What You Will Learn
- 1 Introduction to Scrumban
- 2 Research Approach
- 3 Current State Analysis
- 4 Best Practices of Agile Product Development Processes
- 5 Scrum
- 6 Building a new product development process proposal
- 7 Pilot Testing of the proposed New Process
- 8 Conclusion
Introduction to Scrumban
This research is done in the context of a Finnish case company. The case company was founded in the year 2010. It is a fairly new start-up that has won several awards since its inception. It provides a platform enabling smart content and turns any online content, images, videos and applications into interactive and viral storefronts. With the faster growth of the internet, nowadays almost everyone in the retail industry is turning online. At the same time, media is becoming decentralized; there are several bloggers, publishers and social media clouds. Old and intrusive way of online advertisement is not anymore the answer to current marketing needs. As a result, the concept of smart content is becoming the answer to current online marketing needs. Smart content turns to advertise into service for consumers. It is a non-intrusive service and by simply hovering over or placing a mouse over the smart content on the internet (which can be any media, images, videos, content or anything you see online) displays high relevant information about the product or service in the content. This enables users, publishers and brands to monetize the content and that is always relevant to the context of the online content displayed, For example, brands can display products inside an image or video on some online publisher’s site and those products will be relevant to the context of that particular image or video.
The case company enables brands to turn their content into storefronts and engage directly with their fans everywhere. Publishers can monetize the impulses they generate through the content where the case company added non-intrusive and relevant online products called Kiosks. Through kiosks anyone can buy, want and get more information whenever a user does any interaction with the content. Further, the platform provided by the case company makes it rewarding for online users to discover and share any online content.
Business problem and Objective
The current business problem is related to the Software product development process of applications team within the case company context. In recent years, many organizations have started using agile methodologies for the implementation of software development projects. A few years back, the company in context also started using an agile methodology called Scrum. However, after implementing the scrum methodology for a certain period of time, the software development teams, especially the applications team, faced problems while applying the same. Some of the problems faced were:
Difficult time completing the sprint planning: Many times team members could not agree on the time estimates because of incomplete stories, so the planning meetings dragged on or sometimes effort estimation was canceled altogether for many tasks.
Missing collaboration and problems in self-organization: One of the biggest obstacles that the team members faced was missing collaboration and self-organization. The team members tried to concentrate only on their tasks and not the interdependencies within and outside the team. They assumed that the other team members would handle those interdependencies as they only focused on their own tasks.
Incomplete tasks within a sprint. Often, the team could not complete the sprint tasks as the nature of the tasks was very complex. There were various tasks which included prototype work or research work and it was quite difficult to assess the effort complexity of such tasks. Also, sometimes tasks were getting held in some stage of the workflow. For instance, in some sprints, many tasks were moved to the testing stage simultaneously, and as a result team’s Quality Assurance person could not complete the tasks by the end of sprint iteration.
Product Owners change priority within Sprint: Product owners were changing the priority of tasks in the middle of a sprint. Further, additional tasks were added to the sprints which were not estimated in sprint planning.
Insufficient tasks for some team members: Team members used to commit to X number of story points based on the average velocity of the last 5 sprints. As the scrum methodology focuses on filling sprint backlog based on velocity rather than the number of tasks, members can actually perform, some team members who were specialized in particular technologies were left with insufficient tasks in the middle of the sprint. Further, the team could not pull new tasks from the product backlog as the scrum does not allow that. Sometimes, the team added some new tasks into the sprint which were left incomplete and changed the overall scope of the sprint.
Abrupt switch to Kanban mode in some sprints: Sometimes, the team was shifting from Scrum to Kanban mode because of frequent changes in requirements or the pressure from product management. The management expected the team to get the new tasks done within the same sprint instead of the next one which often lead to the shift from Scrum to Kanban as the team could not simply follow sprint commitments. Further, tasks were getting held in testing or review stage as there were no work policies or maximum limits followed in different workflow stages.
Looking at the issues mentioned above, these lead to delays in the project delivery, increase in the project costs, poor quality of deliverables, and sometimes even failure of sprints. Given this, the objective of this study is to find out a suitable proposal for improved agile software product development process which can handle the situations described above and use the best practices from other agile methodologies such as Kanban or Scrumban.
The output of this study will be improved agile software development process suitable for the applications development team within the context of the case company. Further, this improved process will be pilot tested by the same team for certain iterations. This thesis output could help the team as well as organization to avoid the issues discussed above and handle those situations in a better way.
Limitations & Scope
The scope of this thesis is limited to the case company described in the introduction section 1.1. Within the case company, the scope is further valid only for the applications development team. The aim of this research is not to create any new agile methodology but rather adapt good ideas from the existing methodologies which are suitable for the team under discussion within the context of the case company.
This section describes the research approach used in this thesis which includes overall research design and data collection process.
The research starts with describing the business problem and defining the research objective. This will be followed by the current state analysis of the Agile Scrum Process used within the Applications Development team of the case company. This team consists of five developers and one dedicated quality assurance member. Further, two developers are having dual roles, one is scrum master and the other is the team leader. The main purpose of the current state analysis is to establish reasons for the needed change in the current software development process of the case company. The current state analysis will be compiled based on analyzing the past sprint retrospectives of the team, analyzing the data collected from discussions and interviews of various stakeholders.
The current state analysis is followed by the literature review in order to understand how similar business problems have been handled using the existing literature, and understanding what are the best industry practices in Agile Software Development which can handle the business problem described in this thesis. Accordingly, both the current state analysis and literature review are used together to create a new process model, which is further developed based on data collected from qualitative interviews with all the concerned stakeholders. Finally, new process model will be pilot tested within the development team of the case company for four sprints, so that process can be analyzed in the day to day working environment and accordingly final conclusions will be drawn.
The research design Implemented in this thesis is illustrated in Figure 1.
Data collection and data analysis
The data collection will start with the current state analysis of the agile scrum software development process within the case company. This will include
1. Analyzing all the possible sprint retrospective data of past twenty three (23) sprints, because it is mainly the sprint retrospectives in a scrum process where teams highlight and discuss three main questions namely “What went wrong in the last Sprint”, “What went right in the last Sprint” and “What can be improved in the next Sprint”.
From the data collected from all the sprint retrospectives, each problem which the team was trying to improve in the upcoming sprint was given a score point of one (1). And in case a similar problem was repeated in some other sprint, the score for that problem was incremented by 1. Finally, only those problems were presented in the final report as shown in Table 4 which occurred more than once.
In addition to scrum retrospective data, velocity related data of the past twenty three (23) sprints were also collected. This information is crucial in determining the output of the team in terms of story points completed within a sprint (Velocity) or average story points completed (Average Velocity), and for calculating standard deviation of the velocity. The data was collected from different intranet resources available within the company, and the details of the data collection are presented in Table 1.
2. Interviews and discussions with all the necessary stakeholders within and outside the development team, within the context of the case company. The participants for the interviews were selected from a wider range of expertise areas within the case company in order to get a broader perspective on the current development process and its pros and cons. The participants were selected from within the Applications Development Team, Product Management, Quality Assurance, Ex-Scrum Master of the team and Operations Team. These participants are the main stakeholders who are impacted by the development process used by the applications development teams within the case company. The details of the data collection are presented in Table 2.
The outcome of the current state analysis along with the literature study and review was used to formulate the main research interviews with different stakeholders in order to define the prototype software development process. The participants selected in this round were from within the Applications Development Team, Product Management, Quality Assurance, Operations Team, and ex-Scrum Master of the team.
From the applications development team, two senior developers and the team leader were selected. They have more knowledge of how the team has been functioning from the developer’s perspective, and they know what challenges and problems the team has faced in the long run. The dedicated quality assurance (QA) person of the team was also interviewed in order to know how the team has been performing from the QA perspective, and what challenges and problems need to be addressed for ensuring better quality.
In addition, the product owner/product manager of the team was also interviewed. Product owner observes how a team is performing when it comes to commitments, customer delivery, planning, prioritization and customer feedback. Also, head of overall software development was interviewed, since he is the main owner of processes being followed in different software development teams within the case company. The other responsibilities of the head of the software development include looking after the team’s resource availability, skill set, work culture, and overall delivery of different software products. So his feedback is very important in the current state analysis of this thesis. Further, the quality assurance (QA) head of the case company was also interviewed in order to get his feedback and opinions on the current development process and how it impacts the
the overall quality of software being shipped.
Finally, ex-scrum master of the team was also interviewed in order to get some feedback on how well the current process has worked for the team in past and what were the challenges and issues team was facing during his tenure of team’s scrum master. The other reason was interviewing the ex-scrum master is to bring in new ideas on how well the scrum is working in his current team, know about the challenges and what changes have helped his new team. The details of the data collection used for defining the prototype process model are described in Table 3.
Both in current state analysis and prototype model creation phase, qualitative interviews were conducted and all the interviews were documented. After the new software development model was developed, a power point presentation was presented to all the stakeholders which included product management, development team, team leaders, operations team and quality assurance team. After the presentation, a “Go Ahead” decision was taken by the head of the “Software Development & Services” and product management in order to pilot test this process model for four sprints within the applications development team. The result of the piloted sprints was analyzed and is documented in section 6.
Current State Analysis
Analysis of the Sprint Retrospectives
In this analysis, all the retrospective related data of the last 23 sprints were collected from the company’s various intranet resources such as Jira, Google Drive, Trac Online tool, and Scrum Master’s notes. In the Sprint Retrospectives, every team member answers these three important questions
1. What was good or positive in the last sprint?
2. What was bad or negative in the last sprint?
3. What could be improved in the next sprint?
Based on the above-mentioned questions, Scrum Master would collect all the answers from different team members and accordingly create a list of action points on what could be improved in the upcoming sprint. This information is important for analysing what were the challenges faced by the team in different sprints. The analysis of this data would give very important information on how agile scrum model was used within a team, and whether it was working for the team. From the retrospective data, an eachnegative point that was brought up in the sprint was given a score of 1 and in case the same problem repeated in some other sprint, the score of the problem was incremented by 1. The details of the sprint retrospective data analysis are described in Table 4.
From the sprint retrospective data analysis as described in the Table 4, it can be seen that the team was facing many challenges:
Missing Collaborations or Problems in self-organizing and decision making: Team members only concentrate on their own tasks; in a self-organizing team they should look at interdependences within a team or outside a team and handle those issues as needed. Team members shall organize themselves and not simply concentrate on a task at hand, as is clear from point number 1 in the Table 4.
Some Sprints are simply constant stress: As indicated by points 2, 4, 5 and 10 from Table 4, things keep on changing within a sprint, management adds new tasks within a sprint or modify existing tasks within a sprint. Team members have to complete tasks from the sprint backlog as well as the tasks coming from others. As a result team members are constantly under workload which is not good for the motivation of developers.
Tasks came from every direction: As indicated by point number 4 from Table 4, in addition to sprint backlog tasks other tasks were coming from top management, product owners, designers, and client managers. All these tasks were putting the additional workload on the team and as a result, the team was unable to complete tasks on time or unable to focus on sprint goals. Sometimes incomplete user stories were added to the sprints that were not planned as part of the sprint planning. This could lead to poor quality of deliverables.
