AI Kills the Functional Organisation

Architects think AI makes architecture more important. Data people think AI makes data more important. HR thinks AI makes HR more important.

And they have it backwards.

We built standard functional business units because they help professional mobility across organisations. Not because they help organisations. An HR director can move from a bank to a retailer to a government agency and slot straight in, because the function looks roughly the same everywhere. That’s useful for professionals. It’s considerably less useful for the organisations that employ them.

Mintzberg made the observation that we structure our businesses the way we structure our business schools. Business schools don’t teach management. They teach HR, IT management, financial management, a set of disjointed specialisations. Then we’re surprised, every time, that everything of value needs a cross-functional team. Of course it does. Our organisations are practically designed to prevent anything from getting done without one.

But the cross-functional team is only the most visible symptom. The more telling one is what happens when you try to implement any of these functions seriously.

I wrote about this in 2018. In that perverse type of functional organisation that has a business unit for each profession, there is a predictable progression:

1. The function cannot succeed until it is centralised

2. The function cannot succeed until it has the sponsorship of the CEO

3. The function cannot succeed until it has its own “Chief xxxx Officer”

4. The function cannot succeed until everybody understands that the function is “everybody’s responsibility”

5. The function cannot succeed until it has the resources to engage with “everybody”

6. The function cannot succeed until there is a dedicated business unit that makes the function part of the business-as-usual process

7. The function cannot succeed until it has input into the strategy of the organisation

8. The function cannot succeed until it is part of the organisation’s culture

9. The function cannot succeed until it reports to me

Every function goes through this cycle. Every single one. It’s not a sign that the function is failing. It’s the function working exactly as designed, expanding its claim on the organisation in the only direction available to it.

And here’s the other pattern that doesn’t get talked about enough. The leaders of these functions have a habit of rejecting their own specialisation. Ask an architect what architecture is really about and they’ll tell you it’s about business value. Ask a data leader and they’ll say it’s about culture and change. Ask HR and they’ll talk about relationships, trust, strategic alignment. None of them will tell you it’s about the thing their function is named after.

The effect is that all of these leaders start to sound identical. The specialised knowledge that was supposed to make the function distinct gets quietly set aside at the leadership level. The specialists below know more than ever. Their leaders sound more and more like each other.

So you end up with well-defined knowledge that the leaders don’t engage with. And leaders who all sound the same.

That is not a strong position going into an era of capable AI.

There is almost nothing easier for an AI model than playing the role of a functional manager encountering an idea for the first time. The pattern is highly legible: consider the idea through the lens of the function, identify the risks and dependencies, note what needs to go through the right process, signal alignment with enterprise priorities, request more information. It’s a well-worn script because the function has encoded it over decades. These are the most well-defined and repeatable roles in the organisation. We built them that way on purpose.

The functions arguing loudest that AI makes them more important are describing exactly what AI will replace first.

The underlying knowledge matters. Designing systems that work with AI, managing data that feeds it, thinking through what happens to workforces. All of that is real work. But it requires engaging with the specialisation, not talking around it in the language of culture and business value.

The institutional form of the function is what’s genuinely exposed. The governance, the committee structures, the predictable climb from “we need centralisation” all the way to “we need to report to the CEO”. AI doesn’t need to be smarter than a functional leader. It just needs to be better at executing a predictable role.

And it is.

(Note: this article used my notes over the years on functional organisation to help bring a draft personal note I wrote for myself into something others might like to read.  I think it did an okay job.)