Why Companies Spend More on AI Than It Returns

Key takeaways

• 56% of CEOs say AI has delivered no revenue gain and no cost saving (PwC, January 2026). Only 12% get both.
• The gap is rarely the model. It is AI bolted onto a process no one rebuilt, with no owner and no baseline.
• The profitable minority embedded AI into a redesigned workflow before they scaled the spend. Readiness came first.

Fifty-six percent of CEOs told PwC in January that AI has produced neither higher revenue nor lower cost. The spend went up regardless. That is the quiet story of enterprise AI in 2026: budgets keep climbing while the return stays theoretical.

For most of them the model was never the problem. They bought one and changed nothing around it.

The reckoning has arrived#

The hype phase is over. The market has stopped asking whether AI is impressive and started asking what it returned.

The evidence is consistent across the people who count it. Forrester found that only 15% of AI decision-makers reported an EBITDA effect in the past year, and fewer than a third could tie AI value to the P&L at all. BCG's AI Radar put 60% of companies in the "laggard" group with minimal revenue or cost gains, and only 5% in the group pulling substantial value. MIT's widely quoted figure, that 95% of AI pilots stall with little measurable impact, is from 2025, but it is still the number everyone cites because nothing has replaced it.

Boards have noticed. In July, Chamath Palihapitiya warned that "CEOs and the CFOs, in my opinion, probably have no idea how much tokenmaxxing is going on inside of their organisations." The bill is arriving before the benefit.

This is no longer a hype problem. It is a return problem.

Where AI actually breaks#

If the model were the issue, the fix would be simple: buy a better one. It rarely is. As Ayelet Israeli of Harvard Business School put it, "most AI initiatives fail not because the models are weak, but because organisations aren't built to sustain them."

The pattern underneath almost every stalled project is the same. AI gets added on top of a process that was never redesigned to use it. Deloitte's Bill Briggs described companies "trying to fit AI into preexisting workflows as if it is just another bolt-on, rather than reimagine their processes holistically", and pointed to a revealing split in where the money goes: roughly 93% of AI spend on the technology, and 7% on the people and process changes that make it work. Bain was blunter still: "the single most costly mistake in AI deployment is automating a broken process." AI does not fix workflow debt. It locks it in, speeds it up, and makes it more expensive to unwind.

In practice the money tends to leak in four places, and none of them is the model.

The workflow was never redesigned. The tool lands on top of how the work was already done, so it just accelerates the part that was already the bottleneck. As McKinsey's Alexander Sukharevsky puts it, the job is to "look at the processes end-to-end and reinvent them instead of just infusing technology into an anachronistic process."

No one owns the outcome. A pilot gets a sponsor for the rollout but nobody accountable for the result. Bain names this directly: "ambiguity about ownership ceases to be a nuisance and becomes a performance failure." With no owner and no number to hit, the project drifts and quietly becomes someone's side project.

The data is not reachable. The model needs clean, accessible data and cannot get to it. Bain found this to be the number one reason AI programmes underperform: companies "cannot reliably get access to their own data." Maribel Lopez of Lopez Research says it plainly: "we've never fixed this data quality problem in most organisations, and it comes back to haunt a company in spades as they move to AI."

Governance lags the tools. Teams either lock everything down or trust it blindly, and neither is safe to scale. Gartner expects 40% of companies to switch off AI agents by 2027, largely because governance was treated as an afterthought. IDC's Eric Newmark warns that "handing the data problem back to the customer delays deployment; at worst, it kills the project entirely."

These are wiring problems, and they compound when the IT, AI and revenue sides of the business run as separate layers with no one accountable for the seams between them.

What readiness actually means#

Readiness is not a maturity score or a slide in a deck. It is a short list of things that need to be true before AI can pay off in a given workflow.

• A redesigned workflow. The process was rebuilt around what AI does well, not just decorated with it.
• A named owner. One person is accountable for the outcome, with the authority to change how the work is done.
• A baseline. There is a number, measured before the tool went in, that the project is meant to move.
• Reachable data and clear governance. The model can get to the data it needs, and someone has decided what it is allowed to touch and who checks the output.

None of that is exciting, which is why most companies skip it and reach for a tool instead. But this list is also what turns AI from an experiment into an investment. When all four are true, the same model that stalled elsewhere starts to return something.

