By Stuart J. Green
AI makes existing operating models run faster, so what was already misaligned now misfires at machine speed, and the window for structural redesign is closing.
Large organisations are installing AI on operating models designed for conditions that no longer hold. AI is exposing the structural misfit and is actively changing the conditions the next operating model has to fit. Leaders who diagnose now will design AI programmes they can govern. Those who defer will not.
When a fisheries regulator deployed AI to process catch data, the system flagged thousands of anomalies a day within six months. The first response was that the AI had malfunctioned. It had not. The prior reporting system had been absorbing the anomalies invisibly for years. AI made the condition visible. The condition was structural. Most organisations are not yet seeing the same pattern.
AI as a diagnostic event
In fire ecology, some events do not create new conditions. They reveal conditions the prior structure had been holding hidden. Fire in certain forest systems is one such event. The pre-fire structure sealed capacity inside cones that required intense heat to open.[1] The fire did not build the next forest. It exposed what the prior one had been carrying.
AI in large organisations is operating the same way. It is not building the next operating model. It is exposing the one already in place. It is also changing the environment the next model will need to fit.
The data converges. Bain reports that 88 per cent of business transformations fail to achieve their original ambitions.[2] McKinsey’s 2025 State of AI finds that only 39 per cent of organisations report any measurable EBIT impact from AI, and roughly 5 per cent qualify as AI high performers.[3] The 2025 AI Governance Benchmark found that 58 per cent of leaders cite disconnected governance as the top obstacle to scaling AI.[4] The pattern is not a technology problem. It is a diagnosis problem funded as a technology budget. Organisations are installing the technology on operating models that were never examined against current conditions first.
Misfit at machine speed
AI produces a magnification effect. If a process carries a hidden assumption that no longer holds, AI makes its consequence arrive faster. If a process sits inside an incentive structure rewarding the wrong behaviour, AI scales that behaviour. If the governance layer above is overloaded, AI produces more output than it can absorb.
Three symptoms are now visible across industries.
Incentive amplification. AI rewards whatever the system was rewarding, faster and at higher volume. A sales function hits its numbers on customers the firm will regret onboarding. A grants function processes applications faster while drifting from its mandate.
Governance overload. AI generates more decisions and exceptions than the governance layer can process. A board calibrated for 2010 conditions cannot govern by exception when volumes have multiplied. Board papers swell. Risk registers outpace the risk committee.
Capital exposure. AI investment is approved on the assumption that business models will generate the cash flows that justify the spend. Where those models carry stranded assumptions about customers, regulation, or mandate, the AI investment compounds the exposure rather than resolving it.
Chris Argyris, whose work on institutional learning shaped modern systems thinking, drew the relevant distinction.[5] Single-loop learning adjusts behaviour within existing assumptions. Double-loop learning examines the assumptions themselves. AI is a single-loop technology. It runs the existing system faster. It does not question whether the system is the right one. When the problem is structural, AI runs in the wrong loop.
The pace compounds the problem. Regulatory, technological, and capital environments now shift faster than the five-to-ten-year cycles most operating models were designed against. As AI adoption scales across a sector, it raises the minimum operating standard for every institution in that sector. By the time the misfit appears in financial results, the window for redesign has narrowed. Diagnosis cannot wait for certainty.
What structural redesign looks like
The organisations that succeed do the structural work first and then let technology amplify a fitted model rather than a misfitted one. LEGO is the canonical example. By 2004 the company was approaching insolvency, carrying a bloated product portfolio, a misaligned distribution architecture, and a brand position built for a consumer environment that had moved.[6]
The response was structural and sequential. First, clearing: the product line was cut by a third and obsolete commitments retired. Second, preparation: the supply chain was redesigned, governance simplified, and capability built for the new model. Third, design: identity was anchored around systematic creativity rather than catalogue breadth, and the next operating model was resourced. Only then did digital investments produce returns. The company recovered market leadership over the following decade.
The regenerative redesign framework generalises this pattern. Its three sequential phases are RAZE (clearing what no longer fits), ENRICH (preparing the substrate the next model will grow on), and GROW (designing and resourcing the new operating model). Most institutions skip the first two. They try to grow what they have not cleared, on a substrate that has not been prepared. Technology cannot compensate for a skipped phase. It compounds the skip and amplifies the misfit.
Three questions for the next strategic review
The AI conversation at leadership level is usually running in the wrong direction. The common question is what AI can do for us. The more productive question is what AI is now exposing about the operating model already in place, and what it is changing about the conditions the next model has to fit.
Three questions make the diagnostic workable at a board table, and each maps to a phase of the structural work.
The clearing question. Where is AI producing friction that predates AI? If a function fails after deployment, treat the friction as diagnosis. AI has made a pre-existing operating condition visible. Something that should have been retired remains.
The substrate question. Where is AI producing output the governance structure cannot absorb? If leadership is overwhelmed by the volume of AI-generated decisions, the governance architecture is obsolete, not the technology. The substrate was not prepared before the new layer was added.
The design question. Where is AI making the mission question unavoidable? AI forces choices about where to direct organisational capacity. Those choices force a mission conversation that has often been deferred for years. If the mission cannot answer the AI question, the mission is what needs redesign.
Diagnose before you deploy
AI decisions are structural decisions. They test whether the operating model can survive a compounding environment. Most cannot. Fire does not build the forest that comes after it. Fire reveals which seeds the prior structure was carrying. What the organisation does with that revelation, what it clears, what substrate it builds, what it chooses to grow, is the work. AI is a magnification layer on that work, not a substitute for it. Organisations that diagnose first will design AI programmes they can govern.
Acknowledgements
The author drafted this article independently. AI-based editing assistance (Claude, Anthropic) was used to check structural clarity, verify the accuracy of cited statistics, and support final editorial review. All analysis, argument, and interpretation are the author’s own.
Stuart J. Green
