The case against iteration based re-estimation

Many agile practitioners recommend re-estimating stories at the beginning of each iteration. I disagree with this practice.

For one thing, I believe it's a waste of time. Any value that you might get (which I doubt - see below) from the practice is lost on the time spent.

It's worse than that though. By re-estimating the iteration's stories, you are almost always estimating with a greater level of detail than what you had originally. With this increased level of detail, in my experience, estimates tend to grow.

Why is this a big deal?

Let's try an example.

I come in to my iteration planning meeting with 30 points worth of stories from the backlog. The team commits to those stories, but in re-estimating, the 30 points inflates to 40. In fact, this always seems to happen, as the team gets a little nervous about hitting their historical velocity and they know management is paying attention. Let's assume the team gets them all done. This increases the observed velocity by a third (40 points is a third more than 30). Now, let's say I have 120 more points left in the product backlog to get to the minimal marketable feature set for release. How many more iterations are left?

If you said 3 more iterations (i.e. 40 points per iteration gets you to 120), you are ignoring your team's tendency to inflate estimates. Assuming your estimate inflation rate is consistent (a third), you really don't have 120 points remaining, you have 160 points, or 4 more iterations remaining. Or, calculated another way, if you consider only the initial estimates to calculate your velocity (30), then you can determine that you have 4 iterations of 30 remaining. In both cases, you end up correctly predicting 4 more iterations. Then again - if you use the initial estimates, what value did your re-estimation from 30 to 40 provide you ? I say none.

If you regularly re-estimate at iteration planning meetings, make a note of the original vs. the updated estimates. See if they grow. Consider what impact this is having on the accuracy of your release planning.

OK, I can hear you now. "My team's estimates don't inflate ... some go up; some go down". I haven't seen this, but let's say you do. Let's revisit the example from above with this assumption. You go into the iteration planning meeting with 30 points and walk out with 29. Your velocity is not materially impacted. You are still on track with 3 remaining iterations (roughly). So the question is this: what value did that re-estimation provide? I say none.

When *do* you re-estimate then?

I believe in updating estimates when information arises from experience that pertains to some shared aspect of a subset of stories. For example, let's say that your retrospectives have shown that every time you have a story that hits a certain database, it ends up being much more effort than expected. In a case like this, it makes sense to revist those database stories to ensure that this knowledge is incorporated into those estimates. I call this aspect-oriented re-estimation (adapted from the term "aspect-oriented programming").