several organizations in the 1970s, including IBM [Fagan 1976], GTE [Daly 1977], TRW [Boehm 1976], and Bell Labs’ Safeguard project [Stephenson 1976]. These studies found rather consistently that a post-delivery software change was about 100 times as expensive as a requirements-phase software change. Figure 10-2, taken from the 1981 book Software Engineering Economics [Boehm 1981], summarizes the results of these studies and a few others. It shows that, although the 100:1 ratio was generally true for large systems, a 5:1 ratio ...