There is a version of the AI conversation that most organisations are having, and then there is the conversation they need to be having.
The one they are having tends to centre on tools: which ones to buy, how quickly to deploy them, how to demonstrate return on investment before the next board cycle. The one they need to be having is harder, less comfortable, and more consequential. It is about how the organisation actually needs to work to make any of this succeed.
I was at an event in London last week focused on AI and leadership. Across the conversations in the room, a pattern emerged that I have been seeing in my own conversations: organisations are treating AI as a technology implementation challenge, when it is in fact an operating model transformation and people challenge.
The mistakes that are recurring
A few failure patterns are showing up repeatedly. Organisations assume they can build and maintain their own software, underestimating both the gap between experimentation and genuine capability, and the ongoing demands of support, maintenance, and keeping pace with a technology that is moving faster than most internal teams can track.. They believe they can develop top-tier internal AI talent from scratch. They are waiting for complete data readiness before beginning, when good enough is the operational standard and perfect is, in practice, a delay strategy. And they pursue quick wins rather than identifying which problems are genuine differentiators for the business and focusing there.
None of these are failures of ambition. They are failures of sequencing. And they share a common root: organisations are designing for AI without first understanding how they actually function today.
One voice in the room put it clearly. You cannot identify where AI will make a difference until you have taken a holistic view of the value chain, including the people doing the work, to surface the full landscape of opportunities. Only then can you prioritise meaningfully, build the business cases, and assess what is genuinely needed. The technology decisions follow the diagnostic work.
What this moment is actually asking for
New data makes this more than a matter of practitioner opinion.
KPMG’s first Adaptability Index, built from surveys of 300 C-suite leaders, earnings call analysis across 177 publicly traded companies, and capital allocation data across six industry groups, found that companies making cultural and structural bets, not just technology investments, saw 4.4x higher shareholder returns and nearly triple the revenue growth of more passive peers.
Most transformation programmes are built almost entirely around technology. The cultural and structural work gets framed as the soft stuff. It is hard to budget, difficult to measure, and almost always the first thing to get cut. The data says that is exactly backwards.
There is a further contradiction in the findings. Executives are nearly twice as likely to increase technology spending as to invest in workforce development. Despite this, less than half say technology is very effective at improving adaptability. They are spending most on what is delivering least. Eighty-one per cent say their boards have raised expectations for organisational adaptability, and most of those organisations are not structurally positioned to meet it.
An adaptive operating model does not emerge from a technology deployment. It has to be designed. The organisations that will get this right are not the ones with the most sophisticated AI roadmaps. They are the ones building what I would describe as a human-centric, AI-enabled, continuously adaptive operating model.
That framing matters, and it is worth unpacking.
Continuously adaptive means the operating model is not a fixed design that gets revisited every three years. It is a living architecture, built to evolve as the technology evolves, as the market shifts, and as the organisation learns. This is a departure from how most organisations have approached design. They have treated structure as something to be settled. What is now required is structure that can evolve.
AI-enabled means the technology is not layered on top of existing ways of working in the hope that it will improve them. It is integrated into how decisions get made, how work gets coordinated, how capability gets deployed. That integration requires a clear-eyed understanding of where the organisation is today, not an idealised version of it. It also requires something most organisations have underinvested in: the business context, decision logic, and working assumptions that people carry in their heads need to be surfaced, defined, and documented. AI cannot operate on tacit knowledge.
Human-centric is the part that most AI strategies are underinvesting in, and where I want to spend a moment.
Human-centric is not what most people think it means
In most AI programmes, human-centred engagement means involving people in the process. It means running workshops, gathering feedback, and communicating change. These things matter, but they are not sufficient, and they are not what the moment demands.
What human-centric actually requires is a different quality of leadership. It means putting serious thought into the role of people. What work do they do today, what can be safely handed off to technology for improved efficiency, and what can’t. What will this additional capacity mean for your people. Where can they focus their unique human value when the mundane, repetitive, and unrewarding tasks are removed. For those whose roles will be replaced by technology, what does it mean. Technology advancement has always rendered some roles irrelevant, but we should be deliberate and not cavalier in our planning for them.
People are not failing to engage with AI because they lack information. They are hesitating because they do not trust their organisation’s plans. The media is feeding into this, creating fear about what this means for their work, their roles, and their future. That trust is not rebuilt through a communications plan. It is rebuilt through the quality of the conversation.
It means having more open, more honest, and in some cases more vulnerable conversations than organisations are accustomed to having. Conversations about what is coming, and what is not yet known. Conversations about what the commitments to people actually are, and where there is genuine uncertainty. Conversations that do not paper over ambiguity with confident-sounding messaging.
This is a fundamentally different design challenge from anything most organisations have encountered before. It combines technical integration, structural redesign, and a quality of human leadership that is rarely trained for and rarely rewarded. Getting it right requires all three.
The organisations I worry about are not the ones moving too slowly. They are the ones moving quickly in the wrong direction: deploying tools into structures that cannot absorb them and calling that transformation.
The question to ask is not “how do we implement AI?” It is “what does our organisation need to become, so that AI can do what we are hoping it will do?”
I am curious: where are you finding the human side of this hardest to get right? Is it the leadership conversations, the structural questions, or something else entirely?
If you are trying to figure out what this means for your organisation, a good place to start is the Operating Model Health Check.