There is a pattern that repeats itself across almost every AI programme that stalls. The technology works. The pilot was a success. The board is enthusiastic. And then nothing happens, because nobody designed the operating model that would allow the capability to function at scale.
The pilot team built something impressive, but there is no clear owner for it in the day-to-day business. Nobody has defined how it interacts with existing processes. The governance framework does not cover it. The people who need to use it were not involved in designing it. So it sits there, a working proof of concept with no pathway into the organisation.
This is not a technology problem. It is an operating model problem.
What we mean by operating model
An operating model is the bridge between strategy and execution. It defines how an organisation delivers value: what capabilities it needs, how those capabilities are structured, who owns them and how they interact. When you introduce AI into that picture, the operating model has to evolve, because AI does not just automate existing processes. It changes what processes are needed in the first place.
Consider a mid-market insurer processing claims. AI can accelerate triage and improve fraud detection, but that changes the role of the claims handler, the escalation pathways, the quality assurance model and the regulatory reporting chain. If you deploy the AI without redesigning those elements, you get friction, workarounds and, eventually, a capability that nobody trusts.
The three layers of AI operating model design
Across the organisations we have worked with, effective AI operating model design addresses three layers.
Capability architecture. Before selecting any AI tool, you need to understand your business capabilities, what you do, how well you do it and where AI creates real leverage. This is enterprise architecture applied to AI strategy, and it prevents the common trap of starting with technology and hoping it finds a problem to solve.
Process and role design. For every AI capability you introduce, the surrounding processes and roles need to be redesigned. Who makes decisions the AI used to support? Who oversees the AI's outputs? What happens when the AI gets it wrong? These are not afterthoughts, they are first-class design decisions that determine whether the capability delivers value or creates risk.
Governance and accountability. Every AI capability needs clear ownership, defined performance metrics and a review cadence. Without this, capabilities drift, models degrade and nobody notices until something goes wrong. This layer is especially critical in regulated sectors where accountability cannot be ambiguous.
Why this matters for mid-market businesses
Large enterprises can absorb operating model failures, they have enough resource and enough resilience to muddle through. Mid-market businesses cannot. A failed AI programme in a 500-person company is not just wasted investment, it is lost momentum, eroded trust in technology-led change and a leadership team less likely to try again.
That is why we start with the operating model, not the technology. Getting the design right from the beginning means every AI capability you deploy has a home in the organisation, clear ownership and a pathway to value.
For the strategic framing that underpins this approach, see Why AI Strategy Must Lead Technology. For an example of how operating model design works in practice within a compressed delivery model, see The 8-Day AI Sprint.
If your organisation is ready to design an AI operating model that actually works, Breathe is where we start, and Flow is where we build.