Companies get real AI returns yet most still can't scale past the pilot

A new global study finds nearly three in four companies see real financial gains from AI, but barely one in three can push a project past the pilot stage. The bottleneck isn't the technology. It's legal risk and legacy IT.

Here's the number that should worry anyone pricing AI stocks off exponential enterprise adoption: just 13% of companies say they're fully on track with their AI initiatives. Not because the technology disappoints. Three-quarters of organizations that have actually implemented AI report a measurable financial return, according to a study from consultancy BearingPoint reported by Reuters on October 1. The problem sits one step earlier, in the gap between a working pilot and a system every team actually uses.

BearingPoint surveyed 1,050 C-suite executives and senior leaders across 13 countries for the report, titled "Scaling AI for Measurable Impact." Respondents split cleanly into two camps. Among the most mature organizations, which BearingPoint calls Leaders, 47% scale their AI projects fully as planned. Among everyone else, the Implementers, that figure drops to 6%. Reuters reported that around 24% of companies saw AI-driven cost savings of at least 10%, while only 4% saw revenue growth of that size, and that roughly 40% of firms that had implemented AI reported gains on both fronts at once.

Ask executives what's actually stopping them and the answer has nothing to do with hallucinations or model quality. Forty percent of respondents named legal and regulatory uncertainty as their top barrier to scaling, and 34% pointed to the headache of wiring AI into existing IT systems built for a pre-AI world, according to the BearingPoint findings. A pilot can run inside a sandboxed workflow with a handful of users and nobody's compliance team notices. Rolling that same tool out company-wide means someone in legal has to sign off on data handling, someone in IT has to connect it to systems that were never designed for it, and someone in HR has to figure out what happens to the jobs it touches. None of that is a model problem. All of it is slow.

Geography makes the pattern sharper. China and the US lead on comprehensive AI rollout, at 20% and 18% of companies respectively, while Germany trails at just 8%, Reuters noted citing the BearingPoint data. That's not a story about who has better engineers. It's a story about who has fewer internal gatekeepers between a working demo and a production system.

This echoes a harder number MIT's Project NANDA put out earlier this year: across 300 public AI deployments and more than 150 executive interviews, the research group found 95% of generative AI pilots delivered no measurable profit-and-loss impact. MIT's researchers traced much of that failure to where the money actually goes. More than half of generative AI budgets land in sales and marketing tools, the easiest use case to get approved internally, while the bigger returns sit in less glamorous places like procurement and back-office operations. Two separate studies, a year apart in focus, landing on the same structural story: the technology works, the organization around it doesn't move fast enough to catch the value.

The capex math doesn't wait for the adoption curve

None of this is slowing down what's being built. The five largest hyperscalers are on pace to spend somewhere between $775 billion and $800 billion on AI infrastructure in 2026 alone, and total global AI capital spending is projected to climb toward $1.64 trillion annually by 2031. That spending assumes enterprise demand keeps compounding at the same rate as the chips get built. If only 13% of buyers are actually on track to use what they're paying for at full scale, the gap between the infrastructure bet and the revenue it's supposed to generate gets wider every quarter, not narrower.

That's exactly the kind of data point Michael Burry and the Bank of England's Andrew Bailey have been pointing to when they warn that AI valuations are running ahead of the demand that's supposed to justify them. This study doesn't say the AI boom is fake. It says the boom is real and the bottleneck is organizational, which is actually a worse problem for anyone betting on a near-term payoff, because you can't fix a legal department's risk tolerance with a better GPU.

There's an obvious winner buried in that bottleneck, and it's not a chipmaker. BearingPoint is a consultancy that sells AI implementation and scaling services, which means a study concluding that most companies can't get past the pilot stage without outside help is also, conveniently, an advertisement for BearingPoint's own business. That doesn't make the underlying numbers wrong. Reuters is reporting data from a real survey of over a thousand executives, and the pilot-to-scale gap shows up independently in MIT's research too. But it's worth naming plainly: every consultancy from Accenture to McKinsey is racing to position itself as the fix for exactly this problem, and the firms diagnosing the adoption gap are the same firms charging by the hour to close it.

Also read: California bans AI-only firing decisions under the No Robo Bosses Act • Bank of England Governor Warns AI Valuations Could Face a Sharper Correction • Gemini 4 Argon Posts the Lowest Hallucination Rate of Any Top AI Model

This article is posted in AI News, check it out for more related stories.