Complexity Budgets

November 3, 2020

I have a couple of really cool post ideas in the works. They use the concept of a complexity budget, which seems just a little too big to leave unstated or try to define in a footnote. So this is going to be a short post that describes what I mean by that without a ton of other context.

To interact with a human product in the way that its designer intended, one needs some amount of context[1]. This context--roughly "what you need to keep in mind while you're using a thing"--is what I describe as complexity. A famous and apocryphal quote about system design goes, "Everything should be made as simple as possible, but not simpler." The floor alluded to in that quote--that if you simplify something too much, it no longer works-- is the idea of a complexity budget. System designers should try to make their systems as simple as possible to interact with, as long as the simplification doesn't interfere with important aspects of the system[2].

This isn't a new idea, but it has applications and consequences worth exploring. The first of these is that a complexity budget can be a reason to make a system-design choice. One way to decide which of two features to add to a system is to estimate which one delivers the most extra value with the least added complexity. Just as it can be economically infeasible to build certain types of systems, it can be infeasible to build certain types of systems where the complexity required to use them outweighs their benefit (you could call this one category of economic infeasibility; I'm not sure it makes much difference).

A complexity budget applies to an interaction with a thing, not the thing itself. We can't say what the complexity budget of "a spoon" is--are we talking about using it, making it, or something else? Most systems have a few different obvious complexity budgets to consider. For instance, a coffee shop includes a customer's experience--how easy it is to order, etc--and a barista's experience. These can be broken down further in whatever way is most convenient for the system designer--it might be useful to think about a complexity budget just for the graphic design for the menu.

In software, it can be easy to overlook one very important complexity budget: the complexity involved in trying to understand, change, or update a system. This is often visible in services that prioritize fast initial deployment over other design considerations. Such a system can be deployed easily, but making changes or updating to a new version is a daunting prospect. A system can also acquire complexity over time as pieces are added. When this happens, people who already know the system may not be able to accurately perceive its complexity until they try to explain it to a beginner. Finally, there is a complexity budget inherent in trying to make a new system--the system that gets produced can't exceed the skills, capabilities, and attention of the team that builds it.

I use this idea a lot, both when exploring design ideas for new things and when trying to understand existing systems. I'm working on a longer post exploring the history of some foundational concepts in the contemporary internet, and there are several weird choices that make perfect sense when you look at them in terms of their relationship to a complexity budget that was relevant when they were made.


  1. For the sake of argument, we will also say that it's possible to identify a kind of local context--context about a specific object or system that is sufficient to interact with it. ↩︎

  2. I'll again plug Fred Brooks' "No Silver Bullet–Essence and Accident in Software Engineering" (PDF) as a great discussion of an overlapping idea. ↩︎