Transformation on steroids
Large organisations are currently committing to artificial intelligence at a speed that has few precedents in corporate change. Boards are setting cost targets tied explicitly to the technology, and external deployment providers are being given direct access to core processes. Whether this produces the expected returns will depend less on the technology itself than on whether the organisations’ management layers can absorb the change. The experience of earlier transformation programmes suggests caution.
Two cases
The first concerns a technology subsidiary of a large international bank. Its board has instructed management to reduce costs by 30%, a target attributed explicitly and solely to AI. The expectation at board level is that a digital business should be able to realise such savings faster than the rest of the group. The target has since been reflected in budgets, in the willingness to invest and in the cost expectations placed on the investments themselves. What has not accompanied it is an operational plan. Neither the subsidiary’s management nor that of the parent has so far translated the target into concrete changes in how units work, and discussions at management level remain too general for such a translation to emerge from them.
The second concerns a large European insurance group, which describes its AI strategy as settled. The group works directly with a large model provider, and engineers are moving through the organisation to identify and implement use cases. The pace is notable. Capable language models have been available since 2023, but systems that perform reliably within production environments are considerably more recent. Declaring a group-wide strategy settled within that period is, at a minimum, an ambitious claim.
A familiar pattern
The closest precedent is the introduction of enterprise resource planning software from the 1990s onwards, when vendors such as SAP gained comparable access to core processes. Implementation proved beyond the capacity of the vendors alone, and a substantial industry of integrators and consultants formed around them. A similar structure is now emerging around AI. Under its partnership with Anthropic, Accenture is training approximately 30,000 staff on Claude, including forward-deployed engineers who embed the technology within client organisations1; DXC has announced plans to train tens of thousands of such engineers2.
The working method of forward-deployed engineers resembles that of management consultants. They move through the organisation, examine processes and assemble a portfolio of use cases, which is then consolidated into a formal transformation programme. There is little reason to assume that such programmes will be exempt from the difficulties that affected their predecessors. The scale, however, is different: the momentum generated within the past six months is comparable, in the author’s assessment, to that of two decades of digital transformation.
The middle layer
The difficulty lies less in identifying use cases than in transforming the organisational structure into which they are introduced. Large companies have accumulated successive layers of middle management, and in many of them managerial status is closely tied to the number of people managed. A reduction in team size is therefore experienced not only as an operational change but as a loss of standing. Managers at these levels also have reason to question whether their own positions will survive the savings they are asked to deliver.
Resistance of this kind is seldom explicit. A senior executive at another bank has described it as passive: initiatives are approved, but little follows. Front-line staff are concerned about their jobs; lower and middle managers about their influence and, increasingly, about their jobs as well.
What the data show
The pressure is not confined to individual cases. Several banks have tied workforce reductions explicitly to AI. Intesa Sanpaolo announced in 2024 that it would cut 9,000 jobs by 2027, Mizuho plans to replace around 5,000 administrative roles over ten years, and Standard Chartered announced in May 2026 that it would reduce roles in its corporate functions by more than 15% by 2030, approximately 7,800 positions3.
For middle management specifically, Gartner predicted that through 2026, one in five organisations would use AI to flatten their structure and eliminate more than half of their current middle-management positions4. This is a forecast rather than a measurement, but it is widely cited and forms part of the context in which managers assess their own prospects. Gallup’s State of the Global Workplace 2026 reports that manager engagement has fallen by nine points since 2022, including a decline from 27% to 22% between 2024 and 2025, and that managers have largely lost the engagement advantage they previously held over individual contributors5. Gallup also finds that, aside from integration with existing systems, the strongest predictor of AI adoption is whether an employee’s direct manager actively champions it5; employees whose manager actively supports their team’s use of AI are 1.7 times as likely to use it a few times a week or more6. The data establish associations rather than causes; they do not show that AI is responsible for the decline in engagement. They do show that the group whose support matters most for adoption is currently among the least engaged.
Implications
Three implications follow. First, a cost target does not constitute a strategy. A mandate such as a 30% reduction attributed to AI has to be broken down into specific decisions within individual units, with named owners, before it can be reflected meaningfully in budgets.
Second, responsibility for how AI changes roles and management structures cannot be delegated to providers or integrators. Their competence lies in identifying and implementing use cases; the consequences for individual managers and their teams fall outside their mandate.
Third, organisations will need to reconsider how managerial seniority is defined. As long as it is measured in headcount, successful automation will be perceived as demotion by precisely those who are expected to implement it. Where programmes with strong executive backing produce little visible movement, the absence of open opposition should not be read as an absence of resistance.
The return on AI will ultimately depend on a relatively small number of managers who are asked to lead in ways they have not led before, often while uncertain about their own future. Supporting them is a precondition for the investment to pay off.