Unclear process or weak process: Some aspects of the process were not clear, for instance how to deliver product prototypes? Do tasks related to product prototypes have to follow the same process as for the normal product related tasks? Sometimes workflow process was skipped and tasks were taken from outside the tasks management tool. This is evident from points 3, 4 and 5 in Table 4.
Tasks getting stuck in a particular development stage: By the end of a sprint, many tasks were stuck in some stage of the workflow. As indicated by points 6 and 7 from Table 4, it occurred in many sprints and some stage of the workflow process often became a bottleneck. Reasons were many, for instance when team overcommits to the number of tasks within a sprint and then some stage or other became the bottleneck. Sometimes testers had many tasks in their tasks lists or the developers could not review tasks on time.
A team tends to have too big tasks: From point number 11 in Table 4, it is clear that team was not splitting big tasks in certain sprints and as a result, such tasks were spanning across multiple sprints. This often led to the belief that the team’s estimation accuracy is low, or their visibility and transparency is low.
Poor quality of specifications: As indicated by points 5 and 9 from Table 4, sometimes the scope of the task was changed within a sprint, or incomplete user stories were planned for the sprint. On one hand, the team cannot stop working on tasks because certain aspects of it were not fully specified yet, being as startup company requirements often change as management wants to try and test the results, but on the other, it may lead to poor quality of deliverables.
Conflicting Roles & Responsibilities: As indicated by point number 8 in Table 4, there were certain sprints where non-technical persons were the team’s product owners and they did not have enough competencies to specify requirements clearly. In some other sprints, product owners were having more than one role at a time; also there were no clear owners for the internal or external documentation.
Over commitment of Story Points within a Sprint: As indicated by the point number 10 in Table 4, there were some sprints where the team was overcommitting to the story points compared to what they should have committed based on the average velocity of past sprints. This also led to the belief that the team’s estimation accuracy was low, even though the actual reasons could be many such as pressure from the top management or product owners.
Analysis of Velocity Data
According to whatis.techtarget.com (2013), “Velocity is a metric that predicts how much work an agile software development team can successfully complete with a time boxed iteration called sprint”. The velocity of a Sprint is the number of story points completed within a sprint. The data regarding the team velocity was collected from different intranet sources of the case company such as Jira, Google Drive and Trac. The different sprint metrics, such as, velocity per sprint, velocity per team member, average team velocity, average velocity per team member and standard deviation of velocity data are described in Table 5. Velocity per team member is obtained dividing sprint velocity with the total number of team members present in the team. The total number of team members
in the team under discussion is six (6).
Based on the velocity data described in Table 5, a bar chart representation of the sprint velocity for last 23 sprints is depicted by the Figure 2
As seen from the data in Table 5, the average team velocity for the applications development team using the Scrum methodology is 40.65 with a standard deviation of 10.56. The data also indicates that for the majority of sprints, sprint velocity is within the range of one standard deviation from the mean value. In addition, the average sprint velocity per team member is 6.77 with a standard deviation value of 1.76. This data will help the researcher in comparing the current team velocity with the team velocity once the new process model is pilot tested.
Analysis of the Product Development Process
The current product development process being followed in the case company is agile scrum methodology. In this section, the researcher will be described step by step process, followed by the applications development team within the case company context. At the end of each stage, the researcher has done the analysis of that particular stage. In the scrum methodology, each software development cycle called sprint starts with the sprint planning meeting.
Sprint Planning: Usually last for 2 hours. In this meeting team normally take the tasks from a prioritized set of the product backlog, try to roughly estimate those using a method called scrum poker. Using scrum poker, each team member roughly estimates tasks based on story points, and then the whole team tries to decide on the most appropriate story points for tasks based on what majority votes for. Similarly, the team tries to estimate the rest of the prioritized stories and then roughly selects a subset of tasks into the sprint. The number of story points taken into the sprint is roughly based on the average velocity of last 5 sprints.
- Sprint Planning is so far very helpful; this is the place where the team plans and estimate different stories and tasks.
- Task estimation by the team helps product owners to know the complexity of a task
Sometimes sprint planning is not good enough in case product backlog stories are incomplete or product owner has not properly prioritized the stories, and therefore in certain sprints team end up prioritizing tasks in planning sessions rather than selecting and estimating tasks. In these situations, sprint planning meetings start to drag and often lose focus.
Estimating stories using story points is very much misused by the management who usually compare it to a number of days or man days. On the contrary story, points shall indicate nothing but the complexity of a task.
- Mostly team tends to focus on new stories rather than bugs, so often anything else that blocker bugs keep on pending in the backlog. The team tends to ignore bugs in planning because they are not assigned any story points, and also developers prefer to work on new tasks.
Another drawback in sprint planning is team doesn’t discuss design issues related to tasks, they just estimate the complexity.
Some tasks are so technology-centric that only one or two team members can
roughly estimate the related stories.
Sprint: After planning is done scrum master starts the sprint. Every sprint in the context of a case company lasts for two weeks. Within a sprint team, team members try to focus on completing tasks from the sprint backlog. One by one, team members select tasks from the backlog and each task within a sprint goes through different stages like “To Do (Sprint Backlog)”, “In Progress”, ”Review”, “Testing”, and finally “Done”. Each
In addition, every day, team will have daily scrum (morning stand-up meeting) meeting which usually last for 15 minutes. In this meeting each team member describes what he did yesterday? What he is planning to do today? And is there any impediment that needs immediate attention?
- Sticking to two weeks iteration is good because it helps a team to roughly predict how much work they can complete within a sprint, usually based on work completed in last five sprints.
Daily scrum meeting is very good to keep team focused because it ensures that each team member describes what he did yesterday and what he is planning to do today. In order to answer these questions, team members ensure that they stay focused. Also this meeting helps in resolving work stopper impediments immediately.
Sprint keeps the team focused all the time. Since the team has to fulfil its sprint commitments, there are no dull phases of development which usually may happen in traditional waterfall models.
Reviewing each task within a sprint is very helpful in finding problems at the early stage of development. It also it ensures the quality of the code because someone else has reviewed it. Weakness:
Within a sprint, sometimes the scope of the story is changed by the product owner. This leads to delay in completing tasks and at the same time may result in unfinished tasks by the end of a sprint. 18
Also many times there is pressure
Also many times there is pressure from the product management or the top management to take new tasks in the middle of the sprint in addition to the ones committed by the team during planning. What happens is this usually starts a chain reaction; the flood of additional tasks and changes suddenly break the sprint flow, and for the next couple of sprints process changes to Kanban mode.
In Kanban mode, the team simply processes tasks one by one either from the product backlog or the re-prioritized sprint backlog tasks. There are no policies followed like for instance, rules related to the maximum limit on tasks in a particular stage. As a result, often tasks are still stuck in some stage of the workflow when the sprint is about to end.
Switching between the Scrum and Kanban process modes also reduce team’s motivation, and eventually slow the development process.
Sometimes tasks also get stuck in the review stage of the workflow when many tasks are simultaneously moved to this particular stage. Often these tasks may stay there when team members are too busy with other tasks, and no one has time to review tasks.
Sometimes product owners add new requirements (in addition to sprint tasks), or new user interface (UI) design changes in the middle of the sprint because of customer or management pressure. Even if the developers try to put in extra efforts, this usually leads to many tasks in the “Testing” stage simultaneously. Some of these tasks may remain incomplete by the end of a sprint, even the ones where there was some possibility of getting completed before.
After interviewing the quality assurance (QA) team, it seems sometimes there is a disconnect between the development team and QA. They simply do not tune in or match each other’s expectations, QA and others tend to think that the team does not produce quality code in terms of maintainability and documentation. The development team and QA will need to bridge these gaps, trust and work as single team. Further, they need to have common goals, shared an understanding of how work needs to be done, and also agree on common QA standards.
Sprint Demo: After the sprint ends, the team prepares for the demonstration of the sprint tasks to the scrum product owners, product management, designers (Optional), and sales (Optional) team. In this meeting, the main highlights are the stories or tasks which the team has competed in the sprint that ended. The team starts with what was the sprint goal, what they managed to achieve (committed vs completed stories/tasks), metrics like sprint velocity, and followed by a demonstration of stories/tasks.
Sprint Demonstration is good way of getting the feedback from the product management, designers or other teams.
Product Management knows about the progress of development team and can accordingly plan or prioritize future product requirements
Sometimes management seems to compare a team’s performance based on story points completed within a sprint (i.e. team velocity). Rather than using sprint velocity as a performance indicator of a team, velocity shall only be used for planning the capacity of sprints.
Sprint Retrospective: During sprint retrospective, team retrospect on sprint performance and lessons learned, every team member discusses about negative and positive things related to the sprint. Based on negative points, team tries to create action plan that would help them in improving the next sprint. Every team member writes what went right and what went wrong, then they discuss on how to improve upon in the next sprint and what are the points where they need management’s intervention. Some action points which cannot be handled by the team are discussed in retrospective of retrospectives, where all the scrum masters and product owners of different teams discuss what can be improved at the company level.
- Sprint retrospective is very important tool that helps in achieving constant improvement.
Sometime when team members pinpoint too many problem areas it is very difficult to take action points related to all. One improvement could be restricting the number of issues team members can raise, so that concrete actions can be taken related to those rather than trying to improve everything at once.
Many times some action points are hardly handled because managers or team members are too busy with sprint tasks.
Sometimes there is no right owner who can handle the action points from the management side.
- Product owners are not part of sprint retrospectives even if they are important team members. They can handle many actions points from the sprint retrospective which are related to top management or applicable company-wide.
Product Backlog Grooming: During the sprint, once before the start of next sprint team members along with scrum product owners do the backlog grooming in order to prioritize and foresee what stories/tasks are coming in the next sprint. Optionally, product managers are also part of the meeting in case they want to bring in some additional stories or tasks. Team’s product owner adds new stories or tasks and discusses those with team members. If possible team members can also give some rough estimation in terms of story points. This meeting usually lasts for an hour.
Product backlog grooming helps the team to foresee future tasks & stories.
It helps to keep product backlog in order and prioritized.
In case of product backlog is in order, it simplifies sprint planning sessions. Weakness:
This meeting doesn’t solve its purpose in case the product owner has not done its homework, sometimes product owners are not sure about the scope of tasks or the tasks itself.
Many times product owners are not sure of what new user stories are coming from the management, customers or designers, and as a result, this meeting becomes fruitless. In this case, the team starts to shift to Kanban mode when they do not have enough tasks to select from in the sprint planning meeting.
Sometimes the product owners and product managers are not synching the product backlog effectively well in advance of the sprint planning. Many times sprint planning happened before the product council meetings (product council meetings within the case company are used for defining sprint milestones and product milestones). Some cycle or structure is needed to ensure product backlog is in order and prioritized well ahead of sprint planning or team backlog grooming meetings.
Product Release Testing: Once the sprint ends, quality assurance team creates a new product release based on the work completed by many teams, and henceforth start testing the product release on staging servers. In case some features or user stories are not working as expected, bugs are created for the development team, which they will need to fix in the release branch provided they are blocker/critical bugs. Rest of the bugs are added to team’s product backlog.
