The AI Death Trap: Why Organisations Need to Think Bigger Than Automation – The European Business Review
Companies worldwide are pouring money into AI. Gartner’s September 2026 forecast puts worldwide AI spending at approximately $2.7 trillion in 2026 and $3.6 trillion in 2027. These figures cover AI broadly, including infrastructure, software and services, rather than generative and agentic AI alone. But for many organisations, returns on this investment are proving elusive.1
Deloitte’s 2025 survey of 1,854 executives across Europe and the Middle East found that only 6% reported payback on a typical AI use case in under a year; most reported satisfactory ROI within two to four years. Even for their most successful projects, only 13% reported returns within 12 months. McKinsey’s 2025 workplace report likewise found limited enterprise-wide returns among the executives it surveyed.2,3
Yet investment intentions remained strong: in Deloitte’s survey, 91% of organisations planned to increase their AI investment over the following year, to August 2026. The question is how that investment can translate into lasting value.2
Companies often limit their expectations of AI to performing existing tasks faster, more efficiently and more cheaply. Those savings can be valuable, but the underlying business may remain essentially unchanged: the same products, serving the same needs. If the additional value is bounded while AI costs continue to recur, doing more of the same can become an expensive strategy.
This parallels the electrification of factories in the 19th century.
When electric motors replaced steam power, simply retaining the existing factory layout and workflow left much of the new technology’s potential unrealised. Larger productivity gains depended on redesigning production around what electricity made possible.
New technology does not automatically create a new organisation. The gains depend on how businesses redesign their work and their value proposition around its capabilities. When they do not, they risk what I call the AI Death Trap.
The trap combines three things: an AI cost curve, a traditional value model and digital scale. The cost curve recurs. AI requires heavy capital investment, and models, chips and the systems built around them have to be upgraded and replaced, so the outlay is never paid off once and for all. The value model is traditional: most organisations use AI to do more of what they already do, and each additional unit of the same output is worth less to the business than the last. The scale is digital, but unlike earlier digital technologies, every use still costs money to compute. Together, these produce costs that recur and grow with use, set against value that flattens. Scale does not rescue the investment, because past the point where an additional use is worth less than the compute it consumes, growth adds cost faster than value.
Many organisations walk into the trap because they apply the economics of retrieval to a technology that generates. The last digital era ran on retrieval. Netflix makes a film once, stores it and streams copies to millions of viewers. A payment network adds another cardholder at almost no extra cost. The expense lay in creating the content or building the network. Copying and transmitting it cost close to nothing, so growth pushed the cost of serving each additional user towards zero. Generative AI keeps nothing ready in storage. Each answer, summary or agent action is computed at the moment someone asks for it, and each computation consumes processors and electricity. Nvidia’s chief executive, Jensen Huang, described the shift in March: “We went from a retrieval-based computing system to a generative-based computing system… We need a lot of storage in the old world. We need a lot of computation in this new world.” 5
Computation is different, and this is the biggest mistake that organisations make, I believe. Digital infrastructure captures value through scale, because the marginal cost of scale was almost zero. Adding one more user on your payment rail may not cost much. A generative system is different. And that difference breaks the familiar digital playbook.
More users still spread a fixed cost more thinly, but in a generative business they also mean more computation. A firm that uses AI only to do its existing work more cheaply gets savings bounded by the work it removes, while its costs recur and rise with use.
Consider an illustrative customer-service investment. Suppose a company commits $10 million a year to maintaining its AI capability and saves $2 million a year in employment costs, with no additional revenue or other monetised benefit. On those assumptions, the annual shortfall is $8 million before usage-dependent compute costs. If the $10 million were instead a one-off investment, subtracting one year’s savings would not establish an $8 million annual loss. The strategic problem emerges when recurring commitments remain high, further savings from automating the same work diminish, and each additional use still carries a compute cost. Costs and benefits must be compared over the same period and workload.
Efficiency gains can justify an AI investment. The danger is assuming that they will keep growing fast enough to support recurring costs. Many deployments focus on the intensive margin: doing existing activities more efficiently. AI also offers an opportunity to expand the extensive margin: doing things the organisation could not previously do, creating different products and services, and developing new sources of value. Expanding that frontier is a strategic response to the AI Death Trap.
AI doesn’t just speed up the existing game. It changes the game entirely. If you’re fixated on using AI to make each employee a bit faster, you’re missing out on the opportunity to have that employee create new, more valuable things – and to see that increased capability compound across your organisation. The investment question is whether those new capabilities create value that can sustain the costs of deploying AI. Neither rising investment nor greater usage guarantees revenue growth.
