The horrible truth about software development estimation, and what to do about it
The Horrible Truth
In recent years I've been working with many software development teams and almost all of them struggle with estimating the work. The energy spent on this and the frustration it causes should come to a halt. In this short article, I combine knowledge gained by many people with my experience and explain how to address this.
A significant amount of time is spent on estimating the work required to build software. I've seen teams spending around two days every two weeks trying to understand how much the work will cost, with organizations spending valuable time of managers and experts haggling, pushing and fighting over estimations.
Does it work? Do they get the estimations right? No, they don't. Why? Simply put, building software is complex. We think we know what's ahead of us, but we just don't. This is very difficult to grasp and accept. Requirements evolve. Design evolves. We gain understanding through the work. Things change all the time.
And so, as your expertise increases you come to the conclusion that, while it is usually possible to know the magnitude of order (hours/days/weeks), accurate estimation of the work in hours, days, weeks or months is almost impossible.
This is a great breakthrough. Once you understand this reality, that accurate estimation of software development is not possible, you can start moving forward.
What can be done? How can we have a prediction of when things will get done?
Reduce possibility of development going out of control
The key to the solution is to reduce the possibility of the development process getting out of control. Instead of trying to figure out how much effort it will take to develop features, we take two main measures to increase the predictability of our development process.
The first measure is keeping the batch size, the amount of requirements we develop, small enough. This is because the complexity of development increases at an exponential rate with the growth of the number of requirements.
The main question, therefore, when estimating work is not how much effort it is but rather is it in the right size. If it is too big we need to break it down to smaller pieces.
The right size may change from team to team according to several factors, but from my experience a software development team working with modern tools usually aims towards a few days per development item, something that can be demonstrated to a product manager.
Below see a cycle time control chart of a development team, produced from Jira Software. Each dot is one or more development items that ended on the time indicated by the X axis. The Y axis shows how long it took to develop that item. This team's average cycle time is around 4 days, but from time to time There are items escaping that average - the team discusses these items in order to improve.
To summarize, the first thing to do to handle the problem of estimating work items is to try and bring them down to a size the team is comfortable with. Whether it is 2, 3 or 4 days doesn't matter - what matters is that it is more or less the team's usual size, which gives plenty of space for changes during the actual development.
Increase development predictability
The second approach to adopt to increase development predictability, is combining forces, working together as a team.
To handle the complexity of software development we use a whole team approach. The team plans its work and does it together. The approach of the team leader making the plan and then each team member getting an assignment that they work on alone, doesn't work.
Practically, what does "Teamwork" mean? It means planning together: the team looks at the scope of items, determines together whether they are small enough (see above) and decides on their forecast for the development period. The team then starts handling the various items. They take items according to priority and work on them together. Working on items together sometimes means working on two parallel streams of the same job, sometimes it means sitting together at one development machine (pairing). It means that according to need and priority people switch tasks to help their teammates, and in general the team does whatever it takes to get the work done.
Here is an example of a team who moved from investing a lot of time (2 days per person per 2 weeks) on estimating to working together and avoiding exact estimation. The gray columns are how much was planned for a development period and the green columns are how much was actually done (the changes in throughput are due to personnel changes)
To summarize, instead of being heartbroken over spending endless time on estimation and not hitting the estimates, strive to break your work to smaller, lower complexity items and focus on teamwork. Software development work is thinking. And nothing beats thinking done together.