- Release testing is very important because it is for the first time where sprint related work is tested on the staging servers, a setup which is similar to production release environment.
- Release testing helps to keep a constant check on product quality.
Sometimes bugs detected during release testing are not prioritized properly as far as their criticality is concerned. This means if lots of bugs are treated as critical, team will need to work on those in the current sprint in addition to tasks which were planned as part of sprint planning. This usually may lead to incomplete tasks by the end of a sprint.
Bugs detected in a particular release branch usually need to be fixed in that release branch, and in case there are more than one active release branches (e.g. release in production is also getting patches and a different release in the staging environment needs bug fixes as well). This often leads to software merging issues when team merges software components from two different release branches to the main trunk.
Current Product Development Process Chart
The current product development process being followed in the Applications Development Team within the case company is depicted by Figure 3
As a summary team is facing many challenges while using scrum as the product development methodology, and in certain situations team is switching back and forth to use the pseudo-Kanban approach which is not completely Kanban either. The good and the bad practices in the current process are further summarized in the next chapter.
Overall, the good and the bad practices of the current product development process within the case company are summarized in Table 6.
Table 6. Summary of best and bad practices in Scrum with respect to case company context
Best Practices of Agile Product Development Processes
This section discusses the main features of the product development processes based on different agile methods. According to Wikipedia (2014), “agile methods are focused on different aspects of the Software development life cycle. Some focus on the practices (For example, XP, Pragmatic Programming, Agile Modeling), while others focus on managing the software projects (For example, Scrum)”. Even though a lot of information is available on different agile methodologies, the researcher in this study only concentrated on agile methods which focus on managing the software projects. Accordingly, three main methodologies Scrum, Kanban and Scrumban were studied by the researcher. Further, this section also tries to compare these three agile methods based on the literature study and existing best practices.
Scrum is one of the most commonly used agile software development approach since last 10 years. “Scrum is founded on empirical process control theory, or empiricism. Empiricism asserts that knowledge comes from experience and making decisions based on what is known. Scrum employs an iterative, incremental approach to optimize predictability and control risk.” (Schwaber and Sutherland, 2013). In Scrum model, an organisation is divided into small self-organizing teams with sizes ranging from 4 to 10 people. According to the scrum practice, a scrum team should be self-organized and cross functional, and it should have all the needed competencies to accomplish the project without the need for external competencies. The Scrum framework consists of scrum teams and their associated roles, events, artifacts, and rules. Scrum prescribes four formal events Sprint Planning, Daily Scrum, Sprint Review and Sprint Retrospective. Further, the Scrum Team consists of a Product Owner, the Development Team, and a Scrum Master (Schwaber and Sutherland, 2013).
Every team has a product owner who is responsible for creating user stories based on high level customer/business requirements in a queue called product backlog. The product backlog is a dynamic list of requirements and is the only source for storing requirements and is maintained throughout the life cycle of product development. The product owner is the sole person responsible for managing the product backlog, and the responsibilities include clearly specifying product backlog items, prioritizing the items, ensure the backlog is clear to the development team, and direct the scrum team on what to work next (Schwaber and Sutherland, 2013).
In the Scrum model, product development is done in small iterations called Sprints, facilitated by a person called Scrum Master. Scrum master is responsible for making sure that the Scrum team lives by the values and practices of the scrum. Sprint iteration duration usually ranges from 2 to 4 weeks; however, the duration is fixed within a particular team or an organization. At the beginning of each sprint, teams have a sprint planning meeting where they usually commit to a subset of prioritized tasks from the product backlog. This subset of tasks is called Sprint Backlog and is managed during
the lifetime of sprint iteration. In the sprint planning session, team members try to estimate the effort needed for stories in the backlog. Scrum does not prescribe a single way for teams to estimate their work. However, it does ask that teams not estimate user-stories/requirements in terms of time, but, instead, use a more abstracted metric to quantify effort. Common estimating methods include numeric sizing (1 through 10), t-shirt sizes (XS, S, M, L, XL, XXL, XXXL), the Fibonacci sequence (1, 2, 3, 5, 8, 13, 21, 34, etc.). The product owner needs these estimates, so that he or she is empowered to effectively prioritize items in the backlog and, as a result, forecast releases based on a sprint velocity. The velocity of a sprint is the total number of story points completed within the sprint iteration. This means the product owner needs an honest appraisal of how difficult work will be. Even when the team estimates amongst itself, actions should be taken to reduce influencing how a team estimates. As such, it is recommended that all team members disclose their estimates simultaneously. Because individuals “show their hands” at once, this process is like a game of poker (Scrum Methodology, 2014).
During the scrum sprint, team members pull the tasks from the sprint backlog and start working on those. Everyday team members will have a short stand-up scrum meeting called Daily Scrum, which is facilitated by scrum master where every team member answers three main questions: What he has done since the last meeting? What he is planning to do today? Are there any impediments stopping him to do his work? Throughout the sprint, team members need to be self-organized and ensure that no impediments are stopping their work. In scrum, the whole team is responsible for completing the committed stories by the end of the sprint.
At the end of each sprint, a Sprint Review meeting is held to inspect the sprint targets and accordingly adapt the product backlog. In this meeting, attendees include scrum team, product owner and the key stake owners invited by the product owner, For instance, the key stakeholders may include top management, sales team, or customers. Team members demonstrate the completed stories to the product owner as well as key stakeholders and accordingly answer the question related to the sprint increment. The entire group collaborates on how to sprint can be improved and what can be done next so that valuable inputs are provided to the subsequent Sprint Planning. Usually, the result of the Sprint Review is a revised Product Backlog that defines the probable Product Backlog items for the next Sprint (Schwaber and Sutherland, 2013).
A Sprint can be canceled before the Sprint time-box is over. Only the Product Owner has the authority to cancel the Sprint, although he or she may do so under influence from the stakeholders, the Development Team, or the Scrum Master. A Sprint would be canceled if the Sprint Goal becomes obsolete.
After the end of a sprint, the last thing team members do is reflect on how things went inside the last sprint so that they can improve upon shortcomings and continue to do good things in the upcoming sprint. This is usually done in a meeting called Sprint Retrospective Meeting. The Sprint Retrospective occurs after the Sprint Review and prior to the next Sprint Planning. The main purpose of this meeting is to constantly look for improvement opportunities and retrospect on how the team can do things in a better way in the future. According to Schwaber and Sutherland (2013), the purpose of the sprint retrospective meeting is,
To inspect how the last Sprint went with regards to people, relationships, process, and tools; Identify and order the major items that went well and potential improvements; and, Create a plan for implementing improvements to the way the Scrum Team does its work.
In summary, the Scrum model provides a framework and tools for software development and believes in empowering the team to take its own decisions during small iterations called Sprints. The whole team works in collaboration and commits to tasks taken inside a sprint which they finally demonstrate at the end of a sprint, and at the same time team constantly try to improve by taking on learnings from sprint to sprint. Further, this model defines certain roles to facilitate and manage the work in a team.
Kanban model is known to have originated from the Toyota production system. Kanban is a Japanese word and literally means “signboard”. When used in manufacturing, it is a production control system, aimed to have just-in-time production and making full use of workers capabilities (Sugimori et al., 1977). It is claimed that Kanban is one of the important models which execute lean thinking in practice (Chai, 2008). Kanban system drives project teams to visualize the workflow, limit work in progress (WIP) at each workflow stage, and measure the cycle time (that is, the average time to complete one task) (Kniberg, 2009). Kanban has been used in manufacturing since decades however; it is relatively a new concept in the area of software development. Using Kanban model, the workflow in a software development project is visualized using a board called Kanban board. Kanban board usually is a whiteboard however recently many software/electronics tools are also used to represent Kanban board. On the Kanban board, work items (usually user stories) are represented using Kanban cards, the most commonly used Kanban cards are sticky notes. This board consists of several columns with each column representing a workflow stage of the development process. The number of work items in each column is limited in order to manage the workflow. As a result, developers concentrate on the work items in progress and try to complete those before starting working on new work items. When a work item is completed in a particular stage, it is moved to the next column and a result some other work item can be pulled from the previous column. A simple representation of the Kanban board is depicted in Figure 4
Kanban methodology is based on principles of visualizing the process workflow, limiting work in progress and controlling the lead time.
Process Visualization: means representing workflow stages on a board called Kanban board. It serves as a visual control mechanism, indicating how workflows though different stages of the software development process. This visualization is done by dividing the board into different columns indicating different workflow stages. Further, tasks or user stories are usually represented by sticky notes, which are moved from one column to another as the work progresses from one stage to another. Visualization helps the team to know the progress of each task. In addition, in some stage of the workflow becomes a bottleneck, it prompts the development team to handle it immediately (Mahnic, 2013).
Limiting Work in Progress: Kanban puts a maximum limit on the Work in Progress (WIP) items of every stage within the process workflow. Its ideology is based on the fact that something new shall not be started unless an existing piece of work is delivered or pulled by a downstream function. New work is only pulled into the system when there is a capacity to handle it rather than being pushed into the system from outside. This mechanism is called a “pull” mechanism. WIP limit defines the capacity of each stage of workflow in terms of the number of work items that can be in progress in a particular stage of the software development process. This pull based system ensures sustainable pace without overloading different stages of the workflow. Further, it significantly reduces lead time, which is used as a major measure of the development team’s throughput and productivity (Mahnic, 2013).
Measure Lead Time: Lead Time is one of the important metrics in Kanban a measures average time to complete one work item. In Kanban, the idea is to optimize the process to make the lead time as small and predictable as possible (Kniberg, 2009 Development team focuses on improving the lead time so that they can easily predict how many tasks can be completed within a particular period of time, and further improve the overall process so that lead time is reduced over the time, which ultimately improves the team’s output performance.
In Summary, Kanban focuses on visualizing the workflow, pulling the work when capacity allows it, improving the workflow process by limiting the amount of Work in Progress (WIP) items in different stages, and finally improving the process further in order to have small and predictable lead times.
Scrum-ban is the combination of Scrum and Kanban and tries to use features from both the software development models. On one hand, it uses the prescriptive nature of Scrum to be agile; on the other, it encourages the process improvement of Kanban to allow teams to continually improve their process (Pahuja, 2012). The term Scrumban was first used by Ladas (2008) in his whitepaper on ‘Scrumban-Essays on Kanban Systems for Lean Software Development’. On one hand Scrum model has helped the software development teams to self-organize, collaborate, improve efficiency constantly, work in small iterations, and avoid management overhead, applying lean methods like Kanban can extend these benefits. A lot of literature is available on the web related to Scrumban, and there are two thoughts of people; some apply Scrum to Kanban where the process is more inclined towards Kanban and others apply Kanban to scrum where the process is more inclined towards Scrum. Irrespective of this, both versions seem to take certain principles from Scrum and Kanban and accordingly adjust those to their organization/team needs and requirements. Following are the core principles of Scrumban:
Visualize the workflow: This is one of the most important tools taken from Kanban and applied to Scrumban. Visualizing workflow literally means team visualizes different phases their Product Backlog Items (PBIs) or stories go through starting from the sprint backlog and ending in the done phase, on a whiteboard. Even though there are lots of digital and online tools that can be used to visualize the workflow, many Scrumbanusers still prefer whiteboards because they are easy to manage and easy to change. In a normal Scrum, the team usually starts from the sprint backlog and works on those items and finally moves them to the done stage. However, in Scrumban idea is to visualize the flow of work into and out of the sprint (Yuval, 2012). Once the visualizing of workflow is achieved, it helps team including product owners to know the bottleneck areas within the workflow. Further, visualizing help in knowing who is working on what tasks, and what is progress state of different stories at a particular point of time.
