Deloitte, EY, KPMG and PwC surveyed 7,000+ executives. We read all four AI reports so you don't have to - and the news is better than the headlines suggest.
By Denis Pesa, BlueMetrics
Deloitte, EY, KPMG and PwC have all published major AI studies over the past year, the most recent just weeks ago. Together they surveyed more than 7,000 senior executives across every major industry and region. That’s a lot of PDF. We read all of them, and a pattern emerged that most headlines are missing.
The story isn’t “AI is overhyped” or “AI will change everything.” It’s more specific, and more useful: AI is already paying off, the returns are wildly concentrated in a small group of companies, and what that group does differently is well documented and within reach.
Here are the six findings that matter most for a leadership team right now. After each one, I’ve added a quick note on what we’re seeing with our own clients, because survey data is one thing and Monday morning is another.
1. The returns are real, and they’re growing
Start with the number that should end the “is AI worth it?” debate: EY found that 96% of organizations investing in AI report AI-driven productivity gains over the past year, and 57% call those gains significant. Scale matters too: companies investing $10 million or more are far more likely to see significant gains than smaller spenders (71% versus 52%).
And here’s the detail I find most telling: companies aren’t pocketing those gains as cost cuts. EY found they’re reinvesting them, into more AI capability, R&D, cybersecurity and retraining. Only 17% are using AI gains to reduce headcount. That’s what conviction looks like on a budget line.
The other firms back this up. KPMG’s quarterly pulse shows 76% of leaders saying AI is delivering meaningful business value, up 12 points in a single quarter. Deloitte found the share of leaders calling AI’s effect on their company “transformative” doubled in one year, from 12% to 25%. And 79% told KPMG they would keep AI a top investment priority even in a recession.
So the question boards were asking in 2024, “does this work?”, is settled. The question for 2026 is different: why are some companies getting so much more out of it than everyone else?
From the field: We stopped having the “does AI work” conversation with clients about a year ago. The conversation now is almost always about a specific pilot that worked in the demo and is sitting in limbo. The appetite is there. The proof is there. The path to production is what’s missing.
2. Production is the bottleneck, not the technology
Here’s the uncomfortable number: Deloitte found that only 25% of companies have moved 40% or more of their AI experiments into production. Companies keep funding new pilots because pilots are cheap and safe, while the harder work of shipping the successful ones stalls. One healthcare AI leader in the Deloitte study called it “pilot fatigue”: a hundred pilots, no plan to scale any of them.
The good news hides in the same dataset. 54% of companies expect to cross that 40% threshold within three to six months. The crossing is happening right now, and the companies making it aren’t waiting for better models. A pilot runs on clean data with a small team in an isolated sandbox. Production needs integration, security review, monitoring, and someone on the hook when it breaks. Different job entirely.
From the field: Every stalled pilot we’ve been called into stalled for the same unglamorous reasons: no integration with real systems, no owner, no baseline to measure against. Never because the model wasn’t smart enough. When those three things get fixed, the move to production takes weeks, not quarters.
3. A small group is capturing most of the value, and we know why
PwC put numbers on the concentration: 20% of companies capture 74% of AI-driven returns, and the most “AI-fit” companies see performance 7.2 times higher than everyone else.
What separates them isn’t budget or headcount. PwC tested 60 management practices and found the leaders do a few specific things: they redesign workflows around AI instead of bolting tools onto old processes (2.2x more likely), they build reusable components instead of starting every project from scratch (2.4x more likely), and their employees actually trust and use what gets built (2.1x more likely).
The most encouraging line in the whole report: when companies with weak results put these foundations in place, they see roughly double the payoff from every subsequent AI project. The gap is a practice gap, not a destiny.
From the field: The workflow redesign point is the one companies resist most and benefit from most. Adding an AI tool on top of a broken process gives you a faster broken process. The clients getting real numbers let the technology change how the work flows, not just how fast one step runs.
4. Agents are arriving faster than the guardrails
This is the finding that should be on every risk committee’s agenda. Deloitte: 74% of companies plan to deploy agentic AI within two years, but only 21% have mature governance for autonomous agents.
