program selection system

Authors: Jeff Sutherland, Ph.D. Guido Schoonheim Eelco Rustenburg Maurits Rijk Scrum, Inc. Xebia b.v. Xebia b.v. Xebia b.v. Somerville, MA, US Hilversum, Netherlands Hilversum, Netherlands Hilversum, Netherlands jeff.sutherland@computer.org gschoonheim@xebia.com erustenburg@xebia.com mrijk@xebia.com

Scrum was designed to achieve a hyperproductive state where productivity increases by 5-10 times over industry averages and many collocated teams have achieved this effect. The question for this paper is whether distributed, outsourced teams can consistently achieve the hyperproductive state. In particular, can a team establish a localized velocity and then maintain or increase that velocity when distributing teams across continents. In 2006, Xebia (Netherlands) started localized projects with half Dutch and half Indian team members. After establishing a localized velocity of 5 times their waterfall competitors on the same project, they moved the Indian members of the team to India and showed increasing velocity with fully distributed teams. Running XP engineering practices inside many distributed Scrum projects, Xebia has systematically productized a model very similar to the SirsiDynix model for high performance, distributed, outsourced teams with outstanding high quality. The fully distributed model is now the recommended standard for high performance distributed Scrum when organizations can pass the Nokia test [1] for Scrum implementation and fully implement XP engineering practices [2] in their Scrum teams.

1. J. Sutherland, “Real World Experience Creating Agile Companies,” in Agile 2007, Washington, D.C., 2007

2. J. Sutherland, A. Viktorov, J. Blount, and N. Puntikov, “Distributed Scrum: Agile Project Management with Outsourced Development Teams,” in HICSS’40, Hawaii International Conference on Software Systems, Big Island, Hawaii, 2007

   There is a full length paper that will be avalable at the session and presentation and discussion will address the following key points:

Outsourcing Models

Outsourcing strategies and failure modes will be described with examples from specific companies. For example, PatientKeeper outsourced to an excellent waterfall team in India. For two years they monitored return on investment. When it was discovered that the break even point occured when an Indian programmer cost 10% of a Boston Scrum programmer the Board of Directors terminated all outsourcing. Since Indian resources cost 30% of Boston resources, if we outsourced $2M of development it cost us $6M to get it done in India. Venture capitalists on the Board were unhappy because they had demanded the company do outsourcing and they would wast $4M of venture capital for every $2M of development outsourced.

The fully distributed Scrum model as implemented in this paper achieves the same velocity onshore as well as offshore. PatientKeeper could outsource $2M and get it done for $666K with the model described in this session. Xebia found that this outsourcing style actually made the onshore team function at a higher level because they had to explain requirments issues daily to the Indian part of the team. This kept local team focused on the customer and they implemented the requirements better than when they worked totally onshore.

Many of the findings in this paper are counterintuitive to conventional Agile distributed team strategies. Extensive data will be provided to document every piece of this project.

Strengthening the distributed team model

SirsiDynix solved this problem with fully distributed teams on a singe large project. Each team was half in the U.S. and half in Russia. It is possible to replicate and strengthen this model. Xebia executed multiple projects with fully distributed teams. They kept much better data on velocity onshore and offshore and had better testing and refactoring. Both onshore and offshore team components fully implemented Scrum with XP inside. The distributed model in this paper is a replicated repeatable model that is now used on all Xebia projects.

Development team scaling offshore with this model allows expanding and shrinking offshore team without losing knowledge. For some organizations this is more important than the cost savings from outsourcing.

Process over culture

Critics may claim it because Xebia has better people which is why they were successful. However, quality of people onshore and offshore was the same. Distributed velocity equaled local velocity with the same quality of people throughout. There were some Indian cultural differences and we will discuss how they were resolved with this distributed team model.

Soft factors

There were many soft factors improved by the fully distributed model over conventional distributed teams:

• Focus on customer • No cultural barrier • Shared ownership • High satisfaction – no attrition • Shared knowledge • Feeling for context • Same standards

Typical Indian developer comment “It doesn’t matter to me if I am working in Delhi or Hilversum on this team because we all do exactly the same things.”

• Shared ownership of code across geographies • Shared responsibility for result • High level of initiative from Indian developers with new ideas for improvement • Teams like the room to create a better workplace for themselves. • Very light touch management. Giving trust first and earning trust second.

Detailed graphical data will be presented

A highly experience project manager (co-author Maurits Rijk) developed extensive graphics and tabular data for the projects presented in this paper. Projects are much better documented than previous published papers on distributed teams.

• Velocity and hours per story point
• Quality – total defect and architecture debt
• Quality and scaling strategy done better than SirsiDynix, for example. This will be discussed in depth.