Visualizing on a whiteboard or any digital tool is done by dividing the whole board into different stages represented by columns. And then digital PBI’s/stories or sticky stickers representing the PBI’s/stories are moved around those columns starting from the backlog stage on the left side of the board (or digital tool) to the done stage on right side of the board (or digital tool). A typical visual workflow representation in Scrumban using some software tool like Jira is shown in Figure 5.
An example of visual workflow representation in Scrumban using whiteboard is shown in Figure 6.
Pull Work: In Scrumban, the work is pulled as and when needed into a queue unlike the approach used in a traditional Scrum where all the work to be completed within a sprint is assigned in the beginning of the sprint to the sprint backlog. This backlog queue is different than normal sprint backlog because it can be updated many times within an iteration as and when required. According to Ladas (2008), one enhancement in Scrumban that can enable teams to decouple the process of assigning work from the process of prioritizing work is to add a ready queue in between the backlog and work in progress queue. This ready queue contains items that are pending from the backlog, but have high priority. In this queue, no tasks are bound to any individuals, but as soon as somebody becomes available, they should take one of these tasks instead of picking something out of the general backlog (Ladas, 2008). Ideally, ready queue shall not contain too many tasks which will ensure that the team focuses on just next prioritized PBIs. Also from one stage to another work is pulled to the next step once there is a capacity to do so. This is different from the scrum where work item is pushed to the next step in the process once some stage is completed (Strange, 2013).
Limit Work-In-Progress (WIP) Items: One of the important aspects of Scrumban is to apply limits to the work in progress items at every stage based on team capacity. In scrum context, it means limiting the Product Backlog Items (PBIs) that are in progress at any point of time, including the sprint backlog. The idea is to keep the team focused on completing work at hand rather than starting a new task. This means once a limit is reached within a particular stage of the workflow, rather than start working on something new, it is time to help someone else within the team. This will ensure that the team’s workflow becomes smoother and no stage becomes a bottleneck. Further, this is one of the important aspects of scrum-ban that helps the team to achieve real collaboration and a smoother workflow. Once a limit is reached in some stage of the workflow, team members must prefer helping others rather than starting a new task (For example, developers help developers, developers help testers, testers help product owners, etc.). Down the road, once team observes improvements in their capability and collaboration they can further tighten the limits to catalyze even more improvement in capabilities. Limiting WIP is a great way to drive real collaboration at the team level (Yuval, 2012). Setting limits also include setting multitasking limits for individuals; the team can have rules to like,
Prefer completing work to starting new work, or you might express that as a rule that says: try to work on only one item at a time, but if you are blocked, then you can work on a second item, but no more (Ladas, 2008).
Just that people can work on more than one task at a time does not mean everybody shall work on more than one task at a time. So it makes sense to sets limits to “Progress” queue in such a way that it allows for some team members to work on more than one task but not all. For instance, if there are five members in a team, a limit of 8to the Progress queue will disallow everyone to work on more than 2 tasks.
Make Team Rules Explicit: In tradition scrum, the idea is that teams are self-organized and they will work and coordinate themselves, however in practice there are always gaps between how a team shall organize themselves and how things are working out. “Self-organised teams cannot work if they don’t have a shared the understanding of how work is done” (Yuval, 2012). According to Business Dictionary (2014), the definition of policy is,
A set of policies are principles, rules, and guidelines formulated or adopted by an organization to reach its long-term goals and typically published in a booklet or other form that is widely accessible.
However, as per Heuvel (2011), policies in Scrum-ban/Kanban are;
Mechanisms, rules or processes that govern how a system works. By making these policies explicit, it becomes easy for others observing the system to understand it, and for those inside it to continually evaluate and improve the current mechanisms where necessary.
In Scrumban, team rules or in simple terms process is made explicit so that everyone in the team is empowered how to manage flow, self-organize and coordinate in order to achieve smoother workflow. Making team policies explicit will help the team members manage themselves, make quicker decisions without putting much effort into thinking, and even reduce the likelihood of giving in to special requests under stress. These policies related to the process are something what team decides upon, some policies can even be organization related. And mainly policies are trying to address recurrent situations, where someone needs to make a decision on how to proceed or what to do if such a situation arises.
Planning Meetings: Unlike Scrum, Scrumban has shorter planning meetings in order to update the backlog queue as and when required. According to Ladas (2008), “The planning can still happen at regular intervals, synchronized with review and retrospective, but the goal of planning is to fill the slots available, not fill all of the slots, and certainly not determine the number of slots. This greatly reduces the overhead and ceremony of iteration planning”.
A team shall always plan for shorter period ahead. Having longer planning meetings does not make sense in case the priorities often change. And also, since the team pulls work into a small ready queue before pulling it into work in progress, then from the product owner’s as well as team’s perspective iteration backlog shall just contain prioritized stories which the team shall work on next. Therefore, “the ideal work planning process should always provide the development team with the best thing to work on next, no more and no less” (Ladas, 2008).
In Scrumban, WIP limit is set to the sprint backlog queue as well. Therefore, assuming that in the planning meeting, the team pulls a fixed number of tasks to work on into the backlog queue. And in case the backlog queue is about to empty, the team can decide to have a next planning session, in order to fill the queue with the next prioritized list of tasks. This way the team is not filling the whole iteration capacity with tasks and planning meetings will be shorter. At the same time, product managers will be able to respond to changing requirements (or event-driven requirements) quickly compared to normal scrum where tasks are locked for the whole sprint duration. Further, according to Ladas (2008)
Scrum styled time-boxed planning usually provides a much bigger backlog than what is strictly necessary to pick the next work item, and as such, it is unnecessary inventory and therefore unnecessary waste. Once you’ve broken up the timebox, you can start to get leaner about the construction of the backlog. Agility implies the ability to respond to demand. The backlog should reflect the current understanding of business circumstances as often as possible. This is to say, the backlog should be event-driven. Time-boxed backlog planning is just that, where the event is a timer, but once we see it that way, we can imagine other sorts of events that allow us to respond more quickly to emerging priorities.
Therefore event-driven, time-boxed backlog planning, will make process leaner and introduce the ability to respond to demands faster. This is one of the major improvements in Scrumban compared to Scrum model.
Review, Retrospectives and Daily Stand-up meetings: These are the very important ceremonies Scrumban retains from Scrum. The review provides the team with the direct feedback from product owners and the team’s key stakeholders such as product managers and customers. Usually, customers or product managers prefer to have this meeting at regular intervals like in Scrum.
Scrumban Retrospective is the place where; the team can improve upon their team rules, improve overall process, constantly look for improvement opportunities and retrospect on how the team can do things in a better way in future, and refine ideas to experiment within upcoming iterations.
Daily stand-up meetings keep team members up to date on who is working on what, coordinate activities, know and manage impediments in order to keep the work flow smoother. Again, the main idea is to manage the workflow and use the stand-up meetings as a daily platform to remove impediments.
Metrics and optional estimations in scrum-ban: In Scrum, PBI’s are estimated using metrics like story points and a number of tasks taken into the sprint is done based on average team velocity of last few sprints. Even though velocity gives an idea to the product owners how many teams can complete within a sprint, and according he/she can prioritize and make release plans, the problem with this approach as mentioned by Gambell (2013) is
Often metrics are abused by managers and business stakeholders who want to unnaturally simplify a complex process into a one-dimensional number. Velocity, the amount of story points a Scrum team completes in a single Sprint, is such a metric that incentivizes lower quality at the end of a Sprint as a team scrambles to finish every last story they committed to. When the number fluctuates, as is common with a newer team, the stakeholders begin to question the outputs of the team and even the effectiveness of Agile itself.
Therefore rather than estimating each and every story within an iteration, why not divide the stories into similar size items. And accordingly, the team can decide on selecting the fixed number of prioritized tasks into the backlog based on a statistical analysis of the work items completed in past. Hence, Scrumban prefers metrics like cycle time and lead time over velocity calculation. According to Ladas (2009),
The lead time clock starts when the request is made and ends at delivery. Cycle time clock starts when work begins on the request and ends when the item is ready for delivery. Cycle time is a more mechanical measure of process capability. Lead time is what the customer sees. Lead time depends on cycle time but also depends on your willingness to keep a backlog, the customer’s patience, and the customer’s readiness for delivery. Another way to think about it is: cycle time measures the completion rate; lead time measures the arrival rate. A producer has limited strategies to influence lead time. One is pricing (managing the arrival rate); another is managing cycle time (completing work faster/slower than the arrival rate).
A team shall focus on the average size of items in the backlog. A statistical analysis of all tasks in the project can yield a mean cycle time and standard deviation, which can be very useful planning tool (add up the number of stories and multiply by mean cycle time.) for how many stories can be completed in a certain iteration. And if the cycle time is under control, based on average cycle time team’s capacity can be balanced against the demand, which in other words will control the lead time as well.
In nutshell, Scrum-ban is a methodology which makes scrum leaner and flow oriented. It empowers team, help them to collaborate and organize by utilizing Kanban tools like visual workflow board, WIP limits at every stage of development, team rules, focusing on improving cycle times rather than estimations, etc. Further, it makes scrum flexible towards change by my having shorter planning sessions, avoiding planning for the whole iteration, avoiding unnecessary estimations, late binding of tasks, pulling work than pushing, and all this is synchronized within important scrum ceremonies like sprint planning, sprint review, sprint retrospectives and daily stand-up meetings.
Comparison of different agile methodologies
The comparison of three main agile methodologies Scrum, Kanban and Scrum-ban is described in Table 7. Further, the data in this table is also composed of other references marked with a, b, c and d in case some other author has specified a different behaviour for that particular point. These references a, b, c, and d are defined at the bottom of Table 7.
Conclusion and summary of the framework
After analyzing the literature review, it seems that many problems faced by the Applications Development team within the case company may be fixed by choosing process based on Scrumban methodology. Researcher’s choice is based on the following reasons:
Scrumban framework is the Combination of Scrum and Kanban methodologies and uses good features from both the software development models. On one hand, it uses the prescriptive nature of Scrum to be agile (self-organized teams, self-improvement, constant information flow, etc.) and on the other hand, it encourages process improvement using Kanban.
Scrum ideology is based on spending less time analyzing and estimating work items that may end up as low priority on the backlog. This case is very much valid within the context of case company. The current scrum process is very rigid and Scrumban would provide flexibility to handle event-driven priority changes or handle important requirements which need immediate attention rather than waiting for the next sprint.
Even though Kanban alone may help the team to manage workflow by setting Work-In-Progress limits, and further improve team collaboration by setting the explicit work policies. But these are not the only problems the team is facing. Team needs a process which is iteration driven so that sprint releases are in synchronization with other teams. And in addition, scrum ceremonies bring in many additional benefits as evident from the current state analysis, so using scrum-ban is the most suitable process.
Scrumban is a framework where it provides the development team with best thing to work on next, no more and no less. As a result, the team can focus on what is important next rather than the whole iteration duration.
Scrum-ban also creates a good structure process-wise, Scrum is too strict approach and Scrum-ban gives you added flexibility of Kanban. Accordingly, the summary of the possible improved product development process based on Scrumban methodology is described by Figure 7
Further, the crux of new improved product development process, based on ideas and concepts used in Scrum-ban methodology is summarized below in 7 main points.
1. Pull work and Shorter Planning’s: As per Ladas (2008), “The ideal work planning process in Scrum-ban should always provide the development team with best thing to work on next, no more and no less”, therefore avoid planning for the whole sprint. Team shall have shorter planning session in order to fill the slots available in the backlog. These planning sessions will be time-boxed and event-driven whenever team capacity is available. For example, whenever there are few tasks in the backlog queue or the queue is about to get empty- the team can have next planning session to pull the tasks from the top of the product backlog.
2. Fixed sized backlog and new Ready Queue: Instead of Sprint backlog, the team will have a fixed sized backlog; basically, a backlog with a maximum story limits set. Further, in order to decouple the process of prioritization and work assignment a new queue called “Ready Queue” can be placed between the backlog and the progress queue. Ready queue shall contain the most prioritized set of requirements/stories which management wants the team to work next as soon as someone completes a task. Ideally, this queue shall not contain many tasks; this will ensure that the team simply focuses on the next prioritized tasks. And hence, new requirements or priority changes can be done easily to the backlog to ensure leaner process.
3. WIP Limits: As seen from the scrumban literature and literature review, Scrumban follows Kanban approach to maintain a flow around backlog by putting limits on Work In Progress Items (WIP’s), and accordingly manage the flow. Further, team shall prefer completing tasks in hand rather than starting the new work. This means every stage of the development process will have a limit in terms of a number of work items that can be put to the BACKLOG/TO-DO state, READY state, PROGRESS state, REVIEW state and even TESTING state. By enforcing WIP limits, the process will ensure that bottleneck stages are cleared immediately and tasks do not get stuck in a particular state for long.
4. Explicit Work Policies: Team need to ensure that work policies or rules are explicit in order to improve collaboration. Team shall have clear definitions of done, how to collaborate teamwork when some stage becomes a bottleneck, rules followed in different stages of the workflow, team rules related to intra-team or inter-team communication that will help to manage the workflow, how to improve communication and further work policies related improving quality of work.
5. Continue Daily Scrum, Sprint Review, Sprint Retrospective: Team shall continue to use scrum practices like daily morning scrums, sprint retrospective, sprint review meetings as they constantly help to manage impediments, improve the development process and help to get direct feedback from product owners and other key stakeholders.
6. Avoid estimations: Team shall have PBI’s as small as possible. The best is to have similar sized items which can be completed within the similar if not the same amount of time, such as 1 day or 2 days or 3 days. Once the team starts getting better in creating tasks based on an average size of a backlog item, estimations won’t be needed because product owners can guess release times by simply multiplying a number of tasks taken into the iteration with the average lead time.
7. Focus on controlling Lead Time rather than Velocity: Velocity incentivizes lower quality. In case the team cannot complete tasks on time, they may hurry up things in order to maintain the velocity score. Estimations are good, as long as they mean something to the product owners; most often management uses velocity (which is calculated based on story point estimations) to judge the team’s performance. Better is taking the concept of velocity out, focus on ensuring smoother workflow and controlling the average lead time of a story. Not only will this reduce the stress levels of the team, but it’ll also improve the motivation of the team as well. Further, in normal Scrum, when a team focuses on velocity, they mostly prefer selecting new tasks into the sprint rather than bugs, the main reason being no story points are usually assigned to bugs. When a team will not use velocity for filling the sprint capacity, they will start looking into the bugs as normal tasks and hence quality of code will also improve by fixing bugs.
Building a new product development process proposal
New product development process proposal overview
As discussed in the current state analysis, the current process which is loosely based on scrum is not good enough to solve problems faced by the team under discussion within the case company context. This team needs a process which can handle event-driven requirements as they often are changing within a sprint. Further product management in this team has more pressure to handle urgent requirements from customers and sometimes within the same sprint in order to reduce “go to market” time for certain custom features. Not only that, there are other issues this team is facing related to collaboration, communication and process flow, etc. which were evident from the current state analysis. Therefore, as discussed in the literature review, the product development process based on scrumban can help the team to overcome many such issues. Based on the “summary of the possibly improved product development process” as discussed in the Literature review, section 4.5, qualitative interviews were conducted with many stakeholders to get their feedback and fine-tune the new process model. The details of the data collection are described in Table 3. The theme of the semi-structured qualitative interview is presented in Appendix 1, and the data collected based on those theme-based interviews is presented in Appendix 2. Not all the interview questions described
in Appendix 1 were relevant for all the interviewees. Based on the expertise and skills of interviewees, the researcher asked a different set of questions from different interviewees. After a few interviews, as the knowledge of the researcher increased with respect to the new process and also responses from interviewees started becoming familiar, the researcher added more questions to the theme interview template in order to further understand the problem deeper. The qualitative data was analyzed in various stages: reading and listening to interview transcripts, spending time with data and summarizing it, extracting the ideas and concepts and finally looking for some patterns in the data and how they fit in with respect to the theories (Mayor & Blackmon 2005, 348-49). Based on the data collected from the interviews and the best practices from the scrumban methodology, the new product development process was developed.
New product development process proposal
This chapter will focus on creating the new product development process based on the literature review and feedback received during interviews with different stakeholders and experts within the case company. The details on how the data was collected are described in Table 3, and the summaries of the data collected are described in Appendix 2. The new software development model will be based on Scrumban methodology. In the new process model, some good practices were retained from the current process, and the new process model is further described in different stages below: Pull process and planning meetings: In the Scrumban model, team will start using shorter planning sessions in order to fill the slots available in the backlog. Unlike a sprint backlog, Scrumban backlog may be updated more than once within a sprint. In Scrumban, as described in the literature review, the idea of having iterations is optional. However, after having discussions with different key stakeholders, it was clear that team will continue to use iteration duration of two weeks so that the Sprint demonstrations and Sprint Retrospectives can be held in synchronization with other development teams. In addition, using the iteration model will also simplify the creation of main software release by the end of every sprint. Further, most of the interviewees were in favor of sprint iteration. Lauri Oherd mentioned on 21 February 2014 that “sprint cycles helps the team to focus; otherwise developers would lose the sense of time.”
One of the interviewees, Product Owner of the team, Mika Mannermaa mentioned on 17 February 2014 that, “It is better to have shorter pull sessions than a mammoth sprint planning sessions”. At the beginning of each sprint, team will have a first planning session to fill the slots available in the backlog queue. The size of the backlog queue will be limited to twelve (12) and once the team is about to run out of tasks, it will have another planning session to pull the tasks from the top of the product backlog and fill the available capacity.
Sprint Backlog and Ready Queue: As mentioned in the literature review, one of the ideas in Scrumban is to put a new queue called ready queue in between the backlog and process queue. After having an interview and team discussions, a simpler idea came out to have smaller fixed backlog act as a ready queue itself rather than adding a separate queue in between the backlog and progress queue. A Quality Assurance team member, Ashish Mahindroo mentioned on 19 February 2014 that “working on a smaller list of prioritized tasks will definitely give the flexibility to include changes in priorities within a sprint. So why not have our backlog act as ready queue”. In this way, the team will be only managing one queue and whatever tasks are put in the ready queue (aka backlog) are the ones that the product owner wants team to work on next as soon as someone completes a task. At the same time, the product owner can change the priorities of tasks in the product backlog. The size of the ready queue (aka backlog) was preferred to be limited to twelve (12), which is just double the size of team members, in order to ensure that not too many tasks are put to this queue; this will ensure that product owners can plan anything urgent or important requirements from customers or top management to the next ready queue update within the sprint. In the scrum model, this was not possible as tasks could not be added in the middle of the sprint. As one of the interviewees, Edward Karvinen mentioned on 5 February 2014 that “we cannot be customer oriented if we are having a too strict process. So, our process shall be able to handle the customer driven requirements and their immediate need”. The leaner backlog will ensure that the team focuses on what is important and what needs to be completed next. For product management, this gives added flexibility in a controlled way and hence this new process will be able to handle immediate customer needs within the framework of this process rather than breaking it.
WIP Limits: In Scrumban, maintaining the workflow is important and that is why Scrumban applies Kanban limit to each and every development stage of the workflow. The different stages of workflow with respect to the team under discussion are “To-Do”(aka Ready queue), “Progress”, “Review”, “Testing” and finally “Done”.
For the “Progress” stage, most of the interviewees agreed that it does not make sense for a team member to work on more than one task at a time. Ex-Scrum Master of the team, Maksim Luzik mentioned on 14 February, 2014
One task at a time is good, technically one can never work on more than one task at a time, but there are situations when someone needs to work on more than one task. Accordingly, we shall give flexibility on working more than one task but not to have that as a general practice. In case a developer started working on another task and he already has an existing task in progress, I would prefer moving the old task back to the TO-DO list.
Based on the interview discussion analysis, a rule of (2xN – 2) was selected to limit the number of tasks inside the Progress queue; here N implies the number of developers within a team provided N is greater than 1. This way if we have 5 developers in a team, task limit for Progress queue will be (2×5 – 2 = 8) eight (8).
For the “Review” stage, this is the stage when a team member completes a task, tests it himself and then sends the code changes for review. For reviewing the code (it may even be a documentation task), one of the rule that was implemented by the team is that every developer will make a patch for the code to be reviewed and attach it to the corresponding task in the task management tool, Jira. Once the task is put to the review stage, some other developer who is available will review his task. One of the interviewees mentioned, “It is good to have some limit for the review stage, though sometimes there might be exceptions”. Many interviewees agreed that setting some limit will definitely ensure a smoother workflow. Others were strict in their approach to solve the problem of tasks getting stuck in the review stage. One such interviewee, ExScrum Master, Maksim Luzik mentioned on 14 February 2014,
Whenever a team member completes a task, he shall give first priority to the tasks in the review column, second priority goes to the blocker bugs and then third priority is to take tasks from TO-DO list. The main problem is, developers are not fond of doing something like reviewing a task or testing somebodies task, so as a last resort setting rule will help. If we need to go a bit strict, we shall not have the tasks in the review of more than half the number of developers. So if we have 6 developers, the limit shall be something like 3. The idea shall be that there is at least one developer to review a task.
Another interviewee, Sr. Software Developer of the team, Perry Mitchell mentioned on 21 February 2014, “With review limit does not need to be too strict. If needed, team can put the limit equal to the number of developers”. After more data analysis, in the “Review” stage, a rule of (N – 1) was selected to limit the number of tasks inside this queue; here N implies the number of developers in a team provided N is greater than 1. This way if we have 5 developers in a team, task limit for “Review” queue will be (5 – 1 = 4) four (4).
