Why agile at scale disappoints in both directions
Agile at scale runs up against two opposite failures, and it is common to move from one to the other.
The first is agility that has remained local. The teams work in sprints, hold their rituals, deliver software that works. But each one waits for the others to go into production. Integration happens late, priorities come from several sponsors who do not talk to each other, and the budget remains allocated to projects whose scope was fixed a year earlier. The organisation has agile teams and unchanged lead times.
The second is the framework applied from above. To get out of the first situation, the company adopts a complete framework: new roles, new governance bodies, new vocabulary. The old functions are renamed, the committees change their names, and joint planning becomes a ceremony where everyone presents a plan already settled. Nothing has changed in the way decisions are made or in the architecture that creates the dependencies.
In both cases, the subject has been approached through the teams or through the organisation chart. It is dealt with through the flow: what separates a request from its go-live, and everything that makes it wait.
SAFe and its alternatives: which framework for which situation
Several public frameworks offer an answer. None is superior in absolute terms: they correspond to different sizes, cultures and levels of prescription.
| Framework | Principle | Suitable when | Point of caution |
|---|---|---|---|
| SAFe (Scaled Agile Framework) | Detailed multi-level framework that groups teams into delivery trains synchronised by joint increment planning | The organisation is large, regulated, and needs explicit roles and reference points | Can be applied as a procedure, without addressing the causes of dependencies |
| LeSS (Large-Scale Scrum) | Scrum extended to several teams, with a single product owner and a single backlog | Several teams develop the same product and the company agrees to simplify its structure | Requires existing coordination roles to be called into question |
| Nexus | Scrum supplemented by a team in charge of integration between teams | A few Scrum teams share a product and struggle with integration | Deals with neither the portfolio nor funding |
| Tribes and squads organisation | Autonomous teams grouped by domain, with cross-cutting communities of practice | The architecture genuinely allows team autonomy | This model describes one company’s organisation at a given moment: it is not a framework accompanied by rules |
Our glossary entry on the definition of SAFe details the concepts specific to this framework. In practice, many organisations take a framework as a starting point, then lighten or supplement it. What matters is less the name of the framework than the questions it forces you to settle.
The five questions every set-up must settle
Whatever the framework, a set-up at scale works if it answers five questions clearly.
- Who sets the priorities? A single place for decisions per value stream, and a common backlog whose order applies to all the teams concerned.
- At what rhythm? A shared cadence, which makes it possible to plan together, deliver together and correct course at regular intervals.
- How are dependencies handled? They are made visible during joint planning, assigned, then tracked between two planning sessions. The long-term objective is to reduce their number through the way teams and architecture are divided up.
- When is the system integrated? Continuously, with a regular demonstration of the assembled product and not of each team’s contribution. This requirement assumes an automated delivery pipeline, a subject we cover in our DevOps and platform engineering offering.
- How is the work funded? Through capacity allocated to a value stream, whose priorities are revised during the year. This point links the set-up to project portfolio management, which must evolve at the same time.
Our approach: start from a value stream, not from the organisation chart
We begin by following a real request, from its expression by the business to its go-live. This journey shows where the work waits: approval, dependency on another team, unavailable environment, pending decision. The assessment is based on these waiting times, not on a maturity questionnaire.
We then choose with you a first value stream, important enough for the result to matter and well enough delimited to be reorganised without touching the whole company. Our consultants help redefine the scope of the teams there, establish the cadence and prepare the first joint planning with the product owners.
The extension is carried out stream by stream, drawing lessons from the previous one. Managers are supported throughout the journey. Their role shifts from assigning work to removing obstacles, and this shift is the most difficult part of the transformation. It is a matter of change management as much as of method.
The pitfalls we help you avoid
- Renaming without transforming. Changing job titles and committee names alters neither decision paths nor dependencies.
- Administering dependencies instead of removing them. A dependency board that stays just as full from one increment to the next signals a team structure or an architecture that needs rethinking.
- Planning in order to commit to a scope. Increment planning used as a delivery contract recreates a traditional cycle, with less visibility.
- Leaving the portfolio aside. If the budget remains allocated to fixed projects, teams cannot change priorities, whatever the framework on display.
- Measuring activity. Comparing teams on their velocity pushes them to inflate their estimates, without delivering anything more.
Measuring the value delivered, not the activity
The indicators of a set-up at scale describe the flow and its outcome. We track with you the time between a request and its go-live, the frequency of production releases, and predictability, that is, the share of increment objectives actually achieved. We add to these the number of dependencies discovered during the increment, which measures the quality of planning, and the share of capacity absorbed by defects.
These flow measurements do not say whether what is delivered is of any use. Each increment objective is therefore linked to an expected outcome for the user or for the business, checked after go-live. It is this check that underpins steering by value: a feature delivered on time that nobody uses is not a success.