You started the project. You spent hours, no: days! estimating the project. The project starts and your confidence in its success is high.
Everything goes well at the start, but at some point you find the project is late. What happened? How can you be wrong about estimates?
This story very common in software projects. So common, that I bet you have lived through it many times in your life. I know I have!
Let’s get over it. We’re always wrong about estimation. Sometimes more, sometimes less and very, very rarely we are wrong in a way that makes us happy: we overestimated something and can deliver the project ahead of (the inflated?) schedule.
We’re always wrong about estimation.
Being wrong about estimates is the status quo. Get over it. Now let’s take advantage of being wrong! You can save the project by being wrong. Here’s why...
The art of being wrong about software estimates
Knowing you are wrong about your estimates is not difficult after the fact, when you compare estimates to actuals. The difficult part is to make a prediction in a way that can tested regularly, and very early on - when you still have time to change the project.
Software project estimates as they are usually done, delay the feedback for the “on time” performance to a point in time when there’s very little we can do about it. Goldratt grasped this problem and made a radical suggestion: cut all estimates in half, and use the rest of the time as a project buffer. Pretty crazy hein? Well, it worked because it forced projects to face their failures much earlier than they would otherwise. Failing to meet a deadline early on in the life-cycle of the project gave them a very powerful tool in project management: time to react!
The #NoEstimates approach to being wrong...and learning from it
In this video I explain shortly how I make predictions about a possible release date for the project based on available data. Once I make a release date prediction, I validate it as soon as possible, and typically every week. This approach allows me to learn early enough when I’m wrong and then adjust the project as needed.
We’re always wrong, the important thing is to find out how wrong, as early as possible
After each delivery (whether it is a feature or a timebox like a sprint), I update my prediction for the release date of the project based on the lead time or throughput rate so far. After updating the release date projection, I can see whether it has changed enough to require a reaction by the project team. I can make this update to the project schedule without gathering the whole team (or "the chosen ones") into a room for an ungodly long estimation meeting.
If the date has not changed outside the originally interval, or if the delivery rate is stable (see the video), then I don’t need to react.
When the release date projection changes to a time outside the original interval, or the throughput rate has become unstable (did you see the video?), then you need to react. At first to investigate the situation, and later to adjust the parameters in your project if needed.
The #NoEstimates approach I advocate will allow you to know when the project has changed enough to warrant a reaction. I make a prediction, and (at least) every week I review that prediction and take action.
Estimates, done the traditional way, also give you this information, but too late. This happens because of the big-batch thinking the reliance on estimations enables (larger work items are ok if you estimate), and because of the delayed dependency integration it enables (estimated projects typically allow for teams that are dependent to work separately because of the agreed plan).
The #NoEstimates approach I advocate has one goal: reduce feedback cycle. These short feedback cycles will allow you to recognise early enough how wrong you were about your predictions, and then you can make the necessary adjustments!follow him on twitter