It sounds like a warning, and it is. But flip it around and it’s the clearest opportunity in all four reports. PwC found that AI leaders treat governance as an accelerator: a governance board sets the policies, teams apply them through templates and checkpoints, and routine cases move fast because the rules are already clear. Companies with strong governance ship more, not less. And the market is starting to act on this. EY found responsible AI to be the one area where companies actually followed through on their promises: 60% increased the time spent training employees to use AI responsibly this past year, exactly what they said they’d do a year earlier. Deloitte says it plainly: organizations that build governance now will scale quickly and safely, while those that treat it as a checkbox will stay stuck in pilot mode.
From the field: We’ve started calling this “governance as a feature.” When guardrails, audit trails and human approval points ship with the first use case instead of after it, the security team goes from blocking deployments to sponsoring them. That single shift has unstuck more projects for us than any model upgrade.
5. Someone has to own the number
KPMG found that 75% of CEOs now personally champion AI. Sounds great. But only 24% of organizations can name who is actually accountable for AI outcomes, and that gap is expensive: companies with clearly defined accountability report established ROI at three times the rate of those without it.
Sponsorship and accountability are different things. A CEO saying “AI is a priority” is sponsorship. A named executive whose goals include a specific AI metric, with authority to act when it’s off track, is accountability. The first is everywhere. The second is rare, and it’s where the returns live.
EY’s data shows why this gap hurts. Among companies already seeing productivity gains from AI, 88% of leaders say those gains are a metric they’re evaluated on. Yet 65% admit their organization struggles to tie the gains directly to AI. People are being measured on a number nobody can trace. No wonder 92% say they need better ways to report AI’s value.
From the field: The first question we ask in any engagement is “who owns this number?” If the answer is a committee, or silence, that’s the real project. Tools don’t produce ROI. An owner with a baseline and a dashboard does.
6. You can’t scale what you can’t cost
The quietest finding might be the most practical. KPMG: organizations with full visibility into AI operating costs are five times more likely to report established ROI. Yet only about a third have that visibility, and nearly half have had to pause or scale back an AI deployment because costs crept past the value.
This is a new muscle for most companies. AI runs on usage-based pricing, so a use case that’s cheap in a pilot can get expensive at scale, and a use case that looks expensive can become very cheap with the right model choices. The leaders treat cost per case as a design decision made up front, not a billing surprise discovered later.
From the field: Cost per case is the metric that turns AI from a budget line into a business case. Once a CFO can see “this workflow costs $0.40 per document and saves $12,” scaling stops being a leap of faith. We now design that visibility in from day one, and it changes the entire conversation.
What this adds up to
Read together, the four reports describe a market at an inflection point. Adoption is broad, returns are real, and confidence keeps rising even under economic uncertainty. But value is concentrating fast among companies that do the unglamorous things well: ship pilots to production, redesign the work, govern the agents, name an owner, and know their costs.
None of those requires a bigger budget than you have. All of them require deciding to operate differently. That’s genuinely good news, because practices can be copied and hired for. Head starts can’t be bought back later, though, and PwC’s warning is worth taking seriously: the leaders’ advantage compounds, because every deployment makes the next one faster and cheaper.
This is the first article in a series. In the coming weeks we’ll go deeper on the topics these reports opened up: what mature agent governance actually looks like, how to build cost visibility into AI operations, and how the companies in the “7.2x club” structure ownership. If one of those hits close to home, the next posts are for you.
BlueMetrics is an applied AI firm based in Charlotte, NC, and a member of the Claude Partner Network. We help mid-market companies take AI from stalled pilot to production.
Sources: Deloitte, State of AI in the Enterprise, January 2026 (3,235 leaders, 24 countries) · EY, US AI Pulse Survey, Wave 4 “The dividend age”, December 2025 (500 US senior leaders) · KPMG, Global AI Pulse Q2 2026, June 2026 (2,145 leaders, 20 countries) · PwC, AI Performance Study “Want ROI from AI? Go for growth”, April 2026 (1,217 companies, 25 sectors)