Lessons learned using SCRUM


Paylogic has been using SCRUM for its engineering process for years now. We've made quite a bunch of changes throughout this time and learned a lot. Right now we have a highly repeatable and predictable development process, which makes it very easy for new engineers to start and almost guarantees that a commitment will be met. I would like to share some of the most important things we've learned along the way.

When doing SCRUM, do SCRUM

There are way too many companies out there doing something and calling it SCRUM (or agile), and not actually doing it. You have the opportunity to stand on the shoulders of giants. Use those giants, especially when just getting started. Understand why things are done, and only then modify if needed. Avoid ScrumButs as much as possible.

And if it turns out you don't like SCRUM, use something else, but don't play pretend.

There really isn't much process

There's some, more than with ad hoc programming, and a bit more than with Kanban, but definitely not a lot.

At Paylogic, in an average two week sprint with three teams of in total 14 engineers, we spend two hours on sprint planning, half an hour per team on backlog grooming, about 20 minutes on the sprint demo and about 40 minutes on the sprint retrospective. Then we of course have standups, which take about 10 minutes per day and are done on all days except for the sprint planning day.

This totals to five hours of process per person per two weeks, or two and a half hours per week. Based on a 40 hour work week, that's about 6%.

The process that exists is good!

Now 6% is all well and good, but every bit of process is too much if it has no outcome. However, the structure of SCRUM merely asks you to do what you should be doing already. Basically, you should:

  1. Work in iterations;
  2. Make a plan for each iteration so that the business knows what's coming;
  3. Keep each other updated on your progress and your issues;
  4. Show off what you created; and
  5. Evaluate your iteration and take action to improve.

I find it hard to find fault in any of those steps, and they are basically the only steps you have to do to implement SCRUM.

Timebox your meetings

Meetings have a tendency to drag on (see also Parkinson's Law). It's fairly simple to prevent this however.

In all SCRUM meetings (and most productive meetings actually), you are dealing with a number of consecutive items, and when all are handled you're done. In the case of the sprint planning each item is every story that fits in the sprint, and in the case of a standup it's each member of the team. Simply set a deadline for each item and track and enforce this deadline. People will focus on the most important points to most optimally use their time, and after a few times, almost no-one will get even near to the deadline.

Team members should be in charge of their individual velocity

Engineers take pride in doing their work well. Creation is in our blood. It should be natural to leave it up to the engineer to determine the speed in which they feel they work, or want to work. Velocity should not automatically be computed by a script or mandated by a team leader, and should not be averaged out over a team. It is important to recognize the individuals in a team, and they should have the opportunity to set their own velocity based on past results and future ambitions.

Next to that, it also makes planning a lot easier. If you have 80 story points available in a sprint, it usually doesn't mean that you can plan a single story estimated on 80 story points in that sprint. Work is often not that parallelizable. If two persons can work on this story, and each person has 20 story points as their own velocity, then it's suddenly simple to see that you can do 40 story points of that story.

Use proper tooling

I wrote an article on how we use Trello, but any tool or method that allows you to visualize the work being done and shows your progress towards the end of the sprint is fine. Don't skip this; not having it basically means flying blind, and you won't ever get controlled sprints.

Plan everything

Don't only plan for the user stories scheduled for a sprint. Recognize that other work also has to be done. In our case, we had a period where we had to reduce the commitment every sprint, because one of the engineers had to perform maintenance on the new Continuous Integration server regularly. Keep monitoring which work is actually done, and make sure to create space for this work to be planned as well.

Backlog grooming is essential

We started backlog grooming relatively late, as we didn't think we needed it. Turns out we did. It's really nobody's fault but the team's if it finds out during sprint planning that a user story isn't ready for the sprint. And then there's no time to correct it, with frustrations on all sides.

Don't try to actually do estimations or even think of tasks during the grooming sessions. The team should simply review each user story to see if they can work on it in its current state.

A team also needs freedom when determining what to do

The product owner is of course responsible for the backlog, but to get good software, the team needs to be able to influence what they work on as well. The team needs some room to, for example, improve its tooling and development environment and to iterate over earlier designs. It is essential for a product owner to take this into account.

If this turns out to be difficult to negotiate, balance can be restored by allowing the team to schedule a portion of its time (say 10%) by itself, regardless of the backlog.

Commitment vs forecast

In the official SCRUM description, there actually exists no such thing as a commitment. Instead, there is a forecast. The idea of this forecast instead of commitment is great, the thing is however that often it doesn't matter. Whether you use the term forecast or commitment, business people will still expect you to deliver what you said you would deliver. Because that's how they work.

There's no real cure for this. Communication is very important here, but in the end it's just something you should be aware of. Don't expect you can just win this by changing the term. You'll have to change a mindset, and sometimes even a culture, and that's much harder.

Multiple teams can SCRUM together

Having a separate SCRUM team doesn't mean they have to have their own standups and retrospectives. In fact, that's often a bad idea. If the teams are completely independent, kind of like different companies, then it's of course fine. But usually, when you are part of the same company, you work together on different parts of the same whole, and communication between teams is just as important as communication within teams.

In our case, the three teams consist of a total of about 14 members. This is small enough so that we can still have communal standups and retrospectives. Sprint planning and backlog grooming are done separately however.

When teams get larger, a SCRUM of SCRUMS can be used to keep communication going.

Closing thoughts

At Paylogic we've seen our process grow from being used in a single, small team, to a single large team, to scaling to three teams. We've seen people come and go, but the process is still going strong. I am very happy with it and with the performance of the teams as well.

Regardless, there are other alternatives that can work just as well, or perhaps even better. When it comes to raw productivity, I actually think a Kanban process is just a little bit better. A sprint ending and beginning is still disruptive and does reduce productivity, more than just the hours spent on the meetings. Kanban, being a continuous process, doesn't have this.

It does require more discipline to pull off correctly however, and I believe it is harder for a newcomer to get started with. I would like to try such a process in Paylogic however, so perhaps I'll write another article about that by that time.