Harnessing the promise of AI therefore means rethinking the architecture of work. It means asking how existing workflows could change, what becomes possible when machines rather than humans are doing the processing, and what that might mean for the business model itself.
It’s like opting to pave a cow path. If a cow creates a meandering path through a bumpy field, paving it might make it smoother but it won’t get you through that field any more directly. If you want to build a highway, first you’re going to have to change the route.
The workflow and business model need to change alongside the technology.
Organisations have spent decades redesigning workflows around computers, software and the internet, but much of that architecture reflects human limits on attention and information processing. AI agents can process and coordinate work in different ways. Their outputs still require appropriate checks, and human beings remain responsible for how those outputs are used.
Take recruitment. AI agents might extract applicant data, compare it against defined criteria, and coordinate parts of the process at scale. Hiring teams would still need to check accuracy, fairness and the quality of selection decisions. The strategic question then extends beyond making screening more efficient: what does this free people to do? Where could the unused human effort, ingenuity and creativity be directed? And what new services or forms of value might emerge?
It’s like asking technology to optimise an answer before you have even reconsidered the question.
Instead, leaders need to work backwards.
If you’re going to invest in AI, a far better place to start is with the why. Why does your organisation exist and what problem is it trying to solve? From there, start questioning what you should be producing. Ask yourself this: what products, services or value propositions become possible if your organisation is no longer constrained by the assumptions and limitations of the past? Only then should you start thinking about how technology and AI can be used to deliver these things.
For leaders, this means revisiting the purpose of the business before deciding which activities to automate. What customer problem could the organisation now solve differently? What could people create or accomplish that was previously beyond their reach?
The lesson is to question the problem before optimising the solution. Better execution of an inherited process may leave the more valuable opportunity unexplored.
Questioning the problem also means rethinking the ways that value is created.
Traditional businesses tend to think in terms of a value chain: a company produces something, sells it, and the customer consumes it – value moves through in a linear, sequential manner. AI offers a new way to think about this.
Rather than value being fully designed and created before a product reaches the customer, AI can allow models, data and context to come together at runtime, during the actual moment of use. The interaction itself can generate new information and new possibilities, which can in turn feed back into the next interaction. This creates a new dynamic – a value loop involving the company, AI and the human being.
The strategic aim is to increase marginal utility: to design products and services that help users develop new capabilities and find greater value through continued use. That requires learning from interactions and adapting the offering. More usage, more data or more engagement does not by itself establish that users are gaining more value.
It’s not about maximising engagement or just finding ways to keep people hooked to what you produce. The goal with marginal utility is to use information and data you’re generating at runtime to create and improve products and services that help people – that make them grow, develop new capabilities or aspirations – so they have a reason to come back. Because the experience has become more valuable to them. This is the whole point of the value loop. It’s about creating the conditions for value to keep emerging.
The strategic question that AI poses for organisations extends beyond doing things faster and better, or making something that did not exist before. What can organisations create with AI that people value, and that can become more valuable as they use it?
The companies that end up winning with AI won’t be those that automate human beings most aggressively. They’ll be the companies that use AI to expand what humans are capable of doing, creating and wanting. And in doing so, they’ll generate new sources of demand.
This thinking underpins AI-First Strategy, an open-enrolment programme at the London School of Economics and Political Science (LSE). The programme is taught by myself together with MJ Jabbour, Microsoft’s AI Innovation Officer in the Office of the CTO, and Mickey McManus, Senior Advisor at BCG. Our new programme challenges senior leaders to diagnose the gap between AI investment and value capture, and design AI-enabled value loops for their own business contexts.4
We designed this course to challenge executives from every sector to rethink AI – from task automation to an opportunity to create wholly new economic value – and to consider what needs to change in their organisation; and how to turn these insights into a practical and actionable road map.
Ultimately, leaders need to judge AI by the value it creates and the capabilities it expands, alongside its full recurring and usage-dependent costs. Automation can be part of that strategy. Escaping the AI Death Trap requires a value model that can develop with the technology, so that growing AI capability has somewhere valuable to go.
Mark your calendar for these stimulating events and prepare to be inspired.
Save the dates for the Open Days of your preferred Executive Education programs. Don’t miss out!
Mark your calendars for upcoming MBA events and programmes. Don’t miss out on these valuable opportunities!
None of the information on this website is investment or financial advice. The European Business Review is not responsible for any financial losses sustained by acting on information provided on this website by its authors or clients. No reviews should be taken at face value, always conduct your research before making financial commitments.
source
This is a newsfeed from leading technology publications. No additional editorial review has been performed before posting.
