The data behind our thesis.

The AI shift has generated a lot of activity, but so far little financial return for most companies. McKinsey concludes that the real AI advantage comes from redesigning how work gets done and how decisions are made, not by adding AI to an existing structure. Success ultimately depends on people and operating models supported by technology, not the other way around. Organisations that do this well reorganise their structure around the question of which decisions can be made faster or better, and only then build the technology around it.

The core message: AI is not a separate tool but a teammate. Where leaders once described AI as a personal productivity aid, they now talk about agents with their own names, tasks, and responsibilities that work around the clock. The pattern is that roles are being restructured as collaboration between humans and AI, with work redistributed so teams can take on more. An example from the research: a team that previously handled 50 cases at a time now handles 3,000, not by hiring 60 times as many people but by giving AI the routine work and people the conversations that truly matter. McKinsey emphasises this is not an efficiency push but a fundamentally different business model.

McKinsey’s research across thousands of executives identifies change management and siloed ways of working as the biggest obstacles to scaling AI, ahead of technical infrastructure. Traditional change management, one-off training, and top-down rollout create awareness but do not build the trust and habits needed to sustain new ways of working. Four fears consistently recur among employees: fear of falling behind, fear of responsibility without control, uncertainty about the direction, and fear of losing professional value. Leaders who name these fears and explicitly design around them progress further than leaders who treat them as obstacles.

Nearly 90 per cent of companies now use AI regularly, but only 37 per cent of respondents attribute actual EBIT impact to AI. The gap between leaders and the rest is widening: 60 per cent of market leaders report double-digit revenue growth, compared to 21 per cent of laggards. Organisations that progress furthest are three times more likely to fundamentally redesign individual work processes rather than add AI to what already exists. An example: Reckitt deployed AI-driven scenario planning for pricing and assortment decisions, unlocking more than $500 million in new revenue. McKinsey’s conclusion is that leaders who only set above-average growth goals are not ambitious enough: with AI as the starting point, nearly three times as much value is achievable as most companies imagine.

McKinsey identifies a real challenge: the tasks through which starters once learned their craft are disappearing or changing significantly due to AI. At the same time, uncertainty among early-career professionals is growing rapidly: the share of graduating students pessimistic about their career start rose from 46 to 62 per cent in two years. How much of this shift is truly attributable to AI is not yet fully established, as remote working and other factors also play a role. For organisations, the question is no longer how many graduates they hire, but what those roles are for. McKinsey outlines a four-part approach: knowledge management, role design, learning on the job itself, and active mentoring, so that starters learn to design and direct AI systems rather than simply work with them.

BCG’s fourth annual research shows that organisations that fully redesign work processes score better on both realised value and job satisfaction than organisations that only roll out tools, and that gap widens every year. 72 per cent of respondents say the expected skills have changed, but only 36 per cent feel sufficiently trained. Only 33 per cent find that leadership communicates clearly about AI. BCG explicitly labels this no longer a technical question but an organisational and leadership question: the companies that score best on value are also the companies where employees experience the most job satisfaction. These are not opposing outcomes but are driven by the same choices.

TNO conducted 17 interviews with employees and employers across four diverse AI applications. At HelloPrint, large-scale automation of customer contact led to a significant headcount reduction in customer service. At Dutch Customs, where an AI application helps employees search the Customs handbooks, pilot participants saw clear time savings and improved information retrieval, but TNO stresses that hard productivity figures are still absent: much depends on how the freed-up time is used. At a.s.r., measuring AI impact on customer insight turned out to be more difficult than expected. The common thread across all four cases: AI applications can make processes noticeably more efficient, but the outcome depends strongly on how the surrounding work is redesigned, not on the technology alone.

Deloitte’s annual human capital trends research shows that most organisations (59 per cent) still operate with a technology-first approach to AI: tool first, then the work around it. That backfires. Technology is now available everywhere and therefore increasingly less differentiating, Deloitte concludes, while human adaptability, creativity, and judgment remain differentiating. Competitive advantage no longer comes primarily from technology choices but from how well an organisation aligns people and machines. Deloitte identifies three tipping points that make 2026 different from previous years: the pressures on organisations no longer arrive sequentially but stack simultaneously, causing the classic growth curve to accelerate and requiring organisations to jump to a new curve more often and more quickly.

