Fool's Gold: Time Estimates

Ryan James Spencer

It would be fantastic if we knew the future. With that knowledge we could plan with the utmost precision. But we are not clairvoyant. We actively exercise a process of guesses we dress up in the fancier name of "estimates". Enrico Fermi would feel these guesses are fine so long as you are within an order of magnitude. This form of educated guess is also known as a "back of the envelope" calculation and most random guesses are not doing anything near this level of rigor. Back of the envelope calculations take rough approximations, simplified assumptions, and various tidbits of top-of-the-mind knowledge to calculate a ballpark figure.

Put yourself in the shoes of a manager of a team and imagine asking each person in this team "how long do you think this task will take to finish?" This could well be any type of task. Estimations may sometimes vary wildly and sometimes group around a certain value. Any kind of grouping is a coincidence. Asking the same person about an estimate even the next day may yield different results. How could you help improve the accuracy of these estimations with such fluctuating results?

One option is to decide on all the work you expect upfront. If you know how work is subdivided exactly you'll know, transitively, how long the total task will take, right? Wrong.

Instead you'll starve people of autonomy around how they can tackle a goal. Starving autonomy means people stop thinking and now you'll have to work even harder to keep everyone productive. It also makes people unhappy, whether they realize it or not, and you'll probably wind up with a great deal of turnover because of it.

What about using time taken from similarly sized tasks? If it took someone a certain number of time to complete a task with a rough "t-shirt size" then it ought to take another, or the same, person roughly the same, right? Wrong.

Even if you had the same person doing the same task there is the possibility that some spontaneous act can change timings drastically. People get sick. Their dependents and partners get sick. Trains get delayed and vehicles break down. The human brain decides to watch a video for an hour instead of fifteen minutes. A meeting that was scheduled for an two hours only takes one.

We tend to attach degrees of confidence to our guesses but never seem to discuss those confidence levels openly. We also tend to wrongly consider tapered ranges as the more accurate guess. We feel pressured to pick the most likely range and make it a promise. Resist this temptation and try making this range of possibilities explicit.

Many people focus on what kind of confidence they back on a single range, but our confidence may differ between slices of respective intervals of time. Try to consider that your confidence in your guesses is not a normal distribution centered around a single range; Confidence may be distributed in any number of ways. There may be a high probability that the job gets finished if we focus on it at the beginning and the end of the year, but a low probability otherwise. If we have the time to work on this task on the fortnight's the probability goes up, but the likelihood of completion dips on the weeks in-between.

Does a low probability mean impossibility? No. Nor does a high probability imply absolute certainty. A range of probabilities discusses the full spectrum of what might be feasible. It is better to know you think a project might take only a months work of time all the way up to six months than it is to merely work under the assumption that the single month would do.

Make back of the envelope calculations by decomposing problems into constituent parts.. The refinement of the accuracy of these parts refines the overall estimation. This isn't to say breaking down a random guess into several random guesses will improve accuracy. In fact, with the random-guessing approach you will probably be reluctant to go below a certain lower-bound, which means that a decomposed guess might far exceed the original guess. Back of the envelope calculations try to tie you to real facts, although possibly simplified, without believing the hype that you can estimate down to the minute based on prior similar scenarios (see above). Machine learning won't save you. The great thing about back of the envelope calculations is that they work equally well for both high- and low-level concerns.

Pretending estimates given on work are guarantees is fool's gold. Push back on demands for promises when you know you are only making a guess. We might get better at making guesses with time by practice and research but a guess is still a guess, educated or not.