A Gemba learning for 2019.

Our team were discussing some data about Story sizes. The data showed that our average Story size was around 1 day of effort. Were we right to feel that 1 day Stories are “good enough”?

Now 1 day Stories feel quite manageable. We run 2 week Sprints. If you have 6 people in a Scrum team, that’s maybe 50 Stories closed in a Sprint. That should give a nice flow and a smooth burndown.

But I could see one of the team was concerned. If that’s what the data shows, he mused, then why doesn’t it ever feel that way? It feels like we’re always working on big stories, not changing Stories every day.

In the end we realised we had fallen into a classic statistics trap in understanding the data. “Average story size of 1 day” doesn’t mean “all stories are 1 day”. And that variance causes a problem.

Let’s look at an example. Imagine we had 100 Stories. We’re going to be a bit extreme for this example.

• 99 of these are trivial changes, and can be done in an hour.
• One, however, is a real problem and would take 100 days to complete.

So what’s the average Story size? Well it’s:

( 99 x 1 hour + 100 days ) / 100

That’s about 1 day, just like we have. But let’s say you put in enough work to finish all of the stories. That will take you about 112 days. So how much of that time are you working on the 100 day story? Well, it’s 100 days. That would be:

100 / 112 = 90%

Even though the Stories average only a day, you’re spending 90% of your time on the 100 day story. This is the trap we’d fallen into. No wonder it doesn’t feel like people are working on short stories. Like most teams, we had a mix of Story sizes on the backlog from small bug fixes to larger items that maybe should have been broken down.

Maybe our average was a day, but that was no reason for complacency. People race through those and it’s the larger Stories of five or six days that were taking most of people’s time.

Is there a New Year message for 2019 in all of this? Maybe it’s the good Lean principle, that data alone can mislead. Leaders should combine the data with going to the Gemba (the place where value is really generated) and looking and listening.

Go see, ask why

Fujio Cho (Toyota Chairman)