Once the task is reviewed, the developer will move the task forward to the “Testing” stage. At the same time, the reviewer will ask the developer to commit the task to the main code base called “Trunk”. In the “Testing” stage, Quality Assurance (QA) will test the story/task based on requirements and acceptance criteria, and accordingly move it further to “Done” stage if the task is working as per requirements or back to “Progress” stage in case some bugs or problems were found. One of the main problems the team has been facing as discussed in current state analysis was tasks getting stuck in the “Testing” phase at the end of the sprint. Because of the commitment pressures, the team members sometimes hurried up and moved the tasks to the “Done” stage without testing all the possible test scenarios. Sometimes, when the tester was unable to test all the tasks, they were left in the “Testing” stage when the sprint was over; indicating either the team had overcommitted or they had underperformed. So, managing the work flow around “Testing” stage is very important and these problems were discussed with all the interviewees. Many interviewees were of the opinion to set a maximum limit to the “Testing” stage as well. However, when it comes to handling the situations when testing becomes a bottleneck, opinions differed. One of the interviewees, Maksim
Usually, testing is the responsibility of a tester. In case the tasks are small, it won’t be a problem for a tester, but in case they are big then too many tasks in the testing stage becomes a problem. Automation of tasks might help partially but I do not think we shall have a situation when the number of tasks in testing is equal to the number of developers in a team, it shall be always less than that number.
Another interviewee Team Leader, Edward Karvinen mentioned:
As long as QA ensures that tasks are tested in staging and production, the local testing could be supported by developers if it becomes a bottleneck within a sprint. However, work shall be done in coordination with testers and it shall be done in rare cases. Further, those tasks shall be marked as tested by developers so that the testers are aware that they might need to focus a bit more on those
tasks in the release branch.
Therefore, a more favorable opinion was to set a limit of less than a number of developers within a team and in case of the maximum limit is reached. Further, in order to avoid bottlenecks in the testing stage, many interviewees were of the opinion that other team members can help the Quality Assurance testers with testing someone else’s task under the guidelines of the Quality Assurance team. Accordingly, a limit of (N – 1) was selected to limit the number of tasks inside the Testing queue; here N implies the number of developers in a team provided N is greater than 1. This way if we have 5 developers in a team, task limit for Testing queue will be (5 – 1 = 4) four (4).
Explicit Work Policies:
In order to improve collaboration within and outside the team, it is important for a team member to know what to do in certain recurring situations where he otherwise has no way to decide upon himself. In traditional Scrum, the process assumes that team will be self-organising and will address their issues and impediments whenever they happen. However, often teams struggle to achieve this in the real world and it leads to miscommunication, delay in delivery, weaker collaboration and decision making. Setting explicit work policies empowers the team members to handle decisions with ease and less stress. The team under discussion had some basic unwritten team rules, however, after interview discussions, many more rules were made explicit in the task management tool used by the team and further those rules were improved in the sprint retrospectives.
Most of the interviewees were in favor of making the team rules explicit. In addition, some were favoring setting explicit rules on what to do if some stage becomes a bottleneck, such as one of the interviewees mentioned, “Whenever a team member completes a task, he shall give first priority to tasks in the review column. The problem is the developers are not fond of reviewing a task or testing someone else’s task”. Other interviewee had strong opinions on splitting a task that will help the team in solving over-commitment problems. He mentioned, “It makes more sense to split the tasks into smaller ones”.
After the interview data analysis and Definition of Done (DOD) review meeting headed by the Quality Assurance team, the work policies defined by the team under discussion are mentioned below:
- Limit for Tasks in PROGRESS stage: 2*N-2 (N is number of developers)
- Limit for Tasks in REVIEW stage: N-1 (N is number of developers)
- Limit for Tasks in TESTING stage (N is number of developers): N-1
- No new tasks can be added to any stage if it is already under its maximum limit. In these situations, team member shall first work on task from that particular queue rather than taking a new task from the backlog.
- If the “Testing” queue reached its maximum limit, after completing the existing task, the team member shall work on tasks related to the “Testing” queue. This shall happen under the supervision of the Quality Assurance team who must provide some guidelines for testing different tasks.
- While planning, always split a task if work estimations indicate that task can span across sprints.
- Requirements/tasks shall only come from the Product Owner (PO) or agreed with PO.
- Stories are written and prioritized by the product owner
- Improve the work policies over time using sprint retrospectives as a place to experiment.
- If the feature in production is not working properly, a new bug should be created rather that reopening the corresponding task. A deployed feature must never be reopened.
- All the bugs found in staging and production must be reported to task management tool.
- Blocker bugs will be added directly to the ongoing sprint.
- Blocker bugs must be prioritized by the developer before taking a new task from the backlog.
- Any critical bugs found in the system shall be informed to the Product Owner (face-to-face).
In addition to work policies mentioned above, every developer within the team shall follow the common DOD for each and every story. Definition of done handles rules relating to test case coverage for tasks, review rules and who would approve the tasks. Most of those rules are companywide rules and are further improved for each and every team.
Scrum Ceremonies: In Scrumban, most of the ceremonies related to Daily Stand-up Meetings, Sprint Retrospectives and Sprint Demonstrations are retained from Scrum. This also became evident from the interview discussions where almost all the participants favored those.
One of the interviewee, Product Owner (PO) of the team, Mika Mannermaa, mentioned on 17 February 2014 that “Sprint demonstration is a very important aspect of the current process as far as PO’s are concerned. This is where PO’s know how team has performed when it comes to sprint goals”. Another interviewee, Head of the Software Development, Timo Valtonen, mentioned on 31 January 2014, “Reviews and other Scrum ceremonies like daily stand-up meetings, sprint retrospectives are working nicely in the current process”
Therefore, the team will continue to use daily stand-up meetings, which normally lasts for fifteen minutes, and every member in the team must answer three important questions such as, what they accomplished yesterday, what are their plans for today and are there any impediments related to their work. One of the interviewees, Lauri Oherd on 21 February 2014 mentioned, “daily stand-up meetings keep me focused as the everyday morning I have to answer what I was doing yesterday and it feels good in case I managed to complete my task”.
Regarding Sprint Reviews, they will also continue as usual, every two weeks by the end of the sprint, team will demonstrate what stories and tasks they were able to achieve in the last sprint. One interviewee, leader of the team mentioned, “Reviews are very good because Product Owners, sales guys and even other teams get to know what we accomplished within a sprint”. About sprint Reviews, Ex-Scrum Master mentioned, “Sprint demonstrations are quiet OK in our team, and it acts as what is coming next to sales and Product management, and it also acts as feedback for the team from the key stakeholders, whether things are done as per expectations”.
For Sprint Demonstration, Product Owners and the key stakeholders invited by them such as Technical Architects, Product Management as well as Sales teams join the demonstrations. This helps team to get direct feedback on how the sprint performed and at the same time, Product Management can plan the main release features based on sprint demonstrations.
After Sprint Demonstrations, the team will have sprint retrospective discussions as it
used to happen during old scrum process. However, there were some good suggestions during the interview discussions on how to improve the retrospectives. Ex-Scrum Master of the team mentioned,
Retrospectives shall be used to improve only your own team’s process rather than outside. They are just like therapy sessions to release frustrations about what went wrong and what can be improved but you can rarely change something in sales sides or anywhere else except your team. Grouping the action points and voting will help which action point’s team shall concentrate on and rest action points can wait until next retrospective.
Quality Assurance leader, Erwann Cleudic on 12 February, 2014 mentioned,
Teams shall invite Product Owners to the retrospective discussion; they are the first class team members and they are the ones who know how a team performed based on sprint expectations, so their feedback is vital.
Based on different suggestions and ideas from interviews, team will only try to handle most important action points from retrospectives, rather than focusing on all. The reality is no matter how many action points you come up with, you cannot concentrate on all of those, so team will have to make a choice what action points to implement next. For this purpose, once every member suggests what needs to be improved in next sprint, those action points will be prioritized based on team votes and only most important action points will be addressed in the next upcoming sprint. Further, Product Owners shall also be part of the retrospective discussions, they are part of the team and it will further try to bridge the gap between team and the management. In addition, Product
Owners have valuable information on how team performs from a business perspective and even they can handle any action points where management comes into the picture.
Estimations and Metrics: As discussed in the literature review, in the Scrumban, the team shall have smaller and perhaps similar sized backlog items. And once the team becomes better is creating stories based on average size (Average Lead Time), storyestimations may not be needed at all. In practice, it is often difficult to split a story oreven roughly estimate the story. One of the interviewees, Team Leader, mentioned, “It makes more sense to split the tasks into smaller ones. Sometimes, people are lazy to do that and things keep on dragging for many sprints, however splitting a task doesn’t affect productivity because there are instances when it doesn’t make sense to create too many tasks if only one guy is working on that, it even takes time to create tasks in tools”. Another interviewee, Ex- Scrum Master mentioned, “Splitting bigger tasks will always help, but it is difficult to split a task. Further, there are task dependencies which act are blockers while splitting a task. One task cannot be started before completing the previous one. So, I would say it is often tricky to have a similar size of tasks”.
So in situations where there is a big story which is split into smaller stories and in addition tasks are interdependent, only one person can work on those tasks. Therefore, itmay not make sense to divide this story into smaller ones because of no matter what only that person can work on the next task. Another interviewee, Head of Software Development, Timo Valtonen mentioned on 31 January, 2014,
It is very important to split a story which is estimated more than 8 story points. From the past sprint data, often stories with more than eight (8) story points continue in the following sprint which is not good when it comes to sprint commitments. Further, this gives incorrect visibility to Product Owners
So it may be difficult to split a story into smaller ones but at the same time it is also important to split a bigger story for providing better visibility and achieving sprint commitments. Often the process of estimating the size of a task is a very difficult job, even though people may estimate tasks, but at the end of the day those are just estimation and may or may not be correct. That is one of the reasons why many Scrum practitioners suggest using abstract methods to quantify effort such as estimate the size of tasks in story points based on Fibonacci series or sizes like XS, S, M, L, and XL. Estimationis a difficult task and it may only get better with years and years of experience in that
particular technology (Scrum Methodology, 2014). One of the of the interviewee ExScrum Master mentioned,
The problem is estimation is a very difficult thing to do, so often team overestimates or underestimates tasks. Under pressure teams often underestimate. However rough estimation is still good, it still gives some ideas to outsiders (Sales and Product Owners) how big a task is and how long it may take
Most of the interviewees were in favor of estimating the rough size of the tasks. However, many agreed that it might be difficult to split tasks into similar sizes. At the same time many agreed that it is good to split bigger tasks for more visibility rather than dragging a task around many sprints. Therefore in Scrumban process team will continue to do estimation of tasks based on story points as was done in an earlier process. In addition, team shall try to split the stories which are estimated more than 8 story points.
As described in the literature review, measuring velocity is optional in Scrumban because some practitioners prefer story estimations and others do not, and often it depends upon whether they follow Scrumban which is more inclined towards Scrum or Kanbanrespectively. One of the interviewees, Team Leader mentioned,
Velocity does give some perspective on what can be done in next sprint but often estimations are not accurate as often research work related to tasks is difficult to estimate.