Nearly nine out of ten organisations now use AI regularly, a massive increase from previous years. Yet progress remains unevenly distributed: nearly two thirds of organisations have not yet begun scaling AI across the whole organisation. 64 per cent see AI as a driver of innovation, but only 39 per cent report actual EBIT impact at the organisational level. The organisations that do succeed think beyond efficiency alone: they use AI to redesign workflows and accelerate innovation, with growth and innovation as explicit goals alongside cost savings. That is precisely the difference between organisations that add AI to existing work and organisations that redesign the work itself.

Everest Group’s research places enterprises at three levels: early explorers, structured experimenters, and strategic scalers. Most organisations are still in the first two categories. More than 90 per cent expect measurable performance improvement from agentic AI, but the report is clear about what that requires: success depends not on the technology alone, but on the combination of executive commitment, funding models, governance, and preparation of the organisation’s own people. Organisations that treat agentic AI as a long-term transformation rather than a series of disconnected pilots are the ones that manage to turn early experiments into lasting competitive advantage. The coming year to eighteen months will be decisive for whether enterprises remain stuck in controlled experiments or grow into a truly scaled, governed form of human-AI collaboration.

Celonis' third annual Process Optimization Report finds that 85 percent of organizations want to be an "agentic enterprise" within two to three years, while only 19 percent actually use multi-agent systems today. More than three-quarters say they currently get by with suboptimal processes. The biggest obstacle isn't resistance to change, only 6 percent name that as the main hurdle, it's operational readiness: lack of expertise, siloed departments, and difficulty getting AI to understand business context. 89 percent of leaders say AI only delivers ROI if it genuinely understands the organization's rules, KPIs, and structure. Celonis' conclusion is direct: without a clear view of how processes actually run, AI agents are guessing rather than contributing.

BCG built an independent measure of AI adoption, separate from what companies claim in press releases or surveys. Only 6 percent of the companies studied qualify as genuine leaders, but that group shows striking results: industry-adjusted shareholder returns 9.3 percentage points above the median, driven by revenue growth and margin expansion, not higher valuations. Notably, headcount at these leaders grows 3 percentage points faster than the average company. That challenges the assumption that AI leaders mainly shrink their workforce, among the companies genuinely getting it right, growth is the dominant path to value, not efficiency. Value also doesn't build up gradually but arrives in a step change: companies just below the top show almost no extra return, the gains concentrate sharply at the very top.

A year into the agentic AI wave, McKinsey draws a sober conclusion: getting it right genuinely takes work, and some organizations have even rehired people after agents fell short of expectations. The main lesson is that companies focus too much on how impressive an agent looks and too little on the workflow around it, resulting in good-looking agents that don't actually improve overall throughput. Organizations that do capture value redesign the entire workflow, people, process, and technology together, instead of automating a single task. Another lesson: training and onboarding agents deserves as much attention as onboarding a new employee, including a clear job description and continuous feedback. Without that investment, organizations quickly run into what McKinsey calls "AI slop": mediocre output that rapidly erodes user trust.

Microsoft's "Frontier Playbook" describes its own AI transformation methodology across five elements: anchoring business ambition, building a scalable AI operating model, codifying what makes the company distinct into its systems, investing continuously in people, and applying governance to every agent action before scaling. The document is explicitly cautious: it reflects Microsoft's current experience and perspective, not a guarantee of results for others. Recurring strategic questions that come up in every boardroom: how do you anchor AI to real business value, how do you move fast without losing what makes the company distinct, and how do you bring people along as you redesign their own work? It stresses that AI transformation is as much about people and governance as it is about technology.

The research builds a dynamic measure of how exposed a job posting is to generative AI, and breaks down how that exposure changes over time. Two mechanisms turn out to work together: reallocating where companies hire (52 percent of the change) and redesigning tasks within existing roles (39.5 percent). Senior roles adjust earlier and mainly through reallocation, while entry-level roles change through a mix of both mechanisms. The researchers conclude that labor-market adjustment to generative AI is primarily a process of organizational reconfiguration, companies reshape both who they hire and what a role contains, rather than simply a story of jobs disappearing.

Dutch IT company Team Rockstars IT studied how AI is changing the work of its own knowledge workers. AI now covers more than 30 percent of the work week, double what it was a year earlier, and more than 9 in 10 employees see the technology as an opportunity rather than a threat. At the same time, there's a clear warning, coming shortly after TNO flagged the same pattern: time saved only becomes real productivity and quality once organizations deliberately redesign the work around it. Without that redesign, the risk of "dumb productivity" looms, working faster without quality or value keeping pace. Just how embedded AI has become in daily work became clear when a brief Copilot outage caused visible unease among staff.