When organisations talk about AI alignment, they almost always mean the technical kind — making sure a model does what you intend. But there is a prior alignment problem that goes unaddressed, and it is the reason so many AI initiatives stall after a promising start.

Two kinds of alignment

Technical alignment asks: does the system behave as designed? This is a legitimate and important question. But it assumes the design itself reflects something coherent — a shared understanding of what the organisation is, how it makes decisions, and what it values when trade-offs arise.

Organisational alignment asks: does the design reflect how we actually work? Most teams skip this question entirely. They hand a brief to a vendor, receive a model, and discover six months later that it optimises for something the organisation says it cares about but does not, in practice, prioritise.

The cost of misaligned assumptions

Consider a financial services firm that deploys an AI to accelerate client onboarding. The model is trained on historical approvals. But historical approvals reflect a risk appetite from two leadership cycles ago. The new leadership says it wants to be more selective. The model does not know this. It cannot know this, because no one encoded it.

The result is a system that is technically correct and organisationally wrong. It performs as designed. The design is the problem. This pattern repeats across industries, functions, and system types. The technical problem gets solved. The organisational one gets ignored until it becomes a crisis.

Starting from identity

The solution is not more sophisticated modelling. It is earlier, more deliberate work on organisational identity — articulating what the organisation actually values, how it actually decides, and where it is willing to accept friction in service of those values.

This work is not comfortable. It surfaces disagreements that leadership would often prefer to leave implicit. But implicit disagreements do not disappear when AI is introduced. They become encoded, automated, and scaled.

Alignment begins before the first model is trained. The organisations that understand this do not just build better AI systems. They build better organisations.