Which Industries Are Actually Winning With AI in 2026?

Updated June 22, 2026

A bank can tell you, to the dollar, how much fraud its model caught last quarter. A hospital using AI to triage radiology reads can show turnaround times before and after deployment. A logistics team running demand-forecasting models can point to the inventory write-downs that didn't happen. These are boring wins. They're the only ones that matter.

Most AI spending isn't boring, and most of it can't be measured. In its 2024 State of AI survey, McKinsey found 72% of organizations report using AI in at least one business function — but a much smaller share could tie that use to a specific earnings impact. Goldman Sachs put a sharper point on it in a June 2024 research note titled "Gen AI: Too Much Spend, Too Little Benefit?". The conclusion: capital spending on AI infrastructure is running well ahead of any earnings effect anyone can name.

---

Banks Are Winning — But Mostly at Boring Stuff

The headline finance-AI wins aren't trading algorithms. Algorithmic and high-frequency trading already accounts for most US equity volume. Estimates from academic and regulatory studies — including SEC market structure research — put the combined algorithmic and high-frequency trading share at roughly 50 to 70 percent of US equity volume, depending on the methodology and time period., depending on whose count you trust. Most of that volume has run on rule-based systems and classical machine learning for two decades — not the technology vendors are pitching today. So when a vendor deck claims AI runs 70% of US equity markets, they're counting twenty-year-old rule engines. Discount accordingly.

What's actually working in finance is back-office work. JPMorgan's Contract Intelligence (COIN) contract-review system, first disclosed by Bloomberg in 2017, parsed commercial loan agreements that used to consume roughly 360,000 lawyer-hours a year. The 2024–2026 generative AI wave is mostly extending that pattern — Know Your Customer (KYC) document review, call deflection, suspicious-activity-report drafting. Banks measure return on investment well because fraud teams have always tracked losses by basis points. Drop a model into that pipeline and the comparison is automatic.

---

Retail and CPG: AI as Margin Defense

Consumer packaged goods (CPG) companies live on thin margins — typically low-to-mid single digits in food and staples retail (NYU Stern School of Business industry margin data, 2024 — food retail net margins average 1–3%, compared with 10–15% for branded CPG manufacturers such as P&G) — which is why the sector's AI bets concentrate almost entirely on cost.

Demand forecasting is the marquee use case. Walmart and Procter & Gamble have both publicly described demand-sensing systems that pull point-of-sale data, weather, and promotional calendars into forecasts that refresh daily — or hourly during a product launch — instead of the weekly batch cycle most retailers ran a decade ago (per Walmart and P&G investor relations disclosures and earnings commentary, 2023–2024). The practical payoff is concrete: when a heatwave hits the Southeast on a Tuesday, beverage inventory gets rerouted before the weekend instead of the following Monday. Neither company publishes dollar savings line by line, so treat any article quoting a specific figure with skepticism. The other live use case is store operations — shrink detection, planogram compliance, labor scheduling. These don't show up as "AI revenue." They show up as a few basis points of margin that compound quietly over several quarters.

---

Healthcare's Quiet, Narrow Wins

Healthcare AI hype runs hot. The actual track record is narrower — and more interesting — than the headlines suggest.

Two wins are well-documented. Paige.AI received Food and Drug Administration (FDA) De Novo authorization for its prostate-cancer pathology system in September 2021 — the first AI-powered pathology toll authorized by the FDA for use in oncology. Google Health's mammography model matched or beat radiologists in a 28,000-patient study published in Nature (McKinney et al., 2020). The pattern: AI does well on narrow, high-volume image-reading tasks where the ground truth is verifiable, not as a general "AI doctor."

Operationally, the win is revenue cycle management (RCM) — coding visits, submitting claims, chasing denials. Epic's autonomous coding tools, bundled into its electronic health record suite, are among the most widely referenced implementations in published health system case studies (Epic Systems, Revenue Cycle documentation and client outcomes library). Health systems using them describe coding turnaround moving from weeks to days, which pulls cash forward and shortens the denial-rework loop.

