An estimated 230 million people now use AI tools for health-related queries every week. More than a third consult AI before seeing a doctor. Nearly a million US clinicians have submitted 271 million queries to a single AI platform in just two and a half years. And yet, according to a major OECD report published on 7 October, more than 90% of healthcare executives report no measurable impact on productivity.
That gap — between AI everywhere and AI that actually moves the needle — isn't a healthcare problem. It's an every-industry problem. And the evidence is now too consistent to ignore.
The largest AI adoption study ever — and its uncomfortable finding
The OECD report, produced from a conference in Madrid that brought together 142 experts from 34 countries, was designed to shift the conversation from "should we adopt AI?" to "how do we adopt it effectively?" The finding that stopped that conversation cold: health systems are deploying AI at unprecedented speed, but almost none can demonstrate it's making them more productive.
The scale of adoption is staggering. A JAMA study published 15 October quantified just one AI platform — OpenEvidence — and found 952,008 US clinicians had submitted 271 million queries between January 2024 and June 2026. Monthly active users grew from 355 to 703,265. Physicians averaged 37.1 queries per active user-month, primarily for treatment and management decisions (33.1% of queries) rather than diagnosis (13.9%).
Clinicians aren't dabbling. They're relying on AI for the hardest part of their job — deciding what to do about a patient's condition. But as a Nature Medicine editorial published the same week put it: "A gap remains between AI hype and clinical reality. Evidence remains limited on whether the deployment of these tools is having any real impact on patients, healthcare providers or health systems."
This isn't just a healthcare story
If you run a 30-person services firm, not a hospital, you might be tempted to file this under "not my problem." Don't. The healthcare data is just the clearest, most rigorously documented version of a pattern that McKinsey's 2026 State of AI survey found across every sector.
The headline numbers: 89% of organisations now use AI in at least one business function. But only 6% qualify as "high performers" — organisations attributing at least 5% of earnings to AI. That share hasn't budged in a year. Adoption is spreading, but returns aren't following.
The individual-level data makes the paradox sharper. McKinsey found 80% of respondents say AI has improved their personal productivity, and 50% say it helps them make better decisions. But those individual gains aren't showing up in organisational financials. As McKinsey noted, a study of 5,200 customer-support agents found AI assistance increased issues resolved per hour by 15%. A trial with 4,900 developers found 26% more tasks completed. "Neither finding establishes that adding AI to any business process will improve a company's earnings."
This mirrors exactly what we've seen in previous stories. Uber burned through its entire 2026 AI budget in four months chasing productivity gains that didn't materialise at the organisational level. Starbucks pulled its AI inventory system from 11,000 stores after nine months because operational gains never met expectations. The pattern is consistent: individual tasks get faster, but the business doesn't measurably improve.
The measurement mistake
The most actionable finding in this wave of research comes not from healthcare but from Wharton's 2026 AI Adoption Report, published 13 October. The study, now in its fourth year, surveyed enterprise leaders and — for the first time — compared their perspectives with entry- and mid-level employees.
The central insight: departments that measure AI against growth outcomes — net-new revenue, sales conversion, output volume, customer satisfaction — are nearly three times more likely to report strongly positive ROI than those measuring efficiency alone. Specifically, 35% of growth-oriented departments report strong returns, versus just 13% of those tracking only efficiency metrics.
Yet the most common way organisations judge AI investment is hours saved (40% of departments). Net-new revenue is the least-used major measure (27%).
"Companies believe AI is going to transform their industries, but many are still asking it to prove itself by saving time," said Sonny Tambe, Professor of Operations, Information and Decisions at the Wharton School. "The companies seeing stronger returns are going further. They are redesigning workflows, putting clear ownership and governance in place, and measuring AI against outcomes that matter to growth."
This echoes what the MGMA found when it polled 257 medical practice leaders in May 2026: less than half (46%) said AI tools had made providers more productive. Among those reporting no gains, leaders pointed to resistance to change, poor system interoperability, and a critical observation — even where AI cut documentation time, it hadn't translated into more patients seen or revenue generated.
What this means for your business
The OECD report identified three conditions for AI to deliver measurable impact at scale: trust, foundation, and economics. Those conditions translate directly to any business:
Trust means your team actually uses it. The MGMA data showed adoption rates plateau between 20% and 50% even when tools are broadly available. If half your staff quietly ignores the AI tools you're paying for, no amount of capability will show up in your numbers.
Foundation means your data and workflows are ready. The Nature Medicine editorial highlighted that deploying the same AI system across India, Thailand and Australia required "vastly different implementation strategies shaped by local healthcare priorities and clinical workflows." The tool is the easy part. The workflow redesign is where the value lives — and where most organisations stop short. As we noted when Australia ranked last on AI integration, 88% of businesses use AI but only 10% have actually embedded it into core operations.
Economics means measuring what matters. The Wharton data is unambiguous: if you're only tracking "hours saved," you're almost certainly undervaluing or mismeasuring your AI investment. Start with the question "what would this AI need to do to generate revenue or serve more customers?" and work backwards.
What to watch
The OECD's Expert-Led Madrid Action Plan (EL-MAP) proposes ten priority actions for governments and health systems, centred on building evaluation frameworks that measure real-world impact rather than benchmark performance. If those frameworks gain traction, they'll create a template that every industry can borrow.
More immediately, watch for Gartner's finding that only 13% of organisations report significant or transformational value from AI to become the baseline that boards start asking about. The era of "we're using AI" as a sufficient answer to shareholders, customers, and employees is ending. The next question — "what has it actually done?" — is the one most organisations aren't ready for.
At Heygentic, we've seen this gap firsthand. The businesses getting real returns from AI aren't the ones with the most tools or the biggest budgets. They're the ones who picked one workflow, measured it properly before and after, and redesigned the process around what AI actually does well — not what the pitch deck promised.
Sources
- Proceeding report from Madrid conference on the responsible scale of AI in healthcare — OECD
- Clinician Use of an AI-Enabled Clinical Information Platform — JAMA
- Making AI work in healthcare — Nature Medicine
- The State of AI: Global Survey 2026 — McKinsey
- Why AI's easiest wins are misleading CEOs — McKinsey
- The Adaptive Advantage: Who's Winning With AI and How — Wharton & GBK Collective
- Are AI tools making clinicians more productive? It's complicated — MGMA
- More AI Investment Is Not Creating Enough AI ROI — Gartner
