Successful AI adoption hinges on organizational redesign and an enabling culture

At a time when AI investments are soaring, implementation outcomes don’t always meet expectations. Vamsi Srinivas, partner at Aon, explains that this mismatch often stems from insufficient attention to the “other half” of the implementation mandate: organizational redesign and an enabling culture.

The rise of AI

In the past twelve months, the UAE federal government has committed to delivering half of its services through autonomous AI agents within two years, placed a non-voting AI advisor in the Cabinet and on the boards of every federal entity, and given Dubai government twelve months to integrate every individual- and business-facing service into a single digital ecosystem.

In Saudi Arabia, PIF is weaving AI into the fabric of its giga-project portfolio, while SDAIA pushes the national AI strategy toward a $20 billion ecosystem. Singapore’s trajectory is similar. In Budget 2026, the Prime Minister announced a National AI Council to drive AI missions across four sectors, deployed agentic AI in public services (the first government in Asia to do so) and merged the country’s jobs and skills agencies into a single entity to align workforce planning with AI deployment.

This is not just a UAE, Saudi or a Singapore story. Every organization deploying AI seriously right now is heading toward the same structural questions. The technology to do all of this exists. The harder question – and the one most leadership teams are not yet asking – is what are the organizational and human capital implications of this?

At Aon, we see a consistent pattern across our engagements: organizations are focusing on half of the AI roadmap. The other half – the structural redesign, the new and eliminated jobs, the governance changes, the reward implications, and the cultural shift that holds it all together – is being treated as a downstream HR problem or deferred until the technology settles.

That sequencing gets it wrong. By the time the technology lands, the organizational chart the AI runs inside has not been redrawn. The jobs it has eliminated, created and merged have not been redesigned. The reward system meant to drive performance through it still pays for activity rather than outcome. And the culture that determines whether any of these changes stick is operating on assumptions the AI has just broken.

The consequences of getting this sequencing wrong are no longer hypothetical. Klarna, the Swedish fintech, cut its workforce and replaced hundreds of customer service agents with AI chatbots. At first, it looked like a win – resolution times dropped, revenue per employee climbed. But before long, customer satisfaction fell and the CEO publicly admitted the company had pushed too hard. Klarna is now hiring people back.

Klarna’s mistake was not the technology. It was changing the workforce without rethinking accountability, the job design or the reward system around it. The AI did what it was supposed to do. The organization around it was never redesigned to keep up.

The key lesson? AI is not new software being layered onto an existing organization. It is a different operating environment, and the organization must be reshaped to work inside it.

The Four Pillars

For AI adoption to be successful, four pillars need to be redesigned together: Organizational Design, Job Architectures & Skills, Rewards, and Governance.

Get one right and leave the others unchanged, the change does not hold. Culture is the catalyst that ultimately decides whether any of it sticks.

1) Organizational Design

This is the layer where most of the value gets won or lost. It is also the layer most consistently underestimated, because organizations tend to treat structure as a downstream consequence of technology rather than something that sits alongside it. Most organizations are over-structured for the new operating environment. The shape changes in three main ways.

Layers Compress
Public-sector organizations in the GCC typically run six to eight layers between the front-line and the CEO. Private-sector organizations that have moved on AI in earnest – even before the agentic wave - increasingly run four to five. Dubai’s April 2026 directive, by collapsing dozens of citizen-facing services into a single integrated journey, will require an organizational compression of a similar magnitude across the ecosystem.

The temptation is to remove middle layers first because they are the most visible cost. This is usually the wrong move. Middle layers exist for two reasons: they coordinate work below them, and they translate strategy from above into operational direction. AI takes on much of the first but very little of the second.

The right sequence is to identify which layers were primarily coordinating (those go first) and which were primarily translating (those stay or get redistributed). This is harder than it sounds because most middle managers do both, and the proportions vary by function.

Spans Widen
The classical span-of-control literature, going back to Urwick and Graicunas, suggested optimal spans of five to seven for complex work. The reasoning behind that number is worth understanding, because it is exactly what AI disrupts.

In 1933, Graicunas argued that a manager does not simply manage n people – they manage the full web of relationships those people generate. He broke this into three types:

• the direct one-to-one links (n);
• the interactions between subordinates the manager has to stay across (n(n − 1));
• the exponential combinations of the manager relating to every possible grouping of subordinates (n(2^(n−1) − 1)).

The total: R = n(2^(n−1) + n − 1)

At five subordinates, this gives 100 relationships. At six, 222. Each additional person does not add complexity, it multiplies it. The formula models how relationship complexity scales – not what a manager experiences day to day. The point that matters is that managerial load grows non-linearly with span. When AI enters the picture, it does not change the mathematics. It changes who carries the load.