The Johns Hopkins TREWS sepsis-detection system, published in Nature Medicine in 2022, is the rare clinical-AI study with hard outcome data: earlier antibiotic administration and a measurable mortality reduction across more than half a million patient encounters. One tool, one health system, one condition. Generalize cautiously.

---

What the Winners Have in Common

You don't need a framework to see the pattern. (Editorial note: the original outline called for four traits keyed to scaling-adoption figures from McKinsey, Deloitte, and Litera. We couldn't independently verify those specific percentages against primary documents, so we've replaced the citations with four directly observable behaviors.) The companies pulling AI through to earnings share four traits.

They pick a profit and loss (P&L) line before they pick a tool — fraud loss, claims-processing cost per file, contact-center average handle time. The metric exists before the model does, which makes the comparison honest.

They treat generative AI as one option among several. Sometimes the right answer is a rules engine. Sometimes it's a classical machine learning model that's been in production since 2015. Companies that ask "what's the cheapest way to move this number?" outperform the ones asking "where can we use generative AI?"

They invest in the unsexy work around the model — data pipelines, retraining cadence, drift monitoring. The model is maybe 20% of what makes a deployment work. Skip the other 80% and pilots quietly die a year later.

And the initiative reports to the business owner whose number actually moves, not to a chief data officer with no operating accountability. When the head of fraud owns the fraud-detection model, it gets retrained the week it drifts. When "the AI team" owns it, retraining gets scheduled for next quarter and stays there.

---

Why Most Firms Stall

The blockers repeat across sectors. McKinsey's 2024 survey found that firms reporting earnings before interest and taxes (EBIT) impact from AI are concentrated in a small minority — and the gap usually traces to three things. First, vague goals: "improve the customer experience" isn't a target a model can be measured against, while "cut average handle time by 90 seconds" is. Second, legacy systems: pilots die not because the model is bad but because nobody can get clean data out of a 1990s claims platform. Third, culture: when frontline staff suspect a pilot is a dressed-up layoff plan, the data collection it depends on quietly stops happening.

The fix isn't a bigger budget. It's a shorter list of questions, asked before you sign anything.

---

Sources & Methodology

• McKinsey, The State of AI 2024 — McKinsey
• Goldman Sachs Research, "Gen AI: Too Much Spend, Too Little Benefit?" June 2024 — Goldman Sachs
• US Securities and Exchange Commission, Equity Market Structure resources — SEC
• McKinney et al., "International evaluation of an AI system for breast cancer screening," Nature 2020 — Nature
• Paige.AI, FDA De Novo authorization for Paige Prostate, 2021 — FDA
• Adams R. et al., "Prospective, multi-site study of patient outcomes after implementation of the TREWS machine learning-based early warning system for sepsis," Nature Medicine 2022 — Nature
• Bloomberg, "JPMorgan Software Does in Seconds What Took Lawyers 360,000 Hours," 2017 — Bloomberg

Where dollar-savings figures come from corporate sources, they're attributed and qualified. Where a figure couldn't be sourced to a verifiable document, we left it out. Data as of May 2026.

---

Found an error? At Canopy Press, accuracy comes first. If you spot a claim that needs checking, let us know at [email protected] — we'll verify and correct it immediately.

Sources

• How AI Is Driving Revenue, Cutting Costs and Boosting Productivity for Every Industry in 2026
• From Pilot to Profit: NVIDIA AI in Financial Services Survey 2026
• PwC 2026 AI Performance Study
• McKinsey State of AI 2025: Agents, Innovation, and Transformation
• Goldman: AI productivity 30% boost on 2 specific use cases
• Goldman Sachs: Why AI Companies May Invest More than $500 Billion in 2026
• Morgan Stanley AI Market Trends Institute 2026
• Deloitte 2026 Software Industry Outlook
• Oliver Wyman: AI in revenue cycle delivering results across healthcare
• KPMG: AI adoption in finance doubles