Another interviewee, Perry Mitchell mentioned on 21 February, 2014 that “Velocity is often misused and team may even inflate the estimations to get better velocity. Secondly, there is too much interest from the management side regarding velocity.” Most of the interviewees preferred story estimations and agreed that velocity helps product owners in understanding how much a team can accomplish within a sprint. Many of the interviewees also agreed that velocity shall not be used by management to determine team’s performance. Ex-Scrum Master of the team mentioned that “People can inflate story points just to get more velocity, so it may not be a good metrics if management is using velocity for determining the performance of the team”. There can be sprints where team has to fix many important bugs that might impact their velocity, but it does not mean they performed badly. Average Team Velocity shall only be used by Product Owners to roughly estimate what team may accomplish within a sprint and certainly not what team must accomplish. Since story points are associated with new tasks only that is why team was focusing on new tasks rather than bugs during the sprint planning sessions. As a result, based on average team velocity, team had a tendency to fill the sprint backlog slots with new tasks. This was also mentioned by ExScrum Master of the team, “Partially velocity is a reason that bugs get deprioritized”. Therefore in the new process team will continue to measure Velocity but it would not be used as a direct means to fill the sprint backlog capacity. Accordingly, during planning meetings, new tasks would not be preferred over bugs because they have story points associated. Team shall choose based upon what is important that needs to be completed next, and it can be either a new story or an existing bug.
Although Scrumban prefers measuring Average Lead Time (or Average Cycle time), this approach will only work in case all the stories in the backlog queue are roughly similar sized. It is difficult to have similar sized stories in new product development especially with respect to the team under discussion because many tasks need some research work as well. Also many of the interviewees mentioned that it will be difficult to create similar sized tasks when requirements are changing and scope of the task is not clear. Accordingly, in the new scrumban process, measuring Average Cycle time will be optional but it may give some information in the long run about the average time to
Product Backlog Grooming: means constantly improving the quality and priorities of stories in the product backlog. In Scrum, product backlog grooming is done by the Product Owner and in scrumban as described in the literature review, product backlog grooming is optional and it assumes that product backlog is prioritized. However, the problem is if the product backlog grooming is out of process, pulling stories from the product backlog into the sprint backlog queue will become difficult. Further, as described in current state analysis in case Product Backlog is not prioritized, choosing the right stories for sprint backlog becomes difficult and sprint planning might end up planning priorities rather than filling available backlog slots. Also, if stories are not well defined in the product backlog, it makes it difficult for team to estimate stories. Often the scope of incomplete stories changes inside a sprint and this leads to delays or incompletion of other stories. One of the interviewees, team leader mentioned that “we shall not stop working in case stories are incomplete and rather we shall deliver such story in increments but at the same time management shall understand such tasks will take time”. However, most of the interviewees preferred to have well-defined product backlog grooming process. Head of Software Development teams, Timo Valtonen mentioned, “Majority of the problems within the team will be solved in case Product Backlog is prioritized and the quality of stories is improved”. Another interviewee Ex-Scrum Master of the team was also having a similar opinion. He mentioned, “Product Owners and product managers shall groom the backlog in order to ensure prioritization, in this way when team will be pulling the tasks from product backlog they shall be prioritized already”. Product Owner of the team, Mika Mannermaa, mentioned on 17 February, 2014 that
Product Owners shall ensure that the product backlog is in order; we have been lacking
this lately. We need to break big epics into smaller stories and therefore we must have
weekly grooming sessions with Product Management as well as sales team. Further, we
need to synchronize backlog grooming sessions ahead of sprint planning meetings.
After analyzing the current state and all the ideas from interview discussions, it became evident that product backlog grooming has to be part of Scrumban process, especially with respect to the product development process of the team under discussion. We cannot exclude Product Backlog grooming or assume that our product backlog will be in order without making this activity as part of the process. One of the good ideas was to have the weekly product backlog grooming which involves the Product Owner, Team Leader or any needed team member based on technical expertise required for different stories. This group will do the backlog grooming which includes improving quality of stories, a rough estimation of stories and prioritization of stories so that top of the backlog is always in order and ready to be pulled into the Ready Queue. Further, Product backlog grooming will be done ahead of the beginning of next sprint planning meeting (i.e. before backlog/ready queue update meeting, and the first update meeting always happens just after the sprint demonstration is over). The process of Backlog Grooming which will be part of the new development process is described in Figure 8:
Figure 8. Product Backlog Grooming added to the Scrumban process
Prototype Proposal for Product Development Process
The new software development process proposal (prototype) is depicted in Figure 9, and further briefly summarized in 8 main points.
1. Planning Meetings: In Scrumban model, the team will start using shorter planning sessions in order to fill the slots available in the backlog. The ideal work planning process should always provide the development team with best tasks to work on next, no more and no less. Within a sprint, once the team is about to run out of tasks they will have another planning session to pull the tasks from the top of the product backlog and fill the available capacity.
2. Backlog and Ready Queue: Unlike Scrum, the team’s new backlog will be fixed sized backlog and may be updated many times during a sprint iteration based on available capacity. This backlog will also act our Ready Queue in terms of Scrumban Process as it will only contain the prioritized list of work items that team members must work next. The team has enough development stories to work on in the Ready Queue, not too many not too less. This will help the team to reduce long planning meeting and focus on prioritized tasks. At the same time, new requirements can be brought in making it more agile than Scrum. For product management, this gives added flexibility in a controlled way and hence this new process will be able to handle immediate customer needs within the framework of the process rather than breaking it.
3. Work In Progress (WIP) Limits: The team needs to maintain a flow around backlog and different stages; therefore, they will limit Work In Progress Items (WIP’s) for each and every stage and manage flow. The team shall prefer completing tasks in hand rather than starting new work. This means every stage of the team’s development process has a limit in terms of number of work items that can be in TO-DO state, PROGRESS state, REVIEW state and in the TESTING state. This will ensure that our tasks do not get stuck in a particular state for long or there are not too many tasks in a particular state as team will need to enforce limit WIP policies.
4. Explicit Work Policies: Making the work policies explicit will help the team members to collaborate better within the team as well as outside the team. In addition, it will help in managing the workflow around different stages. The team will have explicit work policies in terms of WIP Limits, maintaining quality, how to collaborate when some stage becomes bottleneck, written and clear definition of done, how bugs and priorities will be handled within team, who can create stories, communication policies, and who will handle external communication. All those policies were described in chapter 5.2
5. Scrum Ceremonies: In Scrumban, most of the ceremonies related to Daily Stand-up Meetings, Sprint Retrospectives and Sprint Demonstrations are retained from Scrum. All these ceremonies constantly help to manage impediments, improve the development process and help to get first hand feedback from product owners, managers and customers. In order to improve the sprint retrospective further, the team will only choose most important action points based on voting rather than focusing on all. Further, the Product Owners will also be part of the retrospective discussions. They are part of the team and they have valuable information on how a team performs from a business perspective. Product owners can even handle any action points where management comes
into the picture. This will further try to bridge the gap between team and management.
6. Estimations: In Scrumban process, the team will continue to do an estimation of tasks based on story points as was done in an earlier process. Unlike Scrumban, the team may not be able to split tasks into similar sizes; however, the team shall try to split the stories which are estimated more than 8 story points.
7. Metrics: In the new process, the team would continue to measure Velocity but it will not be used as a direct means to calculate number of stories needed in sprint backlog. Accordingly, during planning meetings new tasks will not be preferred over bugs because they have story points associated, team shall choose based upon what is important that needs to be completed next and it can be either a new story or an existing bug. Also, measuring Average Cycle time will be optional but it may give the team some information in the long run in case it measures the average cycle time of stories and bugs within different sprints.
8. Product Backlog Grooming: Even though optional and out of scope in Scrumban, product backlog, grooming will be part of the new software development process. Product Management including team leader or the needed team members will do a weekly product backlog grooming, usually ahead of Ready Queue update meetings. This will ensure that correct priorities are set to the
backlog, quality of stories is improved and most importantly work items are at the top of priority list. As a result, team can pull stories from the top of the Product backlog to Ready Queue as soon as there are not enough stories in the ready queue. The flow of product backlog grooming within the new process is also depicted in Figure 8.
Pilot Testing of the proposed New Process
The main purpose of the pilot testing is to ensure the feasibility of the New Software Development Process. The efficiency of the new process can be verified by checking the impact of the new process on the team’s velocity and further analyzing the team’s retrospective discussions whether it has improved the teams day to day functioning in any way.
Pilot Testing Overview
The new product development process described in Chapter 5 was presented to Applications Development Team as well as other important stakeholders like Product Management, Quality Assurance Team and Head of Software Development, using power point presentation on February 25th, 2014. The whole session lasted for around two hours and included a presentation on findings from current stage analysis, describing new process proposal and why process based on scrumban might address various problems described in current state analysis. After the presentation, the Head of the Software Development and Product Management agreed to test this prototype process in the Applications Development team for four sprints, starting from March 4th , 2014 until April 29th, 2014. After every pilot sprint, the process could be improved depending upon the discussions in the team’s sprint retrospectives. Once the results of this new prototype process will be concluded on April 29th, 2014, the next decision will be taken further whether to continue with this new process or not. The timeline indicating different milestones and stages in the piloting of the prototype process is shown in Figure 10.
The applications development team which was going to pilot test this new process consists of six team members, including five developers and one Quality Assurance member. Two of the developers were having dual roles of Scrum Master and Team Leader. In addition, team had a dedicated Product Owner and User Interface Designer. In order to test the new process, there were no changes in the team roles; team’s Scrum Master was driving the new process. Sprint duration was set to 2 weeks same as before, and the maximum limits for the different queues were set as described by the Table 8.
Table 8. Work In Progress Limits for different Queues in new prototype process In addition, the new prototype process model was well documented and listed on team’s intranet pages of the case company.
Pilot Testing Analysis
Impact on Team’s Velocity: One of the keys metrics than can used to analyse whether the new process has improved the team’s ability to deliver user stories is Team’s Velocity. Team’s velocity indicates the number of story points completed with a sprint and the average of those can be used to determine whether the overall velocity has increased or decreased. The data regarding the velocity of four Sprints (piloted) was collected from team’s task management tool (Jira) and is described in Table 9. In this table, Velocity per team member is obtained by dividing the sprint velocity with total
number of team members in the team.
Sprint Velocity per Team Member = Sprint Velocity/No. of team members
As seen from the Table 9, team has a better velocity in the first sprint of Scrumban compared to other sprints. The drop in sprint velocity from second piloted sprint onwards is mainly because the case company reduced the number of team members in the Applications Development Team. The team size was reduced because of reasons which researcher cannot explain in this thesis, as the information is company confidential. Initial size of the team was six members, however after first piloted sprint; the size of the team was reduced to 4 which included three developers and one Quality Assurance Team member. Further, the drop in the team velocity could also be attributed to
the fact that team was no more filling the backlog capacity based on team velocity, the selection was purely done based on Product Owner’s priority and that included bugs as well as new stories. Accordingly, in the four piloted sprints, there is a decrease in the Average Team Velocity from 40.65 to 32 compared to team velocity of the last twenty-three sprints. However, if one looks into the Average Velocity per team member and compares it with the data from last 23 sprints, it has increased in the four piloted sprints from 6.77 to 7.0, which is 3.32% increase in average velocity per team member.
