The Importance of Precise Estimates
We’ve got a rather old bathroom scale.
I’ve found advertisements for this model ranging from 1949 to 1954. My wife inherited it from her mother. When we first got it, we set the “zero” weight to three pounds to give the “correct” weight. Now it’s set closer to five pounds when nothing is on it.
The readings it gives are a little inconsistent. When you bounce lightly up and down, you may get a different reading. (I tend to take the smallest number I can get.)
My weight seems to vary from day to day, also. Some days I eat more. Some days my body seems to be bloated with water. I’m up two pounds one day and down two the next. This can happen within a day, also, prompting my wife to warn me, “Never weigh yourself in the evening.”
We’ve talked about getting a new scale. I almost did, once. But this one works pretty well, all things considered. It’s good enough for me to “watch my figure” as the advertisements put it. I don’t need to “know right away” if I gain or lose a pound. There’s that much noise in the data, and precise measurements won’t take it out.
In fact, knowing that the measurement is imprecise makes me worry less about the small variances. I don’t try to calculate the percent gained or lost in a day. Instead, I just watch the trends. And this scale is great for that.
Recently I was talking with a client who wanted to estimate what functionality his team could release within the next 18 months. He was concerned that they were only estimating stories for the next iteration. I showed him how you stack the features on the vertical axis of a burn up chart, graph the features already done, and eyeball what could be included in the next 18 months. In fact, if you wanted to, you could indicate the potential size range of a feature, and plot both a slow and fast development rate, giving an approximate trapezoid of possibility.
He wanted a mathematical calculation. <sigh/> You can do that, of course. You can convert everything to numbers. You can apply adjustment factors and confidence percentages. And you can calculate some very precise numbers. You can recalculate them daily and worry over small changes in those numbers.
But what does it change?
The estimation of the size of the work, even after it’s done, is a guess. The measurement of progress is fraught with inaccuracies and unknowns. This means that any precise number you calculate from your estimates is still a guess.
Why spend all that effort? A precise guess is of no more use than a rough idea. Maybe less, because the rough idea makes it plain that these results are just the best guess we have at the moment. Precise numbers have a way of hypnotizing you into thinking you know.
My advice is to
- measure your progress
- watch the trends
- project the trends tentatively into the future
and relax. It’ll work out the best it can. False precision won’t make it any better.
Great post George. It’s amazing how many people still mistrust their own gut feeling (let alone those of others) when it comes to estimation, somehow believing that a mathematical formula will be more truthful. The imprecise bathroom scale is a very nice metaphor.
Nice post. I hear so often the ‘number’, what is the velocity of the team, how many points can I get in the next 6 sprints. drives me crazy.
Reminds me of “Measure with a micrometer, mark with a crayon, cut with a chain saw.”
Good post George, especially the advice at the end. In my experience, estimates are rarely right on track, but after awhile you start to see important trends; such as always way over estimate when dealing with a certain part of the domain or codebase. Such patterns are always good indicators of places to either improve customer relations, domain knowledge, or even parts of the system that are in dire need of care. 🙂
Right on, George. This is such an elegant illustration of the penny wise, pound foolish obsession with pie chart precision which is too often doted upon project metrics. We don’t serve the numbers; the numbers serve us.
Great post! Love the metaphore
Simple and great post George. Great metaphor as well. I suppose that your client doesn’t have answer to the question: “Why do you need those numbers?” I guess, “false feeling of certainty” would be the honest answer.
Reminds me of “It’s better to be roughly right, than precisely wrong.”
Very nice, George, and timely! I haven’t had this conversation at a client since, um, yesterday. 🙂
Yes, people has the false idea the mor eprecise numbers are, the more accurate they are. Where in reality it is the opposite.
If I tell a boss it might take between 1 and 2 years to finish a project. That estimate can be very accurate, but sure is not precise. Someone else might say 341,5 days. That estimations is very precise. But it is not more accurate.
(Actually theiris a bigger chance that I’m right and he is wrong.)
y
George, great post. I’m asked this question all the time in interviews. How do you estimate a project, a block of work, etc. and I tell them.. honestly… and you’ve summed it up at the end. Yet still so many people want that perfect equation..even in the presence of actual data that proves that previous estimates were always wrong.. sigh…
I wonder if the people asking for precise estimates interpret the estimates from route planning software as promises too. I’d love to hear them explaining themselves to the law;
– You see I had too speed up in order to get to my destination at the time promised by my GPS. 🙂
Great post!
I’ll tweet to that!
(author of 1976 Software Metrics)
This entry had been republished at PM Boulevard, but it appears to no longer be there.
This is still relevant. I guess too many people didn’t catch this episode of Corner Gas: http://www.cornergas.com/episodes/?season=4&episode=3
Michael Bolton has written a post covering similar ground. (http://www.developsense.com/blog/2014/09/weighing-the-evidence/) It’s well worth a read.