An agentic system that monitors tasks and routes decisions to the manager only when judgment is needed carries most of that coordination weight. Routine coordination – scheduling, status alignment, workload balancing – is partially absorbed. The direct relationships that matter most – coaching, development, calibration – stay with the manager. AI can prepare the conversation; it cannot replace it.

The net effect is that AI compresses the experienced complexity of a span of ten or twelve closer to what a span of five or six felt like before augmentation. The technology is not expanding the manager’s cognitive capacity; it is reducing the cognitive demand.

A single span limit for the whole organization is unlikely to work – it would be prudent to set spans by assessing which categories of relationship complexity AI has absorbed in each function. A span of fifteen in a function where AI handles workflow may be less demanding than a span of eight where the work is largely unaugmented.

Functions Consolidate, Disappear or Emerge
Some functions shrink: transactional shared services, basic case processing, tier-one support. Some merge – most commonly Digital and IT, but increasingly HR analytics, workforce planning and organization design collapsing into one function.

Some disappear entirely where AI removes the handoffs they existed to coordinate. Some grow: model governance, AI ethics, AI product management. The shift is happening across both public and private sectors; the UAE’s agentic AI plan simply makes it an operational requirement on a public timeline.

The question that should drive every consolidation decision is what the function is for, not what it is currently doing. A function whose purpose is to coordinate work between two other functions becomes a candidate for elimination if AI now does the coordination. A function whose purpose is to make judgment calls under uncertainty should stay.

There is a deeper implication. As functions consolidate, the coordination gaps between them do not disappear – they just become harder to see. AI makes integration technically easier, but integration only happens when someone’s performance depends on it. To close the gap, name a single owner for each cross-functional outcome, with real decision authority. Structure follows accountability, not the other way around.

The Cultural Angle
Every structural change above asks managers and leaders to give up something they are used to controlling and to coordinate across something they used to ignore. This is the move from control to coordination, and it is the change most current managers are least prepared for. Promotion patterns, training programs and succession planning all need to reflect that the new managerial job is fundamentally different – not just a variant of the old one.

2) Job Architectures & Skills

Aon’s AI Sensitivity Analysis decomposes each job into its constituent tasks and scores them across fourteen dimensions covering automation, content synthesis, digital assistance, monitoring and related categories. The output is a heatmap by job family showing where AI augments, where it automates and where human judgment remains essential, with concrete numbers attached. Without this, an entity may end up focusing on the wrong priorities.

Once impact is sized, every role falls into one of four cohorts: automated (eliminated entirely), augmented (reshaped), created (new roles that did not previously exist), and repurposed (redeployed to different work). The augmentation cohort is usually the largest.

Each cohort has an appropriate response, structured as the 4Rs: Reimagine, Reskill, Restructure, Relocate. The percentages are illustrative; the Sensitivity Analysis is what calibrates them entity by entity, function by function.

The Chief AI Officer: Case in Point
The Chief AI Officer is a good example of a role in the third cohort, and a role that will be increasingly pivotal for driving success of AI augmentation. Organizations are conflating this role with two adjacent ones.

• It is not a CIO with a new title. The CIO owns systems and infrastructure; the Chief AI Officer owns AI strategy, deployment, value capture, model risk and AI ethics.
• It is not a Chief Digital Officer. The CDO owns digital transformation as a programme; the Chief AI Officer owns AI as a durable operating capability and as a category of risk.
• It is not the non-human advisor at board level. The Chief AI Officer is the human accountable for what the entity does with AI, and the natural counterpart to that advisor.

The role should report to the CEO with a dotted line to the Chair, and own the human-AI decision protocol end-to-end. Its accountabilities and the boundary with CIO and CTO should be set out in writing before the role is filled.

Workforce Composition: The Size and Shape Question
The job architecture answers which roles exist. But there is a second question that carries most of the headcount impact: how many positions are mapped to each role?

In our experience, organizations that run the AI Sensitivity Analysis typically find their target workforce is meaningfully smaller – sometimes substantially so in transactional-heavy functions, less so in judgment-heavy ones. Most focus on which jobs disappear and undercount the quieter question: how many fewer people do we need in the jobs that remain?

DBS Bank in Singapore is reducing approximately 4,000 temporary and contract roles over three years as AI absorbs routine work, while creating 1,000 new AI-related positions. The composition shifts matter more than the size – judgment and oversight work goes up, execution and processing goes down.

