Designing the AI Op Model

Who owns AI here?

The Chief Data Officer thinks they do, and so does the CTO. The CFO owns the funding and general managers are busy using that to demonstrate their ability to integrate AI into their business functions. Ironically, the organisation chart doesn't define it; it only includes humans.

Nobody should own AI. But somebody must own the decisions it makes.

What's at stake

Only one in eight CEOs report that AI has influenced both revenue and costs positively (PwC), and about seven out of eight organisational leaders report that their organisations are not very well prepared to adopt AI in day-to-day operations (McKinsey). Those numbers point to something more interesting than an adoption problem. If this were simply a matter of buying better technology, the problem might have been solved by now.

Early digital programs often digitised existing processes before organisations learned to redesign the customer journey around what digital made possible. AI appears to be following a similar pattern. We are making existing tasks faster before asking the more consequential question: if intelligence becomes cheap, scalable and increasingly autonomous, how should the process work in the first place?

The lesson from past digital transformations is that applying technology to the same organisational design and processes only captures part of the value. Organisations that took a holistic approach and redesigned the experience rather than just applying technology to existing processes captured more of the opportunity. Yet despite the opportunity to redesign work, most leaders still expect AI to function primarily as a support tool over the next two years.

Where to start

Before an organisation can delegate work to machines, it has to understand what it is delegating.

This is where things get harder. Most organisations are far less explicit about how they actually operate than they think they are. Policies describe the normal case, but exceptions live in people's heads. Decision rights are often implicit. Process logic is embedded in systems, and experienced employees have learned how to navigate the gaps.

Humans can work around that ambiguity. Reliable autonomous agents need much more of it made explicit. This isn't necessarily a model problem. It is increasingly an organisational knowledge problem, and it should be treated and funded like infrastructure.