SAFe (Scaled Agile Framework) is the framework published by Scaled Agile, Inc. to have several agile teams contributing to the same solution plan and deliver together at a fixed cadence. For agile project management at scale, it is suitable when several dozen people work on that solution, when their teams wait on one another in order to deliver, and when management agrees to plan and fund this work over time, without breaking it back down into projects.
It is poorly suited to a few teams working on a single product, which lighter arrangements can coordinate, to independent teams, which have nothing to synchronise, and to an organisation that expects the framework to provide a firm scope and date beyond the current increment. SAFe’s concepts are defined in our glossary entry on the Scaled Agile Framework; this article confines itself to the decision to adopt it.
What SAFe changes in agile project management at scale
The search query combines two notions that the framework separates. A project has a start, an end and a scope. SAFe’s unit is the agile release train (Agile Release Train, ART), which the framework’s website (framework.scaledagile.com) describes as a long-lived team of agile teams, organised around a value stream, and which generally brings together 50 to 125 people.
Planning there follows a cadence. The same website presents PI planning as an event that brings the whole train together, every eight to twelve weeks, generally over two days. It produces objectives to which the teams commit and a board of delivery dates and dependencies.
A project manager’s work (planning, coordinating, tracking dependencies) remains. It is shared between the train’s key roles, which the website cites (Release Train Engineer, Product Management, System Architect, Business Owners), and the teams themselves. The scope set for the whole duration gives way to objectives revised at each increment.
Not everything lends itself to this regime. A compliance programme with an imposed date, a data centre move or a contract with a set price and scope keep a deadline, a scope and an overall owner. Programme management can then bring together a train and projects run differently, around shared milestones.
The conditions to check before adopting the framework
We suggest checking five conditions, each against a fact, before any training and any appointment.
| Condition | What makes it possible to check it | If it is missing |
|---|---|---|
| A headcount the size of a train | The number of people contributing to the same solution, compared with the size the framework indicates | A lighter arrangement is enough |
| Real dependencies between teams | Among the latest go-lives, those that required the work of several teams | There is nothing to synchronise |
| Management that backs the approach | A named sponsor, present at planning and empowered to arbitrate priorities | Joint planning takes place with nobody to decide |
| Funding that follows the flow | The ability to change a priority during the year without a new budget case | Increment objectives become a frozen scope again |
| A system that can be assembled | A demonstration of the integrated product, possible at the end of each increment | Joint planning covers pieces that only come together at the end |
The third condition heads the approach that the framework itself publishes. The implementation roadmap published on its website opens with “Reaching the Tipping Point”, the moment when the organisation seeks to make the change succeed rather than to resist it. It places the training of executives and managers before the launch of the first train.
This roadmap includes further training steps, for change agents and then for the teams when the train is launched. Added to this is the time spent on joint planning: the whole train brought together several times a year, preferably in the same place according to the framework. This cost enters into the decision, together with that of the roles to be filled.
The fourth condition concerns the portfolio. If every change of priority requires reopening an investment case, increment planning runs idle. Our entry on project portfolio management presents its concepts.
When LeSS, Nexus or a Scrum of Scrums is enough
Below the size of a train, or when all the teams serve a single product, other public arrangements target a narrower scope. We describe them by their principle, as their websites present them.
Nexus, whose guide is published by Scrum.org, brings together around three to nine Scrum teams around a product. They work from a single Product Backlog, held by a single Product Owner, and an integration team (Nexus Integration Team) ensures that the whole delivers an integrated increment. Nexus addresses a specific difficulty: assembly.
LeSS (Large-Scale Scrum) presents itself as Scrum applied to several teams working together on one product: one Product Owner, one Product Backlog, one common sprint. Its website targets two to eight teams and provides a variant beyond that, LeSS Huge, divided into Requirement Areas. The framework avoids adding roles, on the grounds that doing so takes responsibility away from the teams. It therefore assumes an organisation ready to do without intermediate coordinators.
The Scrum of Scrums coordinates teams without reorganising them. The Scrum@Scale guide describes it as a team of teams that itself functions as a Scrum team: at least one representative of each team takes part in a common daily scrum (Scaled Daily Scrum). This guide puts the optimal number at four or five teams. The arrangement can serve as a first step.
Nexus and LeSS are defined around a single product. As soon as the question concerns several products, the arbitration between them and their funding, it falls outside their purpose. The comparison of frameworks and the way we help with the choice are set out on our page on agile at scale.
The situations where SAFe is not the answer
Refraining is decided on facts, just like adopting.
Teams whose products do not depend on one another have nothing to plan together. The framework would bring them a cadence and events with no purpose.
Dependencies that all lead back to the same component, or to the same team called upon by the others, point to the architecture or to the way the teams are divided. Synchronising the wait does not remove it, and that is the point to address first.
Teams that do not yet deliver a usable result at each iteration would do well to stabilise this way of working before coordinating it across several teams. Scaling a fragile practice spreads the fragility.
A sponsor who expects a firm scope and date over several increments will not find them in objectives revised at each planning session: programme management serves them better. Finally, a management team that delegates the subject to a methods department leaves the third condition in the table unanswered.
A record helps to decide: among the latest go-lives, how many waited on another team, and which one. If they are rare, they justify no scaling framework. If they always hinge on the same team, they call for a local correction. If they are numerous and scattered, in an organisation that meets the five conditions, they justify trying SAFe on a first train.