Workforce planning needs to move from an annual cadence to being continuous. Where AI is being rolled out function by function, workforce needs change every few weeks. An annual planning cycle cannot keep up. Organizations need the capacity to act on real-time surpluses and shortages, not wait for next year’s budget round.

Skills as the Enabler Underneath
Skills influence how individuals operate inside the new jobs, and the basis for moving people between jobs. Most large organizations, particularly in the region, are not yet ready to operate as fully skills-based organizations. The prudent sequence is: redesign the jobs first, then the workforce composition that fills them, then the skills strategy that develops the people in them.

Aon’s A-EYE Wheel – twelve transferable AI-era skills across leadership, agility, data and AI, and ecosystem collaboration – is a useful framework for identifying which skills matter most in the new environment. Within them, the data, AI and ecosystem clusters are the scarcest, and the ones that should anchor any skills strategy.

A recent study from Aon found that only 35% of the employees globally feel motivated to acquire new skills in response to AI. Skills enablement is as much about creating an environment where people want to learn as it is about the frameworks themselves.

The Cultural Angle
The job architecture changes only do their work if hiring, promotion and development decisions actually start to value the new capabilities. If tenure trumps demonstrated capability, every one of these new roles will be filled by someone who interviews well and cannot do the work. No amount of structural redesign survives that

3) Rewards

This is the layer most often deferred and most often regretted. Organizations redesign the structure, rebuild the jobs, invest in the technology – and often leave the reward system untouched. Pay and performance systems must follow the operating model, not lag it.

The lag is where most transformations quietly die. The Organizational Design layer determines who owns the outcome. The reward system determines whether anyone has a reason to deliver it.

Performance Management: From Activity to Outcome
Work done by human-AI teams cannot be sensibly measured in volumes, transactions or hours. A team that processes 40% fewer cases but delivers faster resolution, fewer errors and higher satisfaction is outperforming – but most existing systems would score them as underperforming because the metrics still count activity. What matters is the outcome the team delivers and what each person contributed to getting there.

Goal-setting cadence has to accelerate. Annual targets set in January are stale by March in an environment where AI capabilities are being rolled out function by function. To drive impact on the AI transformation, goals should be set quarterly, and reviewed monthly.

Pay for Critical Skills and Roles
Every organization going through this transition needs to identify the roles that will make or break it – scored on mission criticality and vacancy risk. The Chief AI Officer is the most visible, but the head of model governance, AI product leads and integration architects all carry outsized impact relative to their grade.

The skills behind these roles – particularly the data, AI and ecosystem clusters – are scarce. The regional pool is thin, global competition is fierce, and paying for tenure and credentials does not work when someone’s market value is set by what they can do, not how long they have been doing it.

Compensation for these roles will often require explicit separation from the rest of the band. A critical role at grade X may warrant 30% to 50% more than a non-critical role at the same grade. That is a feature of the system, not a distortion, and it is a mindset most leadership teams still need to adopt.

The cost of mispricing is not the salary differential. It is the milestones these people own. Losing a Chief AI Officer to a competitor is not a recruitment problem – it is six to twelve months of programme momentum walking out the door.

Executive Compensation
The executive incentive plan has to be recalibrated against transformation delivery, as a concrete design change. This means weighting a meaningful portion of executive short-term incentives on integration milestones: services migrated, AI adoption rates within the workforce, measurable shifts in operating efficiency, and critically, the human capital milestones (such as, workforce readiness scores, critical role fill rates, skill acquisition targets, and employee experience metrics) that shadow each technology milestone.

The trap is treating AI transformation KPIs as minor modifiers on the incentive design. A five per cent weighting on “digital transformation progress” signals the board considers it a side project. The weighting has to be large enough that missing the milestones materially affects the payout – and the milestones must be specific enough to resist gaming.

“Launched three AI pilots” is an activity metric. “Reduced average citizen service journey from fourteen days to three” is an outcome. The incentive plan should only reward the latter.

For listed companies, multi-year performance share units vesting against AI transformation milestones create genuine skin in the game over the three-to-five-year horizon that this kind of change actually requires. For organizations without listed equity, including most public-sector entities in the region – deferred cash plans tied to the same milestones achieve a similar effect, provided the deferral period is long enough that the executive cannot collect and leave before the transformation has actually landed.

The Cultural Angle
Reward systems are the most visible expression of what an organization actually values – more visible than any strategy document, any town hall speech, any set of corporate values printed on the wall. Employees decode what matters by looking at who gets paid, who gets promoted, and who gets recognized. And why. Everything else is noise.

