Organisations are investing billions in Generative AI, large language models, and automation. Most will not see a material return on that investment — not because the technology is inadequate, but because technology deployment that isn't tied to a commercial outcome is not transformation. It is expensive capability acquisition.
01The adoption paradox
The headline adoption numbers look impressive. McKinsey's November 2025 Global AI Survey found 88% of organisations now using AI in at least one business function. The value numbers tell a different story: only 39% report any EBIT impact at all, and more than 80% report no meaningful enterprise-wide EBIT impact despite that adoption.
MIT's NANDA initiative went further in its widely cited 2025 study of more than 300 public AI deployments: roughly 95% of organisations deploying generative AI showed zero measurable P&L impact. BCG's September 2025 follow-up, The Widening AI Value Gap, found the picture deteriorating rather than improving — 60% of organisations generating no material value despite continued investment, against just 5% creating value at scale. Gartner's most recent enterprise survey puts it in similarly stark terms: roughly 20% of AI initiatives deliver any return within twelve months, and only 2% deliver what it calls long-term disruptive value.
02Three questions before you deploy a model
The organisations that will realise value from AI investment are those that answer three questions before a single model goes into production.
- What specific commercial outcome does this enable? Not "efficiency" in the abstract — a number, owned by someone, with a baseline.
- How does it change the customer-led outcome? Faster response, better resolution, more relevant service — something the customer would actually notice.
- What organisational capability is required to sustain it? The skills, the data discipline, the process change that has to exist after the pilot team moves on.
Without clear answers to all three, AI becomes another line on the IT budget rather than a driver of competitive advantage. This is not a contrarian view of the technology. It is the consistent pattern across every credible study of where AI investment actually lands.
02.1Where the money actually lands
One detail in the research deserves more attention than it gets: the highest returns are rarely in the most visible deployments. Customer-facing chatbots and marketing copilots absorb the largest share of pilot budgets and produce some of the weakest measured returns. Back-office process automation — claims handling, reconciliation, document processing — consistently produces the strongest returns, precisely because the outcome was defined narrowly and the baseline was already known.
That is an uncomfortable finding for anyone building an AI roadmap around visibility rather than value. The least impressive-sounding use case in the portfolio is often the one paying for the rest.
AI becomes another line on the IT budget rather than a driver of competitive advantage when it isn't tied to a defined commercial outcome before deployment.Dan Collins
03The leadership failure behind the technology failure
None of this is really a story about AI. It is the same execution gap that has defined enterprise transformation for thirty years, now wearing a new technology label. Gartner's own research into why AI projects fail points to expecting too much, too fast, as the single most common cause cited by leaders — a strategy and sequencing failure, not a model-performance failure. The pattern is familiar to anyone who has run a transformation programme of any kind: a pilot launched under hype, no governance framework, no defined success criteria, and no named owner once the novelty wears off.
04What to do instead
Before approving the next AI initiative, anchor it to one of the three questions above and write the answer down. Give it a named, accountable owner — not a steering committee. Set a kill criterion in advance: the threshold at which the initiative is stopped, not quietly extended. And measure the back-office, unglamorous use cases with the same rigour as the customer-facing ones, because the research is consistent that this is where the actual return tends to live.
05The governance gap nobody budgets for
A second pattern shows up consistently once you look past the headline failure rate: AI initiatives that pass a pilot and stall at scale-up almost always fail for the same reason — the governance built for a five-person pilot team does not survive contact with compliance, security, works councils, and department heads, all of whom acquire a legitimate stake once the system touches production data or customer interactions. Nobody is formally responsible for resolving that contention, so the rollout drags, and a project that looked successful in pilot quietly stalls before it ever reaches the scale where the original business case assumed its returns would appear.
This is, again, not a model problem. It is the same organisational-design failure that derails conventional transformation programmes, simply arriving later in the AI initiative's life because the pilot phase masks it. The fix is the same one that works for transformation generally: name the cross-functional owner before scale-up begins, not after the first escalation.
06The maturity question every board should ask
There is a useful, blunt diagnostic that separates organisations that are managing AI investment from those that are accumulating it: can the CFO name, for every AI initiative currently funded, the specific commercial metric it is meant to move, and the current value of that metric against baseline? In most organisations I work with, the honest answer covers a handful of flagship initiatives and goes quiet for everything else. That gap — between the showcased initiatives and the long tail of unmeasured pilots — is usually where the bulk of the unreturned investment is sitting.
Closing that gap does not require more AI governance committees. It requires the same discipline any capital allocation decision should already carry: a baseline, a target, an owner, and a date by which the initiative either shows movement or is stopped. Most organisations apply that discipline rigorously to a manufacturing capex decision and abandon it almost entirely the moment the word "AI" appears on the business case.
07What this means for 2026 budgets
The CEOs who get this right in 2026 will not be the ones who deployed the most models, and they will very likely not be the ones with the largest AI budget line. They will be the ones who could explain, for every model in production, exactly what commercial outcome it was built to change — and who were willing to stop funding the initiatives that, eighteen months in, still couldn't answer that question.