Abstracts

Understanding the effects of agile process design, for example in a Scrum environment, is key for teams to establish a very good team flow. Methods like 'testing first/early' and 'timeboxing' are well known to bring benefit to teams in terms of product quality and time efficiency.
Nevertheless, the questions needs to be answered, how the benefits of both methods can be taught to teams, being new to using these methods. Essential is, besides understanding how the methods can be applied, their impact to the team flow concerning throughput and product quality.
To make these effects visible to learning teams, we have built a simulation environment based on the already existing Okaloa Flowlab (www.okaloa.com). The basic idea is taken from sport training approaches. Thereby you define constrains and train people to experience a predefined learning goal but also leaves enough space to develop own experiences and solutions. This is so called constrained let learning (see Vgl. Renshaw, Ian/Davids, Keith/Newcombe, Daniel/Roberts, Will (2019): The Constraints- Led Approach. Principles for Sports Coaching and Practice Design. Milton: Routledge.  Page. 42).

For example in football training the learning goal could be to have a very high pass rate. The constraints for the simulation could be, that every player only has one contact with the ball while receiving it from another player and passing it again to another player (so called one touch football). Furthermore the field setup is very small and teams of three are playing against each other with the goal to possess the ball.

Transferring this idea to our approach, we used the Okaloa Flowlab simulation environment and build up two simulations. Okaloa Flowlab to simulat ing with a team of four “players” and one product owner. You process a so called work item, which is the representation of real work. Each work item must be processed through two steps, A and B, on the simulation board, represented by two columns, before it finally arrives in the “done” column.
To be processed in a step, the “player” roles a die on every round (representing a simulation day). The die has numbers of points from one to six. You need to have noted down six points on a work item to finish a step. By using a die the simulation injects coincidence to the work speed for work items.

The first extension by our approach shows the effect of early testing compared to later testing in a Scrum team. You can compare the effect of testing early with the idea of a big bang integration at the end of the development cycle. Hence  the learning team can decide, if they want to test in between step A and B (early) or just at the end, when the work item is finished. To see the difference between the two restrictions, the team either plays two simulations rounds or two teams play in parallel with different test approaches. To make the effect of both approaches transparent, the simulation uses KPIs like work item throughput and error rate.

In the second simulation the learning team should get an idea of timeboxing and how new teams develop their workflow. Therefore newly setup teams are asked, how many work items they can process through the simulation board in five minutes (which is the timebox in our simulation). After the estimation is done, the team is doing the simulation for five minutes and checks at the end how many items they have really processed. Now a new estimation takes place, based on the experience of the first round. The second simulation run with again five minutes time is executed and again at the end the processed work items are compared against the estimated number. The last step is a third round of estimation and five minutes simulation with the same approach as before.
At the end of the three simulation rounds, the participants have a look at the results and especially at the differences between estimations and really processed work items. Here the often described effect of over- and underestimation in the first sprints of a newly setup Scrum team gets visible. In the first round the team normally overestimates their performance. By this they will estimate their capabilities in the second round in a more conservative manner and in the end overperform or at least reach their goal. The third round and also further rounds should show, that estimation and performance converge.

In our contribution we are going to show the simulation environment and the approach to learn both methods (test early and timeboxing). We have validated our approach in a case study with our master students.