Aligning pay, performance, incentives and recognition with the new operating model is not the last step in the transformation. It is the step that makes every other step credible.

4) Governance

The arrival of AI in the boardroom is the change that most directly touches an organization’s highest-stakes decisions, and the one with the thinnest existing playbook.

The UAE has made this concrete. Since January 2026, the National Artificial Intelligence System has sat as a non-voting advisor to the Cabinet and the boards of every federal entity. It analyses, simulates and challenges in real time, bringing evidence no human director can match in volume or speed.

For organizations not under a federal directive, the same change is happening less visibly – AI is already embedded in the analytics, the pre-reads, and the materials directors rely on. The question of how the organization governs itself with AI in the room is no longer hypothetical for anyone.

The boardroom has changed. Four changes have to be made in tandem.

Charter and Committee Redesign
Board and committee charters were not written with an AI advisor in the room. They need to be – covering how AI input is recorded, how directors challenge it, what counts as an adequate response to AI-generated analysis, and what happens when somebody decides to escalate.

Most organizations will need a standing AI and Technology Committee with explicit links to Risk and Audit, because the duty of inquiry now extends to AI-generated analysis. The Audit Committee’s traditional remit, which was to verify what management presents – has to expand to verifying what the AI presents, which requires different skills and different evidence.

The Dual Capability Profile
The leadership profile required in this environment comes down to two things working together. First, critical thinking - the ability to read AI output, interpret it, and place it in context. Second, AI capability – understanding the technology well enough to use it and to oversee it.

One without the other does not get you very far. A director with strong judgment but no AI capability becomes a bottleneck. A director with AI fluency but weak critical thinking accepts AI conclusions that should be challenged. The future-ready board has both, in combination.

The matrix is the basis on which director assessment, Chair effectiveness reviews and senior leader development should be calibrated. Most existing assessment frameworks do neither dimension justice.

The Decision Protocol
Most existing decision-making frameworks – RACI, RAPID, governance hierarchies – were built for human-only decisions. They need adapting for human-AI decisions: who is accountable when the AI’s analysis informed the decision, what records are kept of the AI’s input, and how the protocol changes when the AI’s confidence is low or its evidence is contested. This is not a board secretariat exercise. It is a redesign of how the organization governs itself.

Board Performance Evaluation
The arrival of an AI advisor changes how a board should be assessed. Director effectiveness reviews need new dimensions: the dual capability profile tested in combination, quality of engagement with AI-influenced decisions, and rigor of challenge to AI-generated analysis. The Chair sets the tone for whether AI input is genuinely integrated into deliberations. Both should be explicitly evaluated, and the framework should be in place before the first full AI-advised board cycle, not after.

The Cultural Angle
The dual capability matrix is hard. But, the harder change is the move away from valuing hierarchy over evidence. A board that lets the senior director’s view prevail over AI evidence when the two disagree has not actually changed how it governs. The matrix will not work without the shift in culture.

Key Questions Leaders Should Be Asking

The diagnostic that distinguishes organizations that will be ahead of this transition from those that will be behind it.

• Have we run the AI impact assessment, and do we know where each role falls across the four cohorts?
• Where do our directors and senior leaders sit on the dual capability matrix, and what is the plan for those who are not yet where they need to be?
• Have we made an honest assessment of which cultural attributes that drove our success historically are now in the way?
• Is our job architecture being edited or rebuilt? If edited, what is the basis for believing a traditional grade-anchored structure will deliver the new operating model?
• Have we identified our ten to fifteen critical roles using a defensible mission-impact and vulnerability method, and what proportion have ready-now successors?
• What is our Chief AI Officer design – reporting line, RACI alignment, accountability boundary with CIO and CTO, compensation design – and against what benchmark are we setting it?
• Is variable pay actually differentiating between teams delivering the new value and teams that are not, or is bonus still expressed as a percentage of base anchored to grade?
• How is our board itself being evaluated on its conduct alongside AI-generated analysis, and when is that assessment first due?
• For organizations under regulatory deadlines: how are we tracking against each, and who owns each? For organizations not under regulatory deadlines: what is our internal deadline, and who is accountable for it?

The organizations that win across this transition will not be the ones with the most advanced technology. They will be the ones that recognized the importance of both halves.

The recommendation is direct. Pair every technology work stream with an explicit human capital work stream, with a single budget and a single board dashboard. Begin with an honest task-level analysis of where AI lands in your organization.

Redesign the operating model and the workforce together. Rebuild the reward system to follow the model, not lag it. And do the cultural work in parallel with all of it, because without it, none of the rest will stick.