Nearly 90% of Businesses Use AI, But Just 6% See Real Financial Payoff

Nearly nine in ten businesses now run artificial intelligence somewhere inside their operations. Fewer than one in ten call it core to how the business actually works.

That gap, not the technology itself, is what is starting to separate industrial companies that will matter from the ones that will not. Tim Dickson, chief digital and information officer at Regal Rexnord, an industrial manufacturer of motors and power transmission equipment, has laid out the split in blunt terms: a business either uses AI to do existing work faster, or it rebuilds the work itself around what AI can now do. Those are different journeys, and most companies have not noticed they are choosing one.

The 90% Adoption Number Hides a 6% Reality

McKinsey’s State of AI survey, drawn from nearly 2,000 organizations across 105 countries, found 88% of organizations use AI in at least one business function, with 79% using generative AI specifically. That number has become the industry’s favorite headline. It is also, on its own, close to meaningless.

McKinsey’s own data shows why. Only about 7% of companies have scaled AI enterprise-wide, and just 5.5% qualify as what the firm calls high performers, meaning they attribute more than 5% of profit to AI. Deloitte’s 2026 State of AI in the Enterprise report sorts the same population differently but lands in the same place: just 34% are deeply transforming the business, another 30% are redesigning key processes, and 37% are using AI at a surface level with little change to how work actually happens.

Publicis Sapient’s 2026 Global Enterprise AI Report, based on 1,550 decision makers, puts a number on the disconnect between busy and different. Seventy three percent of respondents said AI is used regularly or across most business processes, but only 10% call AI core to operations. Forty two percent said AI is already capable of meeting their business needs, but their organization is not set up to capture the value.

Inside Regal Rexnord’s Own Numbers

Dickson has published his own company’s figures, and they read like a textbook case of the easy layer. Regal Rexnord’s internal GPT handles around 2,000 associate queries a month. Its website chatbot helps more than a thousand customers a week find what they need. Thousands of employees using Microsoft Copilot report saving two to three hours a month, most of it reinvested into higher value work.

Those are genuine wins. Dickson does not pretend otherwise, writing that the results are impressive for both Regal Rexnord and its customers. But he is equally direct that this is the easy part of the AI story, the part almost any organization can now prove works.

The harder test, in his framing, is whether AI gets woven into the processes that actually run an industrial business: forecasting, demand planning, sales and operations planning, materials requirement planning, and the customer service workflows that stretch across half a dozen systems and teams. None of that looks like a chatbot use case.

Why the Easy Part Is Nearly Over

Vendors are everywhere. Pilots are cheap. Nearly any company can now stand up a proof of concept and show a manager something useful within weeks. That is precisely why pilots have stopped being a differentiator.

Running AI inside forecasting or planning systems calls for multiple agents working together instead of in isolation, AI connected directly into core systems rather than bolted on beside them, and human experts accountable for what those agents do. Dickson describes this as the moment a company crosses from using a tool to building a new capability. Deloitte’s AI Institute frames the same shift as the difference between adding AI to a process map built before AI existed and rebuilding the map itself, and warns that companies still running the old map face a compounding disadvantage as competitors redesign around AI natively.

Three Foundations Most Pilots Skip

Why do promising pilots stall before they reach the operations that actually run a factory or a supply chain? Dickson points to three unglamorous gaps that rarely make it into a vendor pitch deck.

• Data readiness – Most industrial data still sits in systems that were never built to talk to each other, with inconsistent definitions and no one formally responsible for closing the gaps. Deloitte’s pulse data backs this up: nearly half of organizations, 48%, have introduced AI without redesigning the workflows or roles it sits inside, versus just 12% that redesigned at scale with a new operating model behind it.
• Governance – Once AI touches forecasting, pricing or customer workflows, someone has to define what a model is allowed to do, on whose authority, and with what oversight. Organizations that skip this tend to discover the gap at the worst possible moment.
• Change management – Transformation is about people first. The best software will not deliver value if the employees expected to use it feel it is being done to them rather than with them.

Publicis Sapient’s survey found 22% of decision makers now identify the way their organization runs, not the technology, as the primary barrier to AI success. Fifty one percent pointed to internal data specifically.

