For CIOs and transformation leaders, the practical issue is not whether AI can create value, but how to stop it fragmenting into dozens of uncoordinated efforts. A maturity model is useful only if it changes funding, ownership and review cadence. The real management question is whether AI sits inside existing functional silos or is governed as a cross-enterprise capability with common standards for data, risk, workflow design and value tracking.
The biggest trade-off is speed versus repeatability. Early productivity wins are easy to celebrate but hard to scale if each team chooses its own tools, prompts, controls and metrics. That creates local uplift, yet no reusable operating model. Leaders should therefore ask which parts of the AI portfolio are intended for broad adoption, which are experiments, and which deserve process redesign. Without that separation, pilots can crowd out the harder work of changing processes, incentives and accountabilities.
Another implication is that AI readiness is as much a people and governance problem as a technology one. If employees are expected to embed AI in daily work, then training, line-manager support and clear guardrails become delivery dependencies, not optional extras. Decision-makers should be asking: Who owns model risk, workflow change and benefits realisation? What evidence will justify moving an initiative from pilot to standard practice? Which metrics distinguish genuine capability building from isolated automation?
The organisations most likely to progress are those that treat AI as a managed transition, not a set of tools. That means building a portfolio view across productivity, process and transformation bets; assigning accountable process owners; and reviewing whether the operating model is actually absorbing change. In practice, the maturity test is simple: can the business reuse what it learns, or does every new AI use case start from zero?
Many executives talk about AI as if itโs a force that will arrive and transform the business on its own. In the philosopher Daniel Dennettโs terms, theyโre waiting for a skyhook: a miraculous lift that suspends the hard work of change from nowhere, requiring little redesign of roles, workflows, incentives or governance.
But AI deployment isnโt only about deciding which technology to purchase โ itโs an organisational change project. The highest level of AI maturity, where true strategic differentiation arises through new and defensible business models, canโt be reached without mastering the foundational stages. In short, organisations need to climb what we call a maturity pyramid.
Level 1: Individual productivity
At the base of the pyramid, AI enhances individual productivity. Knowledge workers use generative AI to do everything from drafting reports and summarising documents to analysing spreadsheets and generating code. Engineers rely on coding assistants, marketers use AI to produce campaign variations and analysts automate initial research.
Besides speeding up tasks, AI can improve the quality of output and decrease the cognitive load of users. But although the impact is real, gains are localised. AI use improves efficiency at the margins without fundamentally changing how the organisation creates value. And since such tools are widely available, competitors can adopt them quickly. The result is cost parity, not differentiation.
Level 2: Group productivity
The second level shifts the focus from individuals to teams as AI becomes embedded in collaborative workflows. At this level, systems summarise meetings automatically, support deliberation and consensus formation, track action items, route tasks across departments and retrieve institutional knowledge in real time. Information flows more smoothly and coordination improves, letting teams move faster.
As with level 1, competitors can implement similar solutions. Although collaboration becomes more efficient and effective, the underlying business model remains unchanged, and the organisation becomes better but not fundamentally different.
Level 3: Business process efficiency
At level 3, AI is integrated into core operational processes: think banks automating underwriting decisions, retailers using demand forecasting and manufacturers deploying predictive maintenance systems. These gains can be significant as costs decline, error rates drop and responsiveness improves, allowing organisations to scale more effectively.
Take Yinson, an energy infrastructure and technology company. It’s embedded a strong digital core across its operations, integrating real-time operations data and making it usable across various workflows. AI helps the company improve efficiency and reduce accident rates by offering real-time analytics for predictive maintenance, route optimisation and automated management.
But even this level is largely defensive. It lets firms compete more efficiently within existing industry structures without necessarily altering what customers value or how revenue is created.
Level 4: Business model transformation
While success at the first three levels of the pyramid is a source of advantage, itโs likely to be transient, as itโs more easily replicable. The highest level of the pyramid is qualitatively different. At this stage, AI doesnโt simply improve how work is done but reshapes how value is created and captured.
Consider Rolls-Royce, historically a producer of jet engines, which now increasingly sells โpower by the hourโ. Airlines pay for uptime, not ownership. This performance-based model predates AI, but predictive analytics and AI-enabled monitoring make it viable at scale. The company shifted from being a capital-equipment manufacturer to a long-term service partner with recurring revenue.
