In Agile, Top Down Ron Jeffries has written a compelling article describing how he would introduce agile methods into an organisation, starting from the top.
“There’s a recent thread on the Scrum list about how an executive or highly-placed manager could get Agile going. I’ve been one of those guys, and I know a bit about Agile, and here’s how I’d proceed. First, focus management attention on cyclic delivery of running tested software. Second, provide the resources to learn how to do that.”
Ron’s advice centres on measuring the development department by running tested software – that is, by measuring throughput. I very much like Ron’s advice and approach. But I do see some hurdles…
First, if you’re in a position to dictate that the only measure is throughput, that’s great. But if you’re not, the experience of the TOC and Lean guys is that you’re going to have a really hard time convincing the bean counters – and other senior management, for that matter – that throughput is the right measure. Common sense notwithstanding, this is a cost-accounting world. The only saving grace here is that most software development organisations are too chaotic to count anything, so they likely won’t be already counting the wrong things. (An obvious exception being those departments that are or have been in the grip of Accenture-like consulting practices. But then we’re not likely to try introducing agile top down in one of those places, I guess.)
And second, I would expect that the department’s throughput of software will initially (first three to four months) dip until the running and tested parts of the equation are working effectively. Because testing everything, and fixing all of the defects before they go out the door, costs. The investment will eventually be repaid many times over in terms of predictability, trust and speed, but it is an investment nonetheless.
I still think Ron has it right – just don’t expect an easy time of it for the first few months.
Update, 1 aug 05
Discussion with Jason (in the comments on this post) leads me to understand that the choice of metric is critical. In the article, Ron was explicitly and implicitly referring to his own metric, running tested features.