End the never ending design discussions: Timeboxed estimation for User Stories, a pattern in Agile adoption

In this post, Jason explains one common problem with the estimation and planning sessions in Agile teams. The problem he describes however, is quite common and has lead to implementation of the Separate Design and Requirements anti-pattern in the traditional software development world.
So I thought I'd write a pattern about a simple and common solution that I've seen and applied to solve that problem.

Timeboxed User Story estimation

Problem:

In the first part of the planning meeting the Scrum team and the Product Owner will review the top User Stories in the backlog to make sure the team and the Product Owner share the same understanding of each User Story.
In the second part of the planning meeting the team will do a short design discussion for each of the items. The discussion should be around what needs to get done for each User Story in order to collect Meaningful Tasks and make the work clear to the whole team.
When the team gets to the definition of meaningful tasks they will get stuck in design discussions often spending too much time in the first one or two stories and then not having a complete decomposition of the other User Stories that would organize the sprint for the team.
Some consequences of this problem are:

• missed tasks in the sprint backlog that end up not being done;
• mis-understanding of the User Story design that leads to inconsistent design decisions;
• unclear definition of the design for a User Story that leads to inconsistent and sometime incompatible architectural patterns in components that need to interact (especially when in the presence of Component Teams);
• Cowboy Coding.
  

Context:

This pattern is applicable in the context of the planning meeting, but can be used also when reviewing a Product Backlog before the planning meeting (see Prepared Product Backlog).
The same pattern can be used to estimate any attribute of a task, such as effort, risk or business value.
In order for this pattern to be applied the team needs to agree to use the same technique. Sometimes some of the team members will refuse to use this practice by stating that it is not "serious" enough or challenging it's credibility. Alternatively people can say that with this method we cannot get "good" design, however it is important to point out that the aim of the planning meeting is not to set the "final" design, but rather the first iteration of the design and especially to achieve consensus with the team on the design patterns and architectural approaches selected. The design, as well as the product, should evolve during the sprint.
In my experience this technique can be successfully used in teams of up to 10 people during the planning meeting.

Solution

The solution is to time-box the estimation for every single User Story, not just for the meeting. Note that the team needs to agree how much time they will spend for each item they accepted for the sprint that is starting. We have used 5 minute time boxes for development stories and 2 minutes for management-related stories.
At the start of the meeting the team agrees on the size of the time box in such a way that it allows to cover all the stories needed within the length of the meeting. For example, if you have 10 stories to cover and 60 minutes meeting that means that after 5 minutes you need to move to the next story (that will give you a 10 minute buffer).
The process for each story is:

1. The facilitator reads the story from the product backlog
   
2. The teams ask questions or define tasks by talking about the design of the implementation
• Example questions are: what modules will this story touch?; do we need help from any other team?; do we already have module X in our version control?; do we have the design for the UI?; will we need to manually test this feature or is it enough to have the automated user acceptance tests?; etc.
  
3. While the team discusses the needed design changes/additions the facilitator writes down the clear tasks (example: refactor class X, change the algorithm for Y, design a view for Z, get support from the DB admin for the DB upgrade in the test environment, etc.)
• For some tasks to be clear, the facilitator may need to ask clarifying questions. See step 2 for examples.
  
4. At 4 minutes (if the timebox is 5 minutes), the facilitator will request everybody to estimate the story (using the Planning Poker method for example).
5. If there is no consensus on the estimation, the discussion continues
• At this time the facilitator can focus the discussion by asking the highest and lowest estimate person: "Why do you think that is the case?"
6. At 5 minutes the facilitator will ask everybody to estimate the story and note down the most popular estimate value
7. The facilitator closes the estimation for this story. Return to step 1.

After the meeting the team has a preliminary list of tasks (that will be improved during the sprint) and a consensus on some basic, yet critical design/architectural decisions.
Even if this method seems "minimalist", our experience is that it works well-enough even for 4 week sprints. Although, due to the size of the tasks it is easier to apply to 2 week sprints.

Resulting context

• The first few times the team will try this technique their discussion may be very superficial, especially if the team has never had design discussions in short spurts before. This is to be expected and the experienced ScrumMaster will keep her ears open for symptoms of this during the daily meetings and call for additional design discussions during the sprint, when needed.
  
• The first few times the team may feel that the process is "artificially" strict. This is a common critique. However the team should stick to the practice for at least 4 sprints before changing it or "customizing" it. Our experience is that people will adapt to the method and, after some sprints, start producing high-quality high-level design already during the short discussions in the planning meeting.
  
• If the first part of the planning meeting was not enough to clarify the aim of each of the stories to be estimated the team may have to discuss the stories from the requirements perspective before they can have a design conversation. In this case the facilitator should stop the discussion and defer the estimation of that task to when the Product Owner has explained well enough the story (this is the Unclear Stories anti-pattern).
  

See Also (patterns and anti-patterns not defined yet):

• User Stories for communication across functions
  
• Meaningful tasks
  
• Component teams (anti-pattern)
• Cowboy coding (anti-pattern)
• Unclear stories (anti-pattern)
• Prepared product backlog
  

Resources

A good practice for the actual estimation is the "planning poker game". Here are some resources:

• A timer for planning poker
• Planning poker explained
• A web-based planning poker for distributed teams