Continuing with the series, this time I want to highlight a very dangerous anti-pattern: using velocity as a performance metric. Before getting into the examples of how it applies to velocity, I want to first explain my view on metrics. I am in favour of metrics and coming up with interesting ways of displaying data (information visualization is a very interesting topic). However, the problem lies in the way that these metrics are used. There are two main types of metrics that I like to categorise as:
- Diagnostics Metrics: these are informative measurements that the team uses to evaluate and improve it’s own process. The purpose of collecting them is to gain insight into where to improve, and to track whether the proposed improvements are taking effect. They are not associated to a particular individual or to how much value is being produced. They’re merely informative and should have a relatively short life-cycle. As soon as the process improves, another bottleneck will be identified and the team will propose new metrics to measure and improve that area.
- Performance Metrics: these are measurements of how much value your process is delivering. These are the ones you should use to track your organisation’s performance, but they should be chosen very carefully. A good approach is to “measure up”. Value should be measured at the highest level possible, so that it doesn’t fall into one team’s (or individual’s) span of control. People tend to behave according to how they’re measured and if this metric is easy to game, it will be gamed. There should also be just a few of these metrics. An example of one such metric would be a Net Promoter Score (that measures how much your custumer is willing to recommend you to a friend) or some financial metric like Net Present Value (read Software By Numbers if this interests you). As you can see, these are very much outside of a team’s control and to be able to score high on them, they should try and do a good job (instead of gaming the numbers).
Going back to velocity, a very common mistake is to use it as a performance metric instead of diagnostics. Velocity doesn’t satisfy my criteria for a good performance measure. Quite the opposite, it’s a very easy metric to game (as mentioned in my previous posts). When approached as a performance metric, it’s common to see things like:
- Comparing velocity between teams: “Why is Team A slower than Team B?” Maybe because they estimate in different scales? Maybe their iteration length is different? Maybe the team composition is different? So many factors can influence velocity that it’s only useful to compare it within the same team, and even then just to identify trends. The absolute value doesn’t mean much.
- Measuring individual velocity: as highlighted by Pat, this is a VERY DANGEROUS use of velocity, and it can actually harm your process and discourage collaboration.
- A push to always increase velocity: it’s common to have a lower velocity in the beginning of a project, and that it tends to increase after a number of iterations. Inspite of that, I’ve seen teams pushing themselves to improve it when they reach a natural limit (Who doesn’t want to go faster, right?). Velocity measures the capability of your team to deliver and, as such, tends to stabilise itself (if you have a stable process and the number is not being gamed). A Control Chart could help you visualise that. As noted by Deming, in a stable process, the way to improve is to change the process.
It’s important to remember that velocity is a by-product of your current reality (your team, your processes, your tools). You can only improve your process once it’s stable and you know it’s current capacity. Velocity is just a health-check number that will tell your team’s capability. It will not tell you about how much value is being delivered or how fast you’re going. You can deliver a lot of points and make trade-offs on quality which, no matter how you measure it, will impact your ability to go fast in the long run. As uncle Bob says:
“The way to go fast, is to go well”
So let’s stop using velocity to measure performance and look at it as a diagnostic metric to improve our software delivery process.
Continuing with the series on how to misuse velocity, the second anti-pattern I would like to highlight is when teams start making up points. Because the definition of velocity is so simple, it’s easy to game the metric to show what looks like apparent progress. If a team is being measured on velocity (more about this on later posts), it’s quite easy to just start increasing estimates: “If we just double all estimates, the relative sizes stay the same, but our velocity doubles!“. This is an extreme behaviour that would be quickly noticed as a discrepancy, but the same thing could happen in a smaller scale and pass unnoticed.
This problem can not only be originated from the team, but I’ve also seen Project Managers/Scrum Masters coming up with “clever” ways of making up points to count as velocity:
- Counting percentage or half points (as mentioned in my previous post)
- Deciding to split a story to count the partially finished work as complete, and track whatever is left in a separate story (splitting should be business-driven and not tracking-driven: it should only happen when you come up with simpler/incremental ways of delivering value in smaller chunks)
- Counting points on technical tasks. I’ve seen a team that spent a lot of effort in an iteration to make up for accumulated technical debt, and did not have a lot of time to work on new stories. The Project Manager decided to come up with a “refactoring card” and gave it a 16 to try and demonstrate how much effort was spent on such refactoring
- Counting points for in-release bug fixing. In a team, stories were deemed completed on the first iteration, but bugs started to show up in later iterations, impacting he team’s ability to deliver new functionality. Instead of allowing the decrease in velocity to demonstrate how the lack of focus on quality was impacting the team (bugs should be prevented in the first place, right?), the Project Manager decided to estimate and count points on bugs, which kept velocity apparently constant, when in fact a lot less value was being delivered
The next time you catch yourself asking “Should X count as velocity?”, stop, reflect, and ask instead “Should I worry about X happening at all?”. If you are worried about having to track or show progress on things that should be embedded parts of the process (such as activities to prevent bugs or refactoring), chances are that the problem lies somewhere else. Some of these questions might make as much sense as “Should time spent on retrospectives count as velocity?” or “Should going to the bathroom count as velocity?” :-)
I’m sure that these examples drawn from my personal experience are just a few examples of how to make up points and misuse velocity. What other similar experiences did you have in your own projects?
Dan North wrote an interesting post about the perils of estimation, questioning our approach to inceptions, release planning, and setting expectations about scope. This made me think about the implications of those factors once a project starts, and I came up with some anti-patterns on the usage of velocity to track progress. This is my first attempt at writing about them.
Before we start, it’s important to understand what velocity means. My simple definition of velocity is the total number of estimation units for the items delivered in an iteration. Estimation unit can be whatever the team chooses: ideal days, hours, pomodoros, or story points. The nature of items may vary as well: features, use cases, and user stories are common choices. Iteration is a fixed amount of time where the team will work on delivering those items. Sounds simple? Well… there’s one concept that is commonly overlooked and that’s the source of the first anti-pattern: what does delivered means?
One of the most common anti-patterns I’ve seen is not having a clear definition of done. Lean thinking tells us that nothing is really done until it’s delivering value, which in software means: code running in production and being used by real users. Although I know very few teams who can deploy code to production at the end of every iteration (some even do more than once per iteration), once a story is considered done, it could be potentially shipped, if the business decides so. There shouldn’t be a lot of extra work after that.
Another bad implication of this anti-pattern is that some teams decide to change the definition of done and count half-completed work to show progress. Some of the symptoms to help diagnose if your team is suffering from this anti-pattern are:
- The team starts tracking dev-complete stories
- “It’s done, but [we need to write the acceptance test/it’s not integrated with the other system/…]”
- “It’s done, but not done-done”
- It takes a lot of extra work to get the story deployed to production
- After finished, the story goes into the next team’s backlog
- Hearing terms like “development team velocity” or “test team velocity”
- Counting half-points or percentages because “if we don’t count it will look like we haven’t worked”
The solution? Remember that velocity is just a number that provides information for the team to understand and improve it’s process. Forget that you’re tracking it and focus on the entire Value Stream and on what’s really value-added to get things into production. Anything else is just waste. If it’s not done, it’s not done. Accept it, move on, and don’t overcomplicate, because it will only add noise and mask what could have been important information to the team.