The Enterprise AI ROI Gap: 90% Adopt AI, Only 18% See Revenue

Nine in ten enterprises say generative and agentic AI are transforming how they work. Fewer than two in ten say that transformation is showing up as revenue. That is the finding of a global HCLTech study released on July 21, 2026, and it is not an isolated data point. Put alongside separate 2025 and 2026 surveys from MIT, Gartner, Deloitte and McKinsey, a consistent pattern emerges: enterprise AI adoption has become close to universal, while enterprise AI payoff remains rare, concentrated, and hard to measure. This piece lays out the evidence for that gap as neutrally as possible, then works through why it exists and what separates the organizations that are closing it.

The number behind the headline

The HCLTech figures come from "The Blueprint for AI Leadership," a global study HCLTech ran with research firm Raconteur among 500 enterprise decision-makers, published July 21, 2026. Ninety percent of respondents said generative and agentic AI are transforming their workflows, 91% cited improved data access, and 90% reported productivity gains. Against that backdrop, only 18% said AI is delivering significant revenue impact, a gap the report's authors describe as a widening divide between operational progress and business outcomes. Pawan Vadapalli, HCLTech's Corporate Vice President and Global Head of Digital Business Services, put it directly: "The organizations pulling ahead are not just running more pilots; they are rethinking how the business works, embedding AI into everyday decisions and workflows."

This is not one outlier study

The HCLTech numbers land inside a pattern other independent surveys have been documenting for over a year. MIT's Project NANDA published "The GenAI Divide: State of AI in Business 2025" in July 2025, based on more than 300 public deployment reviews, 52 executive interviews and 153 leader surveys. It found that despite an estimated $30 to $40 billion in enterprise generative AI investment, 95% of pilots produced no measurable profit-and-loss impact, while only about 5% of integrated deployments extracted significant, ongoing value. Gartner's June 2025 forecast is a forward-looking version of the same story: it predicts more than 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value, or inadequate risk controls, and it notes that most agentic AI initiatives today are early-stage experiments driven more by hype than by proven return.

McKinsey's global AI survey, fielded in the summer of 2025 among nearly 2,000 respondents across 105 countries, found that more than 80% of organizations report no tangible enterprise-level EBIT impact from generative AI, and only 17% attribute 5% or more of their organization's EBIT to it. Just 1% of executives in developed markets describe their gen AI rollout as "mature." Deloitte's 2026 State of AI in the Enterprise report, surveying more than 3,000 business and IT leaders, found a similarly shaped gap on the revenue side specifically: 74% of organizations hope to grow revenue through AI, but only 20% are actually doing so today, and while 54% expect to move 40% or more of their AI experiments into production within three to six months, only 25% have reached that milestone so far.

Why the gap exists

None of this means the underlying models are the problem. The more consistent explanation across these reports is that the gap sits in the distance between a pilot and a production system, and in everything that distance requires.

MIT's researchers frame it as a "learning gap." Generic chat tools work well for individuals precisely because they are flexible, but that same flexibility means they do not retain feedback, adapt to a specific team's context, or improve with use, so they stall when asked to run inside an actual enterprise workflow rather than beside it. A separate HCLTech report from earlier in 2026, "The AI Impact Imperatives," surveyed 467 senior executives at companies with more than $1 billion in revenue and found 43% expect their major AI initiatives to fail, with much of the shortfall tied to deploying AI without adequate change management, even as nearly half of leaders expect measurable value within 18 months.

Deloitte's data reinforces the same point from the production side: a pilot can succeed with a small team, clean data, and an isolated environment, but production demands integration with legacy systems, security review, compliance sign-off, and ongoing maintenance, which is exactly the work that a 54%-expect versus 25%-achieved gap describes. And McKinsey's research is explicit that the shortfall is disproportionately an execution problem rather than a technology one: the organizations it identifies as high performers are distinguished by redesigning workflows around AI, rather than dropping AI into workflows unchanged, and by putting senior leaders directly in charge of AI governance rather than delegating it.

The tool fragmentation tax

A less obvious driver shows up once organizations move past a single pilot and start running AI across many teams at once: sprawl. Salesforce's 2026 Connectivity Benchmark Report, run with Vanson Bourne and Deloitte Digital among 1,050 IT leaders, found the average enterprise now runs about 12 AI agents, a number it expects to reach 20 by 2027, and that roughly half of those agents operate in complete isolation from one another, with no shared data or handoffs. Eighty-six percent of the IT leaders surveyed said AI agents would add more complexity than value without stronger integration. Separately, Futurum Group's 1H 2026 Enterprise Software Decision Maker Survey of 830 decision-makers found 66% of enterprises now favor a platform-first strategy over a best-of-breed one, citing exactly this kind of friction, and cited WalkMe research estimating that unmanaged technology friction costs employees the equivalent of 51 workdays a year.

That points to a structural reason the revenue-impact number stays low even as adoption climbs: value that AI creates in one tool can leak straight back out in the cost of coordinating across a dozen disconnected ones. It is one reason some organizations are consolidating around a single, model-agnostic workspace rather than adding another point tool for every new model release, so that switching or combining models is a routing decision made in one place instead of a fresh integration project each time. Metir AI is built around that same logic, giving teams access to multiple leading models inside one workspace rather than one more silo to reconcile.

What separates the enterprises that see impact

HCLTech's report splits its 500 respondents into "AI Leaders," who convert adoption into growth, innovation and customer-experience advantage, and "AI Followers," who mostly capture efficiency gains without much revenue effect. Leaders are four times more likely to scale agentic and autonomous AI than Followers, and the gap between the two groups is concentrated in a small number of concrete practices rather than in which vendor or model either group chose.

Seventy-three percent of Leaders define measurable use cases up front, against 22% of Followers. Sixty-three percent of Leaders have secured senior leadership sponsorship, against 36% of Followers. Ninety-three percent of Leaders run structured upskilling programs, against 20% of Followers, the single widest gap in the report. And Leaders are roughly eight times more likely to express confidence in their underlying data foundations, at 74% versus 9%. None of those four practices is exotic or expensive relative to the AI spend itself; what they share is that they are organizational commitments rather than technology purchases; a defined use case, an executive sponsor, a training budget, and clean data are decisions a company makes about itself, not features a vendor ships.

The takeaway

The honest reading of five independent 2025 and 2026 surveys, run by a Big Four consultancy, a systems integrator, an industry research firm, an analyst house and a university lab, is that they agree with each other more than any one of them would if the finding were an artifact of methodology. Enterprise AI adoption is genuinely close to universal. Enterprise AI revenue impact is genuinely rare, and the two facts are not contradictory: they describe an industry that has finished the easy part, buying and switching on the tools, and is now in the much harder and slower part, redesigning how work actually gets done around them. The organizations narrowing the gap are not the ones with access to a better model; every survey here points to the same handful of unglamorous decisions instead: a measurable use case, an accountable executive, a trained workforce, trustworthy data, and fewer disconnected tools to reconcile along the way.

Sources:

• HCLTech report exposes widening AI divide with only 18% of enterprises seeing revenue impact despite near-universal adoption | Webnewswire
• HCLTech Report Exposes Widening AI Divide With Only 18% of Enterprises Seeing Revenue Impact | IndianWeb2.com
• HCLTech report warns 43% of enterprise AI initiatives may fail as leaders face shrinking timelines for impact | PR Newswire
• MIT Says 95% Of Enterprise AI Fails, Here's What The 5% Are Doing Right | Forbes
• State of AI in Business 2025: The GenAI Divide | MIT NANDA (PDF)
• Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 | Gartner Newsroom
• The state of AI: How organizations are rewiring to capture value | McKinsey
• From Ambition to Activation: Organizations Stand at the Untapped Edge of AI's Potential | Deloitte US Press Release
• The hidden ROI of AI: What leaders should actually measure | Fortune
• IT leaders grapple with AI agent sprawl | CIO Dive
• Will Technology Friction Derail the ROI Promise of Enterprise AI Investments? | Futurum Group

Image credits

Header image: an empty, modern corporate conference room, generic and not tied to any company named in this piece, by Excluzobusinesscentre via Wikimedia Commons, licensed under CC BY-SA 4.0. In-body photograph: a team meeting in an office, by Klean Denmark via Wikimedia Commons, licensed under CC BY-SA 2.0. Neither image depicts HCLTech, MIT, Gartner, McKinsey, Deloitte, Salesforce, or any respondent in the surveys cited; both are used to illustrate generic enterprise office settings.