What the profitable few did first#

The 12% in the PwC survey who get both revenue and cost gains share one trait: they embedded AI instead of bolting it on. BCG's 5% look the same. Their edge was not a bigger budget or an earlier start. It was that they rebuilt the process first and then put AI inside it, with an owner and a baseline attached.

Order is the whole game. The companies still stuck did it the other way around: they bought the tool, then hoped the organisation would reshape itself to fit. It rarely does. Bain found the most telling number of all: 90% of companies that miss their AI savings target raise the next budget anyway. Spending more against a broken wiring diagram only makes the leak more expensive. Redesign the workflow first, give the result an owner and a baseline, and only then hand AI the part it is good at.

A note for European mid-market companies#

There is a version of this story that runs in Europe's favour. Only 20% of EU enterprises with ten or more staff used AI in 2025, up from 13.5% a year earlier, according to Eurostat. Adoption is still early and still narrow: the most common uses are text analysis (11.8%), image or audio work (9.5%) and text generation (8.8%). For a mid-sized company, that gap is not a reason to panic and copy US-scale spending. It is a reason to move deliberately.

The mid-market disadvantage is rarely ambition. It is integration capacity and governance. Vendor research suggests many mid-market firms already run several AI tools each while only about a third have a governance framework they apply consistently, which spreads effort thin and leaves little owned end to end. The companies that pull ahead are the ones that pick one or two workflows, rebuild them properly, and can prove the result, rather than the ones with the longest tool list. Disciplined adoption, not volume, is the European edge.

Readiness before spend#

If your AI spend is climbing faster than anything it has returned, the most useful thing you can do is diagnose the spend you already have before adding to it. Find where it is leaking first, then decide what to buy.

That is what an AI Readiness Audit is for: a senior-led diagnostic that tells you where to invest, what to leave alone, and what it will actually cost. No pitch deck, no tooling recommendation you did not ask for. Just an honest read of where AI would pay off in your business and where it would only add cost.

The 95% did not lose to a better-funded competitor with a smarter model. They lost to their own wiring. That gap is fixable, and it is the first thing worth looking at.

Sources#

• 56% of CEOs see no revenue or cost gain, only 12% get both (embedded vs bolt-on) - PwC 29th Global CEO Survey, January 2026, via Forbes.
• Only 15% report an EBITDA effect; fewer than a third can tie AI value to the P&L - Forrester, "Predictions 2026: AI Moves From Hype To Hard Hat Work", October 2025.
• 60% "laggards" with minimal gains, only 5% "future-built" - BCG AI Radar, "The Widening AI Value Gap", September 2025, via PR Newswire.
• 95% of AI pilots stall with little measurable P&L effect (2025 figure) - MIT Project NANDA, "State of AI in Business 2025", via Fortune.
• Tokenmaxxing quote - Chamath Palihapitiya, CNBC, 14 July 2026.
• "Most AI initiatives fail... organisations aren't built to sustain them" - Ayelet Israeli, Harvard Business Review, November 2025.
• "Just another bolt-on" + 93% tech / 7% people spend - Bill Briggs, CTO Deloitte Consulting, via Fortune, December 2025.
• "The single most costly mistake... automating a broken process" + "cannot reliably get access to their own data" + 90% raise budget anyway - Bain & Company, "Your AI Budget Is Growing. Your Returns Aren't.", 2026.
• "Ambiguity about ownership... becomes a performance failure" - Bain & Company, "An Operating Model for the Age of AI", May 2026.
• "Reinvent them instead of infusing technology into an anachronistic process" - Alexander Sukharevsky, Managing Partner QuantumBlack (McKinsey), The Innovator, June 2025.
• "We've never fixed this data quality problem..." - Maribel Lopez, Lopez Research, via CIO.com, July 2026.
• "Handing the data problem back to the customer... kills the project entirely" - Eric Newmark, Group VP & GM, IDC.
• 40% of companies will switch off AI agents by 2027 (forecast) - Gartner (Shiva Varma), May 2026.
• EU AI adoption 20% in 2025 (up from 13.5%); usage 11.8% / 9.5% / 8.8% - Eurostat, "Use of artificial intelligence in enterprises", 2025.
• Mid-market runs several AI tools, only ~1/3 with consistent governance (directional, vendor survey) - Freshworks, "The Global Cost of Complexity Report", 2026.