Twelve years. That is how long I have spent deploying SAFe in organizations ranging from 500 to 50,000 people. At Renault, Allianz, La Poste, Orange Business. Here is the unfiltered summary of what works -- and what does not.
What works: the 5 levers
1. In-person PI Planning, at least for the first one. The quality gap between a first PI done in person and a first PI done remotely is dramatic. Dependencies reveal themselves in hallway conversations.
2. An RTE who comes from the field. The best RTE I coached at Renault was a former Scrum Master who had spent 3 years embedded in the teams. He facilitated with credibility.
3. Start with a single ART. Pick one value stream, identify an ART of 50 to 125 people, and make it work for 3 PIs. At Allianz, this approach enabled scaling from 1 to 6 ARTs within 18 months.
4. Flow metrics. Lead time, throughput, PI predictability. Not story point counts.
5. Inspect & Adapt as a real lever. At Orange Business, an I&A revealed that 60% of cross-team dependencies came from a single API. Within 2 PIs, dependencies dropped by half.
What does not work: the 4 traps
"SAFe light." You take PI Planning, ignore the Continuous Delivery Pipeline, skip the Architectural Runway. SAFe is a system. You can adapt it, but you cannot gut it.
Certification without field coaching. Every ART needs 2 to 3 PIs of intensive coaching before it can run autonomously.
Ignoring system and platform teams. They become a bottleneck that nobody anticipated.
Middle management left out. This is the number one failure factor. Without a dedicated program for managers, the transformation stalls.
After 12 years and more than 30 ARTs launched, my conviction is simple: SAFe works when you deploy it with rigor, pragmatism, and sustained field coaching.
Agile coaching in the age of AI naturally evolves toward IAgile, the convergence of agile principles and artificial intelligence.