(Prepared by Thérèse Lalor and Steven Vale - September 2013 - Updated March 2016)
The idea of holding sprint sessions to develop standards for official statistics is fairly new, but has already it seems to have a lot of promise. This approach can shown its ability to considerably reduce development time ( and probably cost – but hard data on this are not really available). This note section sets out the experiences and lessons learned from the five sprint sessions held to date, in the context of projects overseen by the High-Level Group on for the Modernisation of Statistical Production and Services Modernization of Official Statistics (HLG-MOS). It also makes recommendations for the conduct of future sprint sessions in the context of developing statistical standards development.
What is a sprint?
The idea of sprints comes from the “Agile” approach to software development. A sprint is generally defined as “a ‘timeboxed’ effort”an intensive effort, restricted to a specific duration, normally between e.g. one week and one month, with the goal of delivering , to deliver a pre-determined output. This means that in In the official statistical community, a sprint can be explained as being something like a very intense and focused workshop with a specific deliverable.
During a sprint, participants break down the task to be accomplished into small components, each of which can be resolved in one or two hours by a small group of two to five people. This means that sprints . Sprints involve a lot of parallel working, and that so one person can not cannot be involved in everything. The aim is to build on the best ideas and experiences of all participants to reach an outcome that is “owned” by all. This outcome is seldom exactly the one envisaged at the start of the sprint, as ideas should evolve through the sprint process. However, however, it should be at least as useful , and probably better as a result.
The first two HLG sprints (in Slovenia and the Republic of Korea) -MOS sprints were two weeks long, whereas the other three (Netherlands, Canada and Italy) others have been completed within a week. Both types met their goals. Some participants to of the two-week sprints felt that they were rather slow to get going, so for later sprints much of the introductory material was covered in pre-sprint web conferences. This meant that the participants felt they were starting the real work from day one. Some participants, particularly those with family responsibilities, also felt that two weeks is rather a long time to be away from home. Changing to one-week sprints did, however, put more burden on the sprint team, requiring much more pre-sprint and post-sprint work to compensate for the more limited time available during the sprint itself.
The ideal set-up is to have the exclusive use of at least three rooms for the whole duration of the sprint. To save time, it is preferable that these rooms are located close to each other. At least one of the rooms should be large enough to comfortably accommodate the whole group seated at desks, which should preferably be organised in a U shape, facing a screen for presentations. All rooms should be equipped with whiteboards, flip-charts and writing materials, and ideally it should be possible to stick posters and flip-chart sheets on the walls. Plenty of stationery, particularly paper, pens and post-it notes should be available. WiFi Internet access is important, along with easy access to an on-line collaboration space, e.g. a wiki. Participants should bring their own laptops / tablets. Finally, to save time, it is helpful if coffee and lunch facilities are available close by.
Travel and accommodation
For the HLG sprints held so far, the vast majority of participants have paid their own travel costs. In just a handful of cases, where people were considered essential and there was no other way for them to attend, project funding has been made available to subsidise their costs.
From a practical point of view, it is very useful if all participants stay in the same hotel during the sprint. This helps build the team spirit, and facilitates discussions in the evenings and over breakfast. It is helpful if the hotel has a public seating area where people can work in small groups. It is very helpful if someone from the host organisation can help to make local arrangements and advise on practical matters such as travel between the airport, the hotel and the sprint venue, including where and how to buy tickets for public transport.
Ideally a sprint should have 10-15 participants. If there are more, it is difficult to maintain a sprint atmosphere and to get people to work as a single group. Normally there should be no more than two participants per organisation, to ensure that as many organisations as possible can be represented. Participants should come from a range of backgrounds, and, between them, should have all the necessary skills and experience to produce the output. For HLG-MOS sprints, participants have been mainly at the “expert level”, and have included methodologists, IT specialists, metadata experts, business / information architects and external experts, depending on the topic. However, having a more senior person present can help ensure a sufficiently strategic focus.
Ideally there should also be representatives of “the business”, i.e. subject matter statisticians. In practice it has proven difficult to achieve this, as organisations are more interested in spending money to send “specialists” as participants. Two possible solutions are to try to select “specialists” who have previously worked in statistical subject-matter areas, and to ask the host organisation to make subject-matter people available for consultations if necessary.
The sprint master should also use the pre-sprint period to research the topic. The aim is not to become an expert, but to have enough knowledge to be able to create the first inputs for the sprint team. It is easier for the team to start with something other than a blank page, even if the final output is something completely different.
At least three two pre-sprint web conferences are essential, both to help the participants to get to know each other, and to cover all the basics about how a sprint works. It is helpful if participants outline their backgrounds, ideas and expectations for the sprint, so that some degree of consensus can be reached before they physically meet.
During the sprint
Most HLG-MOS sprints have been opened or addressed by the head of the host organisation. Active interest from this level can help to inspire participants. Towards the end of each sprint, a presentation is usually given to the staff of the host organisation. This serves several purposes, but mainly helps the participants to focus on how to communicate their work, and explain it to people outside the sprint group.
It is also useful to take stock at this point about the organisation of the sprint. What worked, and what should be done differently in future?
Whilst most HLG-MOS sprints have been face-to-face events, the idea of virtual sprints has also been tested, with some success. In most cases, this approach has involved bringing sprint participants together several times, each for a few hours, by web conference. For example, a virtual sprint can take place over a week, with a two-hour plenary web conference each day, at the time most suitable given the different time-zones of participants. In-between these plenary sessions, participants work individually or in small groups (with people in similar time zones) to produce inputs to the next plenary discussion.
Virtual sprints are clearly much cheaper than face-to-face sprints and can be effective in certain circumstances, particularly if the “sprinters” have already met beforehand. They work best for topics that are already reasonably well developed, such as revising an existing standard but are less suitable for developing new material which usually requires more intensive discussion.
Sprints have proved very valuable in the context of the HLG-MOS projects. The Generic Statistical Information Model (GSIM) Development Project was completed within a year, which is less than half of the time taken for comparable initiatives using more traditional methods. There is a cost, both in money and time, but feedback from chief statisticians so far is that the value of the outputs more than outweighs the costs.