Or take banking and financial services firm DBS. Since 2014, it systematically built AI and data capabilities into its operational core. Then, a qualitative change happened, emerging from the operational and data capabilities DBS built patiently across years. The bank began launching consumer marketplaces for the likes of cars, property and travel, generating revenue from facilitating major transactions rather than just financing them. DBSโs business model and competitive positioning fundamentally changed.
Competitive advantage rests not only on algorithms (which are increasingly considered a commodity), but on data, redesigned workflows, human capital and organisational capabilities that competitors canโt easily copy. Thatโs why level 4 cannot be mandated into existence โ it emerges from mastery of the levels below.
Why firms stall before transformation
Many leaders are tempted to leap straight to transformation. However, vision alone, in the absence of capability, canโt reshape the business model. Without widespread AI fluency and operational integration, there is the risk of top-down strategy becoming disconnected from execution. Most importantly, without buy-in from employees, no change initiative โ which is what AI adoption is โ can succeed.
Yinsonโs leadership didnโt treat AI as a pilot or a collection of disconnected experiments, but as an organisational journey: build broad fluency and workflow muscle first (levels 1 and 2); use that base to redesign core processes (level 3); and, only after that, make selective large-impact business-model bets (level 4). Had leadership jumped straight to a large transformation bet, it would likely have stalled due to insufficient distributed AI fluency.
At level 1, Yinsonโs emphasis was on widespread individual adoption and capability-building. In our survey of over 300 employees, respondents reported using AI across 6 out of 10 work activities on average. This breadth matters because it creates shared fluency and normalises experimentation as a default way of working.
At level 2, the focus shifted from โI use AIโ to โwe use AI in how we work togetherโ. Examples cited by employees include embedding AI into team routines, such as generating post-discussion action items and handover templates. Recognising these patterns and automating such repeatable tasks turn isolated productivity wins into reliable collaboration practices.
With that foundation in place, the company could expand into level 3 opportunities that sit within operational and functional workflows. In fact, among Yinson employees who proposed substantive AI opportunities, around 26% pointed to ideas that scale up current operations or make them more efficient. This solid base provides Yinson a pathway to level 4, where AI reshapes value creation and capture by enabling differentiation-oriented activities.
Top-down or bottom-up adoption?
A common debate around AI adoption is whether it should be top-down or bottom-up. But this is a false choice, because the answer is both โ though with distinct roles. The problem with the skyhook is not that it comes from above; itโs that it has no basis to be hung.
In reality, the value of AI requires an anchor on which to be mounted: one built on clear decision rights, trusted data pipelines, model governance, accountable process owners and metrics that distinguish novelty from impact. When those anchors are missing, top-down AI initiatives become theatre and bottom-up efforts remain local experiments that donโt translate into organisational capability.
Leadership must create enabling conditions, which include investing in widespread AI literacy and infrastructure, setting guardrails and defining criteria for success in scaling initiatives. Without this scaffolding, experimentation remains fragmented. At the same time, innovation can also flow upwards, whereby employees experimenting at the individual level generate insights and teams embedding AI into daily workflows identify improvement opportunities.
Seen this way, the crux of the question isnโt about top-down or bottom-up; itโs about which level of the pyramid your company is currently on, what must be built to reach the next level, and who owns and leads that process.
Advantage comes from strong foundations
Consider Accenture. Its AI journey didnโt begin with proclamations about radically new consulting business models. Instead, the company announced that all 700,000 employees would become AI literate. Why? Because even for a firm that helps others transform, level 4 rests on levels 1, 2 and 3.
Before AI can reshape client offerings, it must reshape how consultants work. Before it can redefine value capture, it must become embedded in everyday workflows. Before strategy shifts, behaviour must change.
In AI, as in mountaineering, you donโt leap to the summit. You climb.
Edited by:
Rachel Eva Lim
About the author(s)
Vivianna Fang He
is an Associate Professor of Strategy and Entrepreneurship at the School of Management, University College London.
Phanish Puranam
is a Professor of Strategy and the Roland Berger Chaired Professor of Strategy and Organisation Design at INSEAD. He serves as Academic Director of the INSEAD-Wharton alliance and leads the AI for Business and Transforming Your Business with AI programmes.
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.