Context

AI adoption was expanding across multiple business units, each with distinct priorities, capabilities, and delivery pressures. Some teams wanted freedom to move quickly and tailor AI to their specific needs, while central functions were accountable for risk, coherence, and long term sustainability. Previous attempts at central control had slowed adoption, while periods of unchecked local autonomy had resulted in duplicated effort and incompatible practices. The organisation was operating at a scale where neither full centralisation nor complete decentralisation was viable.

‍

The Challenge

The challenge was not designing a federated model in theory, but making it work in practice. Central teams needed enough control to manage risk, reuse, and alignment, without becoming a bottleneck. Local teams needed autonomy to deliver value in their context, without being forced into one size fits all patterns. Informal coordination was no longer sufficient, but formal control mechanisms risked pushing teams back into siloed behaviour or shadow solutions. The organisation needed a way to hold the centre and the edges together without defaulting to command and control.

‍

The Decision

The organisation chose to federate AI adoption explicitly, rather than letting it emerge implicitly. Central teams retained authority over a small number of non negotiable expectations, such as shared principles, accountability boundaries, and conditions for reuse. Local teams were given freedom to execute within those boundaries and to make context specific decisions about how AI was applied. Importantly, the organisation rejected both extremes: a single central AI factory, and a hands off model where each business unit operated independently.

‍

What Changed

The relationship between central and local teams became more intentional. Disagreements shifted from “who is allowed to do this” to “where does this decision belong.” Local teams moved faster because expectations were clearer, not because controls were removed. Central teams spent less time intervening late and more time shaping early decisions. Some inconsistencies remained by design, but they were visible and understood rather than accidental. The organisation traded uniformity for coherence.

‍

Why This Matters

Federated AI models fail most often when roles and boundaries are left ambiguous. Too much central control suppresses adoption; too little creates fragmentation that is hard to reverse. Balancing control and autonomy is not a structural diagram, but an ongoing operating choice. Enterprises that make this choice explicit are better able to scale AI without repeatedly reorganising or re centralising when problems surface.