Monday, July 13, 2009

How to make good estimates?

All of us faced the problem of giving estimates. The problem here is that sooner or later we have to meet the milestones planned out of our estimates. When we can't - we get our due portion of frustration from managers.

Estimation is a projection. In order to do it precisely you need to know of the task as much as possible in advance. If a task is new to you then you hardly be able to predict all the details with due preciseness. Best estimates are made for the task that we know well. The more we know of the task, the more precise our estimate can be.

We can compare the task against other known task we performed in the past or we can just use our experience with similar tasks. If we never did anything alike before then it makes sense to do research or a sample that will provide you with clues on how long it make take. For example, you need to translate all the code from Visual Basic into .NET but you didn't do that before. Before giving estimates, you can select some piece of code for a sample, port it to .NET and extrapolate the result to the whole bunch of code.

Estimate is the measure to something that did not happen yet. When we talk about future we always consider different variants how things may go. Many defects created by newbie, new development environment or even power supply may bring us some surprises. As far as we assume and measure future, every estimate implies probability. Point estimate (like 8 or 10 days) has almost 100% probability to fail (because the actual work can take 7.5 days or 11 days). So, this is not very unwise to give point estimates. Make it a range and share your vision on probability. Whatever you chose for a range of values it should be big enough to make you confident you can make it.

For example, if I would be asked to provide estimate on the weight of a space ship (that I have no idea about) I would provide a very rough estimate. Rough means that the range would be really big (50 to 200 tons).

Rough estimates are not what will make your manager happy. Actually he or she will be a bit confused and perplexed by it. They can't use it for planning. To resolve his or her confusion you need to express what you are going to do about it. Tell your manager what other information or what prototyping or online investigation you have to do to learn more about the problem.

As I have written above, the more we learn of the task the more precisely we can project on its pace and timing. Do not hesitate to ask for help. Managers can also be useful ;) They can help you finding the required hardware or procure tools. They can even suggest who to ask for the consultation on a specific problem. Inform you manager what else you need to learn about the task before you can make a more precise estimate. Also, mention when you can revise your estimate so a manager can adjust his or her plans accordingly.

A good estimate may undergo several revisions before it becomes reliable. A reliable estimate is what you believe you can make almost for sure. "Almost" is for the contingencies that could hardly be predicted in advance (power down, system crash, network problems, illness of key people, etc.).

In order to improve, keep tracking the preciseness of estimates on different stages (project initiation, design, implementation, testing, and maintenance). Analyze deviations to find out what to focus on in the future. Achieving preciseness of 10% is considered really good.

Following these recommedation you will find out pretty soon that making estimates is not a pain but fun. Good luck!

P.S. Next post will be on how to elicit good estimates. Stay tuned!


  1. Thanks for the article, this article was highly helpful for me to learn how to make good estimates?. Cheers

  2. You are welcome :) If you have any question regarding estimation I would gladly answer. Just pots it here in the comments.