Wednesday, January 04, 2023

Even the Best Chefs can't make a bad recipe taste good; why someone else's Agile Transformation recipe may not work for you

 I've been reflecting on some of the agile transformations I've been involved in over the past few years. In some cases, I see some real progress made. Other times it seems like just re-arranging the deck chairs on the Titanic. An organization follows some recipe, but they don't get the results. Why?

I was working with a large financial institution. They had adopted the "Spotify Model" (yes I know, it's not a real thing) and after that they adopted another scaling framework. I came in and started looking at what was going on and discovered they had a huge challenge with dependencies across teams and products. They failed to realize how much the dependencies on other teams was slowing any one team down.

I took some on-line training recently, a course called Agile Physics - the Math of Flow led by Troy Magennis. In it, he talks about Amdahl's law, which was developed in the late 60s to explain parallel processing. The bottom line is that there is always going to be a limit on how much additional output you can get by adding more processing power. Troy points out the law applies to people and teams as well as CPUs.

One source of the limit is how much can be done in parallel, or said another way, how much dependency there is between teams. So two processors (teams) in parallel with no dependencies can produce twice the output of one. However, as less of the work can be done in parallel (ie, dependencies), the output drops. Using the formula Amdahl developed, if you have 4 teams and 80% parallelization, your output is 2.5 (not 4). If you want to learn more, take the's free and some really interesting ideas.

Back to my client with all the dependencies. What could have helped them? Maybe if before they applied the recipe for scaling, they took some time to look at how their teams were set up and came up with a better approach there, that is, one that reduced/eliminated dependencies. More on this idea in a future blog post. 

No comments: