Why 71% of Enterprises Still Aren't Seeing ROI from AI

In January 2025, Forbes ran a piece pointing out that roughly three-quarters of enterprises weren't yet seeing ROI from AI. It landed hard because the gap between spend and outcomes had become impossible to ignore.

Sixteen months later, in May 2026, the obvious question is: did the picture get better?

The honest answer is - barely. The capital has scaled. The pilots have multiplied. The dashboards are prettier. But the gap between AI investment and AI return remains the defining problem of enterprise technology in 2026.

Here's what the latest data actually says - and, more importantly, what separates the companies that are winning from the 70%+ that still aren't.

The ROI picture in mid-2026 reads almost identically to early 2025, just with bigger denominators and sharper evidence:

In other words: AI is delivering productivity at the individual level almost everywhere. It is delivering transformation at the enterprise level almost nowhere.

The "75% aren't seeing ROI" framing of January 2025 has become, in May 2026, a more nuanced - but equally uncomfortable - "the bottom 70% are still stuck, and the top 5% are pulling away faster."

This is the widening AI gap, and it is the real story of 2026.

Three things shifted meaningfully in the sixteen months since the original Forbes piece.

1. CEOs Stopped Delegating It

In BCG's 2026 AI Radar, nearly three-quarters of CEOs now say they are their organization's main decision-maker on AI - roughly double the share from the prior year. Half of CEOs surveyed say their own job is on the line if AI doesn't pay off. [6]

That's a material shift. In 2024 and early 2025, AI was largely a CIO/CTO portfolio item. In 2026, it has migrated to the CEO chair. The accountability has moved with it.

2. Agents Stopped Being Theoretical

A year ago, "agentic AI" was a conference word. In 2026, it is a budget line. BCG reports that agentic AI is already driving 17% of total AI value, expected to nearly double by 2028. [2] Trailblazer companies are now directing more than half of their 2026 AI corporate investment toward agents and are roughly 2x as likely to deploy them end-to-end across a workstream rather than as point tools. [6]

3. The Failure Patterns Are Now Well-Documented

Sixteen months ago, the failure causes were anecdotal. In 2026, the data is consistent across MIT, BCG, Deloitte, NVIDIA and WRITER. The reasons enterprises don't see ROI are no longer a mystery - they're a pattern.

Pulling across the major 2025-2026 studies, the same root causes appear again and again.

1. The Money Is in the Wrong Place

MIT's research found that over half of corporate GenAI budgets are directed at sales and marketing, despite the strongest returns showing up consistently in back-office automation - business process automation, reduced outsourcing, and operational efficiency. [3]

Companies are funding the demo-friendly use cases, not the ROI-friendly ones.

2. Pilots Don't Integrate

MIT's lead author, Aditya Challapally, named it explicitly: the issue is rarely model quality. It is a "learning gap" - generic tools that don't adapt to enterprise workflows, don't retain organizational context, and can't be governed at scale. [3] Pilots run beautifully in isolation and fall apart at the integration seam.

3. Build-Your-Own Underperforms Partner-Sourced

The same MIT study found that AI tools sourced through specialized vendors and partnerships succeed roughly 67% of the time, while internal builds succeed only about one-third as often. [3] The instinct to "build it ourselves" is, statistically, the more expensive path to no ROI.

4. Productivity Is Not ROI

WRITER's 2026 survey captured the paradox precisely: 97% of executives report benefiting from AI personally; only 29% report significant organizational ROI. [1] Individual productivity gains of up to 5x are real - and they're invisible on the P&L because there's no operating-model change to absorb them. Hours saved by individuals don't show up in financials until departments are redesigned.

5. The Workforce Wasn't Redesigned

Deloitte's 2026 report flags this as one of the largest gaps: the #1 way companies adjusted their talent strategy was education - not role redesign or workflow redesign. [5] You cannot get AI-native outcomes from a workforce model built for pre-AI work. Training people to use AI inside the old job description leaves most of the value on the table.

Across BCG, MIT, Deloitte, WRITER and NVIDIA's 2026 reports, the "future-built" minority - the 5% capturing outsized returns - share a strikingly consistent profile. [2][3][5][6][7]

They tie AI directly to revenue outcomes, not to vague productivity metrics. Time saved is logged. Outcomes are what's measured and rewarded.

They invest in reshaping, not bolting on. Leading companies allocate more than 80% of their AI investment to reshaping functions and inventing new offerings, not to incremental productivity tooling. [8]

They go heavy on agents, early. Trailblazers direct over half their 2026 AI spend to agents, with end-to-end workstream deployment rather than chatbot point-solutions. [6]

They govern before they scale. Deloitte's 2026 research is explicit: enterprises where senior leadership actively shapes AI governance achieve significantly greater business value than those that delegate it to technical teams. [5]

They treat data and context as the moat. NVIDIA's 2026 State of AI found that 48% of enterprises cite data issues as the #1 challenge, and the leaders are the ones treating curated context and proprietary data as a first-class capability - not a cleanup project. [7]

They expect compounding, not instant, returns. BCG's research shows initial efficiency wins typically materialize in 6-18 months, meaningful financial impact in 18-36 months, and enterprise-level competitive effects in 3-5 years. [2] The 5% plan against that curve. The 70% don't and quietly defund before the compounding kicks in.

In January 2025, the Forbes framing was "75% aren't seeing ROI yet." In May 2026, the more accurate framing is:

Almost everyone is getting individual productivity from AI. Almost no one is getting enterprise transformation from it - because almost no one has redesigned the operating model that would let the value compound.

The technology side of AI is no longer the bottleneck. Models are abundant. Tools are mature. Costs are falling. What's missing is the harder, slower, less glamorous work: redesigning processes, redesigning roles, opening up enterprise capabilities to agents, governing the result, and measuring outcomes that show up on the P&L.

The 5% know this. They're running a different play. And the gap they've opened in 2025-2026 is wide enough that BCG, in its 2026 reports, has started warning that the window to close it is already narrowing. [4]

The 70% will need to stop running AI as a technology program and start running it as an operating-model program. The leaders who internalize that shift in 2026 will define the next decade. The ones who don't will spend it explaining to their boards why the AI ROI slide still looks the same.