The AI ROI Evidence Base: What 7 Major Reports Actually Prove

Table Of Contents
- The Gap Every Executive Needs to Understand
- Report 1 — McKinsey State of AI 2026: The Productivity-to-Profit Disconnect
- Report 2 — IDC/Microsoft Business Opportunity of AI: The $3.70 Benchmark
- Report 3 — MIT Project NANDA: Why 95% of Pilots Fail to Move the P&L
- Report 4 — IBM Institute for Business Value: The CEO Expectation Gap
- Report 5 — Stanford HAI AI Index 2026: Adoption Is Solved, ROI Is Not
- Report 6 — PwC Global AI Jobs Barometer 2026: The Superstar Effect
- Report 7 — BCG Applied AI Index 2026: Nearly Half Are Generating Value
- What All 7 Reports Agree On: The Three Separators
- From Evidence to Action
The AI ROI Evidence Base: What 7 Major Reports Actually Prove
There is no shortage of AI statistics. What most executives actually lack is a way to read them together — to understand why one credible report says 74% of companies are seeing ROI while another says 95% of pilots produce zero measurable P&L impact, and why both numbers can be simultaneously true.
The answer lies in what each report is measuring, and for whom. When you read the seven most significant AI ROI research reports published between 2024 and 2026 as a single evidence base rather than competing claims, a coherent and actionable picture emerges. Adoption is near-universal. Individual productivity gains are real and consistent. Enterprise-level financial returns, however, remain concentrated in a small group of organizations that share specific practices — not superior technology.
This article breaks down each of the seven reports, extracts what it actually proves, and identifies the three patterns that separate AI value-generators from the majority still waiting for their investment to show up on the income statement.
The Gap Every Executive Needs to Understand {#the-gap}
Before examining the individual reports, it helps to understand why the headline numbers appear to contradict each other. The core issue is that AI ROI research measures different layers of the same journey. Task-level productivity gains (a developer coding faster, a support agent resolving more tickets per hour) are common and measurable within weeks. Function-level financial impact (cost reductions in supply chain, revenue lift in marketing) is achievable within months to a year or two. Enterprise-level EBIT movement — the kind that shows up in a company's overall earnings — is rare and typically takes two to four years to materialize after initial deployment.
Most optimistic surveys measure the first layer. Most pessimistic surveys measure the third. The gap between them is not a measurement error. It reflects a genuine organizational challenge: getting task-level gains to compound into enterprise-level financial results requires deliberate workflow redesign, strong governance, and committed leadership. That is the central finding across all seven reports reviewed here.
Report 1 — McKinsey State of AI 2026: The Productivity-to-Profit Disconnect {#mckinsey}
Source: McKinsey Global Survey, August 2026 (1,719 respondents, 97 countries)
McKinsey's annual State of AI survey is the most widely cited longitudinal benchmark for enterprise AI adoption. The 2026 edition, titled On the Road to ROI, captures a defining tension in the current moment.
On the adoption side, the picture is unambiguously positive. Nearly nine in ten respondents say their organizations use AI regularly in at least one business function. 44% report AI is scaling across their enterprise, up from 38% the year prior. Among large organizations (annual revenues above $1 billion), the share scaling AI agents jumped from 27% to 40% in a single year.
Individual impact is also strong. 80% of respondents say AI has improved their personal productivity, and 50% say it helps them make better decisions at work. These are not trivial numbers — they represent a genuine shift in how knowledge work gets done.
But the enterprise financial story is flat. Only 37% of respondents attribute any EBIT impact to AI use, a figure essentially unchanged from 2025. Only 6% of respondents qualify as AI high performers — defined as attributing at least 5% of EBIT to AI and describing its impact as significant. That share, too, is unchanged from the prior year, despite rising investments and broadening adoption.
What separates that 6%? McKinsey finds that high performers are 3.3 times more likely to intend fundamental business transformation through AI (rather than just efficiency gains), nearly three times more likely to be scaling AI agents across most business functions, and twice as likely to have senior leaders who demonstrate visible commitment to AI initiatives. They also redesign workflows rather than layer AI onto existing ones — nearly three-quarters of high performers report fundamentally redesigning workflows because of AI, compared with just one-quarter of other respondents.
The McKinsey data also reveals where functional-level value is appearing even when enterprise-level EBIT hasn't shifted. Cost reductions are most commonly reported in supply chain management, service operations, and manufacturing. Revenue gains concentrate in marketing and sales, followed by product development and software engineering.
What this report proves: The productivity-to-profit translation problem is real and structural. Individual gains are widespread. Enterprise financial impact is not. The difference is operating model change, not technology access.
Report 2 — IDC/Microsoft Business Opportunity of AI: The $3.70 Benchmark {#idc}
Source: IDC InfoBrief sponsored by Microsoft, November 2024 (4,000+ enterprise respondents globally)
The most cited positive ROI figure in the AI space comes from IDC's 2024 study: for every $1 invested in generative AI, organizations report an average return of $3.70. Among the top-performing segment of adopters, that figure rises to $10.30 per dollar.
IDC also found that generative AI deployments take less than eight months on average, and organizations typically realize initial value within 13 months of deployment. These are meaningfully faster timelines than most enterprise technology transformations, and they help explain why AI investment budgets continue to rise even as measured enterprise-level returns remain uneven.
The report identifies productivity as the primary current ROI driver — 92% of AI users deploy it to improve employee productivity, and 43% say productivity use cases have delivered the greatest ROI so far. However, IDC anticipates a shift: within two years, more organizations expect to move from productivity-focused deployments to functional and industry-specific use cases in product development, manufacturing, and beyond.
One important context note: this study was sponsored by Microsoft, and its sample skews toward committed, relatively advanced adopters. The $3.70 average describes companies that have invested substantively in AI and integrated it into operations — not the full population of organizations that have purchased AI tools or run pilots.
What this report proves: Committed AI adopters that integrate the technology across operations can generate meaningful per-dollar returns, with productivity as the near-term ROI engine and functional depth as the next frontier. The 13-month value realization window gives finance teams a realistic planning horizon for initial deployments.
Report 3 — MIT Project NANDA: Why 95% of Pilots Fail to Move the P&L {#mit}
Source: MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 (300 public AI initiatives reviewed; ~150 executive interviews; 350 employee survey responses)
If the IDC figure is the most cited optimistic data point, the MIT NANDA finding is its counterweight: 95% of enterprise generative AI pilots produced no measurable profit-and-loss impact, measured six months after deployment.
This sounds devastating. Read in context, it is clarifying. MIT NANDA notes that six months is a short measurement window by enterprise software standards. More importantly, the research identifies that the divide between successful and unsuccessful AI deployments is not driven by model quality or regulatory environment — it is driven by approach.
The roughly 5% of pilots that succeed share a recognizable pattern: they target a single, well-defined pain point, execute with discipline, and build genuine partnerships with the technology providers whose tools they use. The majority that fail share a different set of characteristics: AI is bolted onto existing workflows without redesign, baseline metrics were not established before deployment, and pilots have no clear path from proof-of-concept to production.
This pilot-to-production gap is a consistent theme across the broader evidence base. MIT's data makes it concrete: without a measurement framework and a redesigned workflow, task-level gains stay trapped at the individual level and never reach the income statement.
What this report proves: The failure rate of AI pilots is a workflow and measurement problem, not a technology problem. Organizations that pick one pain point, measure rigorously, and redesign the relevant workflow consistently cross into the successful minority.
Report 4 — IBM Institute for Business Value: The CEO Expectation Gap {#ibm}
Source: IBM Institute for Business Value CEO Study, 2025 (2,000 CEOs surveyed)
IBM's research brings the CEO perspective into sharp focus. When asked how their AI programs had actually performed, only 25% of AI initiatives had delivered the ROI that CEOs expected. Only 16% of organizations had scaled AI enterprise-wide.
Despite this, 85% of CEOs anticipate a positive return from scaled AI efficiency and cost savings by 2027. This combination — lower-than-expected current returns and sustained future optimism — reflects what IBM describes as CEOs balancing pressure for short-term ROI with longer-term innovation goals.
IBM also identifies an organizational barrier that compounds the problem: half of CEOs surveyed said the pace of recent AI investment had left them with disconnected, piecemeal technology rather than integrated capability. This fragmentation is expensive. IBM's data suggests that companies carrying significant AI-related capability debt see project returns drop by 18 to 29%, with implementation timelines stretching by up to 22%.
What this report proves: The expectation gap is wide and acknowledged at the highest levels. Three-quarters of AI initiatives have not met CEO expectations, primarily because of fragmented deployment, capability debt, and the absence of enterprise-wide integration. Organizations that build a coherent platform before scaling show materially better outcomes.
Report 5 — Stanford HAI AI Index 2026: Adoption Is Solved, ROI Is Not {#stanford}
Source: Stanford Institute for Human-Centered AI, AI Index Report 2026, Economy Chapter (April 2026)
Stanford's AI Index provides the broadest view of AI's economic footprint. It reports that 88% of surveyed organizations used AI in 2025, up from 55% in 2023 — adoption, in Stanford's framing, is essentially a solved problem. Generative AI is now used in at least one business function at 70% of organizations globally.
On productivity, Stanford synthesizes the academic literature to find gains of 14% to 26% in customer support and software development — meaningful improvements in structured, high-volume tasks. The same research identifies weaker or even negative productivity effects in tasks requiring more judgment and contextual reasoning.
One of Stanford's most striking findings is a consumer surplus estimate: the value of generative AI tools to U.S. consumers reached $172 billion annually by early 2026, with the median value per user tripling in a single year. This measure of unrealized economic value suggests that the social returns from AI technology far exceed the financial returns being captured by the organizations deploying it.
For business leaders, the Singapore context is particularly relevant. Stanford's data shows Singapore at 61% individual AI adoption — among the highest globally, well ahead of the United States at 28.3% — which reflects both government-level AI investment and enterprise readiness across the region.
What this report proves: Adoption is no longer a differentiator. The question every organization now faces is not whether to use AI, but whether its operating model can convert usage into measurable financial results. Productivity gains are real but task-specific. The value creation opportunity is large and largely uncaptured.
Thinking about how to build that operating model? Business+AI's consulting services help executive teams identify the highest-ROI use cases and design the organizational conditions that allow pilots to reach the P&L.
Report 6 — PwC Global AI Jobs Barometer 2026: The Superstar Effect {#pwc}
Source: PwC Global AI Jobs Barometer 2026 (analysis of close to 1 billion job postings across 27 countries and territories)
PwC's research adds a dimension that most AI ROI surveys miss: what separates the best-performing companies from the rest is not just that they use AI, but how dramatically they diverge from the average.
Overall, companies most exposed to AI are seeing productivity growth of 34% relative to a 2018 baseline. But zoom in on the top 20% within that already-exposed group and the number rises to 163%. This is what PwC calls the superstar effect — a small population of organizations that have unlocked compounding returns from AI by redesigning how they operate, not just adding tools to existing processes.
PwC also finds something counterintuitive: greater AI exposure is linked to headcount growth, not decline. The most AI-exposed organizations are growing headcount at double the rate of the least exposed, with wages also growing significantly faster. The organizations achieving the largest productivity gains are the ones taking the redesign route — using AI to create new capacity rather than simply reduce headcount.
The data further shows that companies applying AI widely across products, services, and customer experience report margins nearly 4 percentage points higher than those applying it only internally. The strategic ambition of AI deployment, in other words, has a measurable margin impact.
What this report proves: The distribution of AI returns is highly skewed. Most organizations see modest gains. A small group, defined by their willingness to redesign operations rather than just automate tasks, is capturing outsized value. The superstar effect is not accidental — it is the result of deliberate strategic choices about where and how AI gets integrated.
Want to understand how leading organizations in Asia are making those choices? The annual Business+AI Forum brings together executives and practitioners to share what is actually working — with evidence, not aspiration.
Report 7 — BCG Applied AI Index 2026: Nearly Half Are Generating Value {#bcg}
Source: BCG Applied AI Index 2026 (1,300+ senior leaders and executives across 20+ sectors)
BCG's 2026 research offers an encouraging mid-year signal: nearly 50% of surveyed companies now generate some value from AI, up from earlier BCG data showing only 26% had moved beyond proof of concept. Within BCG's classification framework, 7.5% of organizations qualify as "future-built" — the highest maturity tier — while 41% are actively scaling AI with room to improve results.
BCG also projects that agentic AI will account for approximately 40% of all AI value by 2030, while noting that only 5% of companies currently have the full governance and control frameworks in place to operate AI agents at scale. This gap between agent ambition and agent readiness is a significant near-term risk.
Perhaps the most practically useful BCG finding concerns investment allocation. BCG's 10/20/70 framework — allocating 10% of AI investment to algorithms, 20% to technology infrastructure, and 70% to people and processes — reflects the empirical finding that workflow redesign and change management are the dominant determinants of AI ROI, not model selection. BCG data shows that leading companies solve the value gap by allocating the majority of investment to reshaping key functions and inventing new offerings, focusing on an average of 3.5 use cases compared to 6.1 for lower-performing peers. PwC adds a complementary finding: organizations with established responsible AI frameworks and enterprise-wide integration are three times more likely to report meaningful returns.
What this report proves: Progress is real but unevenly distributed. The organizations generating the most AI value are investing more in organizational change than in technology. Those that try to cover too many use cases simultaneously are consistently outperformed by those with focused deployment discipline.
Looking for structured guidance on AI use case prioritization and deployment? Business+AI's workshops and masterclasses are designed to give leadership teams the frameworks and practical tools to move from pilot to production — with BCG-aligned deployment logic built in.
What All 7 Reports Agree On: The Three Separators {#three-separators}
Read side by side, the seven reports share a consistent set of findings about what separates organizations that capture AI value from those that do not. The differences are not about which AI tools an organization uses. They are about how the organization is managed.
1. Workflow Redesign, Not Task Automation
Every report that examines high-performing AI organizations finds the same characteristic: they redesign workflows around AI capabilities rather than inserting AI into existing processes. McKinsey's high performers redesign workflows at three times the rate of average adopters. MIT NANDA's successful 5% all targeted a specific workflow and rebuilt it. BCG allocates 70% of recommended AI investment to people and processes. The implication is direct: if AI is being added to an unchanged workflow, the enterprise P&L is unlikely to move.
2. Measurement Before Deployment
Harvard Business Review analysis of 847 enterprise AI deployments found that companies establishing success metrics before rolling out AI tools are four times more likely to achieve ROI than those that deploy first and measure later. MIT NANDA confirms the inverse: missing baseline metrics is one of the three primary causes of pilot failure. KPMG's Global AI Pulse finds that organizations where the CEO is accountable for AI-driven outcomes report established ROI at 14% — versus just 4% at organizations without that accountability. Measurement discipline is not a reporting exercise; it is a value-creation discipline.
3. Strategic Depth Over Breadth
BCG's data shows that organizations generating the most AI value focus on 3.5 use cases on average, compared with 6.1 for lower performers. McKinsey's high performers deploy AI to pursue growth and innovation alongside efficiency, and are 3.3 times more likely to intend fundamental business transformation. PwC's superstar companies apply AI to products, services, and customer experience — not just internal operations. Spreading AI investment thin across many shallow deployments is a consistent characteristic of organizations reporting no enterprise-level financial impact.
From Evidence to Action {#action}
Seven reports, thousands of executive respondents, multiple measurement methodologies — and the same three lessons keep appearing. The evidence base for AI ROI is not contradictory. It is hierarchical: adoption is near-universal, individual productivity is well-established, and enterprise financial impact belongs to the organizations willing to do the harder organizational work.
The number that matters most right now is not $3.70 per dollar (IDC's optimistic benchmark) or 95% failure rate (MIT's cautionary signal). It is 6% — the share of organizations McKinsey identifies as genuine AI high performers, with their share unchanged for two years despite rising investment and adoption. That figure is both a warning and an opportunity. The path those organizations took is well-documented. The question for every executive team is whether they are following it.
Businesses across Asia that are ready to close the gap between AI adoption and AI returns — through structured use case selection, workflow redesign, and the leadership practices the evidence consistently identifies — will find that the tools are available and the playbook is proven. The variable is organizational intention.
Take the Next Step with Business+AI
Business+AI is Singapore's dedicated ecosystem for executives turning AI investment into measurable business results. Whether you're benchmarking your organization against the evidence in this article, selecting your highest-priority use cases, or building the leadership and governance capabilities the research identifies as critical — we have the resources to help.
Join the Business+AI membership community to access forums, workshops, masterclasses, and Asia's most practitioner-focused annual AI event — built for business leaders who want returns, not just adoption.
Meta title suggestion: The AI ROI Evidence Base: What 7 Major Reports Prove | Business+AI Meta description suggestion: McKinsey, IDC, Stanford, MIT, IBM, PwC, BCG — 7 major AI ROI reports tell a coherent story. Here's what the evidence base actually proves about AI returns.
