Definition of SAFe
SAFe, for Scaled Agile Framework, is a framework published by the American company Scaled Agile. By way of definition, SAFe can be summed up as follows: a set of roles, events and practices that extends agile and lean principles to organisations where dozens, or even hundreds, of people contribute to the same product or the same system.
The original agile methods, Scrum first among them, are designed for a single team. They say little about what happens when ten teams depend on one another, share an architecture and have to deliver together. SAFe offers a complete and very detailed answer to that question. This exhaustiveness explains both its spread in large companies and the criticism it receives.
What SAFe is for
The framework pursues three objectives. It aligns teams on common priorities, so that each one does not optimise its own backlog. It synchronises them on the same cadence, so that their work is integrated on fixed dates rather than at the end of the project. Finally, it links strategy and funding to the teams’ work, through a portfolio level.
The levels of SAFe
SAFe comes in several configurations, from the simplest to the most complete, which stack up to four levels. Their names have varied from one version of the framework to another.
| Level | What it organises | Characteristic elements |
|---|---|---|
| Team | The work of one agile team | Scrum or Kanban, short iterations, team backlog |
| Train (essential level) | Coordination of several teams on the same product | Agile release train, PI planning, system demo |
| Large solution | Coordination of several trains and suppliers | Complex systems, often hardware and software |
| Portfolio | Choosing and funding investments | Strategic themes, value streams, lean portfolio management |
The basic configuration brings together the first two levels. The other two are only justified for very large systems or when the organisation wants to rethink the way it funds its development work.
The agile release train
The agile release train (ART) is the heart of the arrangement. It is a long-lived team of teams, built around a value stream and not around a project. It brings together all the skills needed to design, develop, test and put into service.
Four key roles run it. The Release Train Engineer facilitates how the train works and removes obstacles. Product Management sets content priorities. The system architect ensures technical consistency. Business leaders, called Business Owners, assess the value of the objectives and results.
PI planning
A train moves forward in planning intervals, known by the abbreviation PI (Planning Interval, formerly Program Increment). A PI groups several iterations and usually lasts eight to twelve weeks.
PI planning brings the whole train together to plan the next interval. The teams build their plans there, identify their dependencies and formulate their PI objectives; the framework also provides for a confidence vote. The outcome is a set of PI objectives and a train planning board on which dependencies are shown. Our PI planning entry describes the event in detail.
SAFe, Scrum and portfolio management: not to be confused
SAFe is not a “big Scrum”. Scrum remains one possible way for a team to work inside the framework, which adds levels of coordination that Scrum does not describe.
Nor is SAFe a project management method. It thinks in terms of long-lived products and value streams, whereas a project has a beginning and an end. Its portfolio level takes up, in another form, the questions of project portfolio management (PPM): what to fund, in what order, with what capacity. The difference is that it funds value streams rather than individual projects.
Limitations and alternatives
The criticisms levelled at SAFe are well known.
- Heaviness. The number of roles, events and terms to absorb is high, and so is the cost of training.
- Agility in name only. The framework can be laid over a hierarchical organisation by renaming existing functions, without changing anything in the decision-making channels.
- Rigid planning. A PI planning run as a firm three-month commitment recreates a small-scale waterfall cycle.
- Treating the symptoms. Synchronising highly interdependent teams is expensive. Reducing those dependencies, through architecture or the way teams are split, is often more effective.
The most frequently cited alternatives are LeSS (Large-Scale Scrum), Nexus, Scrum@Scale and Disciplined Agile. The choice depends on the number of teams, their technical coupling, existing agile maturity and the degree of prescription the organisation is prepared to accept.
Example: eight teams, a single product
Eight teams develop the different parts of the same contract management application. Each one keeps to its iterations, but features that cut across several teams arrive late, because dependencies are only discovered at integration time.
The company sets up an agile release train. The teams adopt the same iteration calendar and meet for a first PI planning. Two findings emerge: one team is called upon by all the others, and two planned features cannot fit into the increment. The scope is adjusted on the spot, with the business leaders. The framework has not removed the dependencies. It has made them visible early enough to decide.
For choosing and implementing a framework suited to your context, see our page on agile at scale.