The Governance Bill Comes Due

The 95% failure rate for enterprise AI solutions represents the clearest manifestation of the GenAI Divide.

That is how MIT’s Project NANDA initiative summarized its own headline finding in its State of AI in Business 2025 report, built from more than 300 public deployments, 52 organizational interviews and 153 executive surveys. The researchers found that companies buying AI tools from vendors and building partnerships around them succeeded roughly 67% of the time, while internal builds succeeded only about a third as often.

Gartner has since pointed to a more specific culprit than bad models. Gartner’s newsroom warned that treating AI agent governance as binary guarantees failure. “Enterprises are treating AI agent governance as binary, either locked down or fully trusted, and that is the root cause of failure,” said Shiva Varma, senior director analyst at Gartner. The firm has separately projected that more than 40% of agentic AI projects will be canceled by the end of 2027 over escalating costs, unclear business value or weak risk controls.

A separate study on enterprise AI governance found that CIOs held accountable for AI they cannot control describes almost the identical dynamic one level up the org chart from Gartner’s agents.

Not everyone accepts the 95% figure at face value. Where the experts split:

• MIT’s Project NANDA researchers – say the divide is not about model quality or regulation, but about organizational approach, and that it explains why some pilots scale while most stall.
• Marketing AI Institute – calls the methodology too narrow, arguing a six month, pure profit and loss window ignores efficiency gains, cost reductions and other value AI already delivers.
• Gartner’s Shiva Varma – locates the failure a layer deeper, in uniform governance policies applied to agents that carry very different levels of autonomy and risk.

Which Industries Are Actually Running Agents?

Manufacturing sits in the middle of the field, not at the back. Across industries, roughly 31% of enterprises now have at least one AI agent running in production, according to a mid-2026 synthesis of S&P Global Market Intelligence and McKinsey survey data. Telecommunications, retail and banking lead the pack; manufacturing trails them but sits well ahead of healthcare and government.

Healthcare and government trail not for lack of sophistication but because their regulatory perimeter is harder to defend around a non-deterministic system, and because the underlying data work costs more in those estates. That mid-2026 synthesis also credited manufacturing’s 30% share of agents in production to a wave of master data investment that only started paying off this year. Pricing is shifting alongside deployment. Some vendors now sell agents on outcomes instead of seats, a structure that comes with its own rising token bill tied to results rather than a flat license fee.

The Next Three Years Sort Winners From Bystanders

Dickson’s bet is that the companies pulling ahead over the next three years will not be the ones running the most agents. They will be the ones that did the unglamorous work first, on data, on governance, on the people expected to use the systems, and can now move faster because of it.

For now, Regal Rexnord’s own numbers, the 2,000 monthly queries, the thousand weekly chatbot customers, the couple of hours Copilot buys back each month, sit squarely in the tool column. Whether they move into the transformation column is a question the next three years, not the next product demo, will answer.

Frequently Asked Questions

How many companies have actually scaled AI agents into production?

McKinsey’s State of AI survey found 62% of organizations are at least experimenting with agentic AI, but only 23% have scaled at least one agent system into production, a much smaller slice than headline adoption numbers suggest.

Why do most AI pilots fail to show a financial return?

MIT’s research found that successful deployments typically spend 60% to 80% of project resources on data preparation alone. Companies that underestimate that work tend to see delays or outright failure rather than a model problem.

Is it better to buy AI tools or build them in-house?

MIT’s data shows purchasing from specialized vendors and building partnerships around their tools succeeds about 67% of the time, while fully internal builds succeed only about a third as often, largely because vendor tools adapt to workflows faster.

Are companies making more money from AI, or just saving time?

Mostly saving time so far. Deloitte found two thirds of organizations, 66%, report productivity and efficiency gains from AI today, while 74% hope for revenue growth from it eventually and only 20% say they are already seeing that revenue.

How much are companies expected to spend on AI in 2026?

Gartner has forecast worldwide AI spending will reach roughly $2.59 trillion in 2026, a 47% increase over 2025, even though most organizations still have not converted that spending into measurable returns.