Sprint Retrospective Analysis:
Being the Scrum Master of the team, the researcher also collected and analyzed the sprint retrospectives of piloted sprints. From all those sprints retrospectives, the data collected related to the positive points and the things to be improved in every sprint, is listed in Table 10. In this table, however, some team-specific points, which were not related to the overall new process, have been omitted by the researcher.
Analysis of First Piloted Sprint: From the retrospective data of first piloted sprint, it isclear that team appreciated having shorter planning sessions in order to update ready queue capacity, and all the meetings were short and to the point. Accordingly, the researcher can conclude that the new process has helped the team members to focus on what is important next, and provided added flexibility by planning for shorter periods rather than the whole sprint at once. At the same time, workflow across different stages has improved, and there was hardly any development stage that became a bottleneck. With the earlier process, team was often struggling with this issue. In the new process model, every stage has a maximum limit and team members ensure that they need to process the queue which has reached its maximum limit, before continuing with newer tasks.
In addition to positive points from the retrospective, team members discussed on improving handling pressure situations from the top management. Since management wanted some important customer-requested changes to be completed in the hurry, some additional tasks were added to the ready queue than the permitted capacity. Accordingly, it was agreed by team not to fill the ready queue beyond its maximum limits until its size is improved further. In addition, team agreed on adding on more rules to the explicit team policies. This new rules were related to improving code quality, and says “Before any code check-in, use Js-Hint rules for improving code quality”. This rule will further help the team to improve the quality of code, and every team member has to ensure that they will follow this rule before adding or modifying any new code to the existing code base. This is the best example of how team started to use scrumban retrospectives to improve the team policies.
Analysis of Second Piloted Sprint: The notes from the second sprint retrospective were also positive as far as new process model is concerned. There were no bottlenecks, planning meetings were short, and team members were helping each other in case the maximum limit was reached in some development stage of the workflow. For instance, developers started helping quality assurance guys in reducing the testing queue size once the queue reached its maximum limit. This kind of team collaboration was missing in the earlier process, as team members were only focusing on their individual tasks. Setting the limit on queues and making work policies explicit has helped the team to improve collaboration. In addition, team members discussed that Product backlog grooming shall happen weekly and product owners shall take the responsibility of reserving time slots.
Analysis of Third Piloted Sprint: From the retrospective data of third piloted sprint, team realized the power of focusing on workflow rather than focusing on team velocity. Using the new process framework, team was no more having sprint planning sessions which filled the whole sprint backlog based upon the team velocity. Rather, team was simply filling the available backlog queue capacity with new stories as well as bugs; this is a significant change to include bugs into the sprint as well. In addition, ready queue updates within a sprint gave flexibility to the Product Owners in case they want to change priorities within a sprint.
In the retrospective, team members also agreed to improve code review rules, so that no matter how small a task is or no matter how urgent the bug fix is needed, code review shall always be done by the team members in order to maintain code quality.
Analysis of Fourth Piloted Sprint: In this sprint retrospective, team members especially the team leader appreciated the stability and flexibility of new prototype process, even though team would like improvements in other areas such as, improvements in quality of story specifications, improvements in communication from product and sales departments so that team is in synchronization with management expectations and sprint goals. Improvements in user story specifications can only happen with constant backlog grooming as described in the new process. Further, to improve communication, Product owners will need to ensure that sprint goals and expectations are in synchronization with Product Management and Sales team’s expectations.
Pilot Conclusion: Summing up, as a result of the new development process described in Section 5.1, sprint retrospectives indicates that team’s new process has provided the needed flexibility to the product owners and hence enhance the time to market of new features. And at the same time, team has the flexibility to handle urgent customer requirements within a sprint. Further, work flow across the different stages has improved, and team is managing the bottlenecks and collaborating better based on explicit rules and policies. Scrum ceremonies such as sprint retrospectives were used effectively to improve the scrumban process further, and outcome was often a small set of prioritized
action points. Overall, the output of team has changed little as far as Average Velocity per team member is considered. Average Velocity per team member has increased from 6.77 to 7.0, which is just a marginal increase of 3.32%. Only future sprint velocity calculations can point out whether the output has increased or decreased over a longer term. Using the new process model, team was still trying to improvise some aspects of the new process in order to improve collaboration and handle different situations which are not addressed 100% by the new process model. There is a scope of improvement when it comes to product backlog grooming, review process and handling pressure situations. After the Pilot ended, team agreed to continue using the new process model in future sprints as well. Head of the Software Development Team of the case company, Timo Valtonen (2014) mentioned, “Eventually it is the responsibility of the team to follow the process, no matter how good a process is on the paper, if it is not followed in the right spirit, it is worthless”.
This section summarizes the research process from the initial stage of setting objective to the final proposal of the improved software development process. In addition, the research methodologies used in this research are described, followed by the final conclusions and thesis validation.
The objective of this thesis was to improve the agile software development process within the case company which could help them in addressing problems described in the current state analysis of this study. One of the teams, namely Application Development Team, within the case company was following Agile Scrum model as a software development process; however, the team was facing many problems that could not be addressed by Scrum. An analysis of this process based on interview discussions and sprint retrospectives found many issues related to: process flexibility, frequent priority changes and pressure situations, team collaboration and communication, workflow, incomplete tasks within sprints, shift in process and many other issues. If these issues were not addressed, they could lead to delays in software delivery, increased project costs and low team motivation. As a result, a new or improved software development process was needed that could address many of these issues.
The proposed model was developed and verified in four iterations. Further, this study was mainly based on qualitative analysis of discussions with different stakeholders which included: Product Manager/Owner, Team Leader, Scrum Master, Team members, Quality Assurance (QA) head, QA team member, and Head of Software Development Teams. The interview discussions were conducted with same stakeholders both in the current state analysis as well as the new process development.
The researcher started with defining the study objective which was followed by the current state analysis of the current process. Based on the findings from the current state analysis data, the researcher focused his search for best practices in literature as well as literature review. In literature study and review, the researcher mainly focused on three agile models namely Scum, Kanban and Scrumban. From the literature study and current state analysis a new software development processes created. This model was further developed based on the data collected from qualitative analysis of interview discussions. Finally, the new proposed model was further improved and pilot tested in four sprints during the thesis study, which resulted in the final process proposal and conclusions.
Thus, the outcome of the thesis is the new software development model for the case company based on Agile Scrumban model. And the results from the pilot testing indicate that the output of team has slightly improved as far as the average velocity per team member is considered; there is just a marginal increase of 3.32%. The results also indicate that many process-related things have improved such as team collaboration, improved workflow across different stages, flexibility to address important and urgent business requirements, some aspects of the team communication, process stability, etc. There are still some areas where process improvement is needed further, and many of those issues can be addressed easily within the framework of new software development process.
Even though the new software development model has addressed many of the issues faced by the development team, there is still some scope of improvement in certain areas such as, the code review process, product backlog grooming and handling high-pressure situations. For instance, in the new process backlog queue and the ready queue are same, which means that the new user stories or tasks can only be processed by the team on the next ready queue update meeting within a sprint. In order to add more flexibility and separate the process of task assignment from task prioritization, the ready queue can be separated from the backlog queue. Further, ready queue shall only contain a fixed prioritized subset of backlog queue. As a result, developers can pull the new tasks from the ready queue and product managers will have the flexibility to update the backlog queue as and when needed. Product owners will not need to wait for the next Ready Queue update, as it is a separate queue from the backlog queue. Other issues like improving the quality of specifications can be addressed by properly implementing the product backlog grooming sessions, based on the new process model defined by this study. In addition, over the time team could also improve the team related work policies. Further, the pilot tests for few sprints were done with less number of team members; therefore it would be interesting to know the actual changes in Average Velocity once the team size is increased back to the original size of six members.
Outcome Vs Objective
The objective of this thesis was to improve the agile software development process of the Applications Development Team within the context of the case company. The study proposed a new model based on Agile Scrumban and it retained the vital elements from the current model based on data from current state analysis and qualitative analysis. Overall the thesis was successful based on two main decision points; first the proposed software development model was approved by different stakeholders within the case company for pilot testing, and second, after the completion of pilot testing the team continued to use the new process in future. Further, the positive feedback of the new process based on retrospective data of piloted sprints indicates that new process has added flexibility, simplified the planning, improved team collaboration, made the workflow smoother, and addressed many other issues found in the current state analysis of this study.
Reliability and validity
In order to ensure the trustworthiness of one’s thesis, the researcher has evaluated it in two important aspects: reliability and validity. Reliability is the tendency towards consistency found in repeated measurements of the same phenomenon. A research is said to be highly reliable if it consistently gives the same results if measured repeatedly and if the results are less consistent, the reliability of the study is low (Carmines & Zeller, 1979). Once the data is analyzed, it should be possible to be certain that if the study is conducted again, the same results will be produced. If there is uncertainty in this aspect, then the research will not be considered reliable (Mayor & Blackmon, 2005).
In this Thesis, the proposed software development model was produced based on literature study, inputs from current state analysis, and data collection based on qualitative interview discussions. Further, the research was done in various stages, with each stage providing inputs to the next phase. As a result, knowledge of the research problem, as well as solution ideas, improved from one stage to another. The reliability of the literature data was realized by using the data sources for the literature study and literature review from a broader range of leading academic journal articles, reputed online sources and books published related to the application of agile software development processes. In the data collection and analysis phase, reliability of data sources for qualitative analysis was realized by choosing the participants from various backgrounds within and outside the team in order to get broader perspectives and viewpoints. Reliability of the data collection for the current state analysis was realized by using the case company’s reliable intranet resources and online tools. Finally, the new process model creation was also done in iterations, where it was further refined and improved gradually. The consistency in the results of the iterations also indicates that the study is reliable.
Validity is affected by the researcher’s perception of validity in the study and his/her choice of paradigm assumption (Creswell & Miller, 2000). As a result, many researchers have developed their own concepts of validity and have often generated or adopted what they consider to be more appropriate terms, such as, quality, rigor and trustworthiness (Davies & Dodd, 2002; Stenbacka, 2001). The research is valid if it captures the truth of the situation and is not influenced by outside influences or personal preferences (Mayor & Blackmon, 2005).
In this thesis, the validity of the research is accomplished since the researcher achieved the main objective he had established in the beginning of the research. The new proposed model addresses most of the issues that were presented in the current state analysis of this study. Validation of the new process model was further realized by pilot testing of the proposed model in several iterations. The data collection for this study at various stages was based on the data from reliable and valid sources, which accordingly reflected the true situations in current state analysis as well as in the results of the pilot testing phase. Further qualitative interview discussions were handled one to one with different stakeholders so that the interviewee is not influenced by other opinions. In order to get comparable data, the same structure of interviews was followed to collect common and diverse opinions from different stakeholders.
- PBIs – Product Backlog Items
- PO – Product Owner
- UI – User’s Interface
- QA – Quality Assurance
- WIPs – Work In Progress Items
- XP – Extreme Programming