Labor productivity measures the efficiency of the transformation of labor into something of higher value. It is the amount of output per unit of input: an example in manufacturing terms can be expressed as X widgets for every hour worked. Labor productivity (typically just called productivity) is a fairly simple manufacturing concept that is useful in IT. It is a powerful metric, made even more powerful by it’s simplicity. At its heart, productivity is a measure of the efficiency of a process (or group of processes). That knowledge can be tool to target and facilitate change. The problem with using productivity in a software environment is the lack of a universally agreed upon output unit of measure.
The lack of a universally agreed upon, tangible unit of output (for example cars in an automobile factory or steel from a steel mill) means that software processes often struggle to define and measure productivity because they’re forced to use esoteric size measures. IT has gone down three paths to solve this problem. The three basic camps to size software include relative measures (e.g. story points), physical measures (e.g. lines of code) and functional measures (e.g. function points). In all cases these measures of software size seek to measure the output of the processes and are defined independently of the input (effort or labor cost).
The standard formula for labor productivity is:
Productivity = output / input
If you were using lines of code for productivity, the equation would be as follows:
Productivity = Lines of Code / Hours to Code the Lines of Code
There are numerous factors that can influence productivity like skills, architecture, tools, time compression, programming language and level of quality. Organizations want to determine the impact of these factors on the development environment.
The measurement of productivity has two macro purposes. The first purpose is to determine efficiency. When productivity is known a baseline can be produced (line in the sand) then compared to external benchmarks. Comparisons between projects can indicate whether one process is better than another. The ability to make a comparison allows you to use efficiency as a tool in a decision-making process. The number and types of decisions that can be made using this tool are bounded only by your imagination and the granularity of the measurement.
The second macro rational for measuring productivity is as a basis for estimation. In its simplest form a parametric estimate can be calculated by multiplying size by a productivity rate.
1. The lack of a consistent size measure is the biggest barrier for measuring productivity.
2. Poor time accounting runs a close second. Time account issues range from misallocation of time to equating billing time to effort time.
3. Productivity is not a single number and is most accurately described as a curve which makes it appear complicated.
Variants or Related Measures:
1. Cost per unit of work
2. Delivery rate
3. Velocity (agile)
There are several criticisms of the using productivity in the software development and maintenance environment. The most prevalent is an argument that all software projects are different and therefore are better measured by metrics focusing on terminal value rather than by metrics focused on process efficiency (artesian versus manufacturing discussion). I would suggest that while the result of a software project tends to be different most of the steps taken are the same which makes the measure valid but that productivity should never be confused with value.
A second criticism of the use of productivity is a result of improper deployment. Numerous organizations and consultants promote the use of a single number for productivity. The use of a single number to describe the productivity the typical IT organization does match reality at the shop floor level when the metrics is used to make comparisons or for estimation. For example, would you expect a web project to have the same productivity rate of a macro assembler project? Would you expect a small project and a very large project to have the same productivity? In either case the projects would take different steps along their life cycles therefore we would expect their productivity to be different. I suggest that an organization analyze their data to look for clusters of performance. Typical clusters can include: client server projects, technology specific projects, package implementations and many others. Each will have a statistically different signature. An example of a productivity signature expressed as an equation is shown below:
(Note this is an example of a very specialized productivity equation for a set of client server projects tailored for a design, code and unit testing. The results would not representative a typical organization.)
A third criticism is that labor productivity is an overly simple metric that does not reflect quality, value or speed. I would suggest that two out three of these criticisms are correct. Labor productivity is does not measure speed (although speed and effort are related) and does not address value (although value and effort may be related). Quality may be a red herring if rework due to defects is incorporated into productivity equation. In any case productivity should not be evaluated in a vacuum. Measurement programs should incorporate a palette of metrics to develop a holistic picture of a project or organization.