BLOG

The state of AI in mid-market operations 2026: what the numbers actually say

Tucker Bern, Founder

Every few months the big AI surveys land, and I read them the way I read a deal memo: skip the adjectives, find the tables. I deploy AI systems inside mid-market companies at Meet Caddy, and before that I built them inside my own family of companies, utility locating, military drones, manufacturing, real estate, healthcare. So when the 2026 numbers came out, I pulled everything relevant to operators in the middle of the market and lined it up in one place.

This report is built entirely on public data. Every number below is linked to its source, I verified each one on the page before citing it, and where the surveys disagree with each other, I tell you why. What follows is the data first, my read second.

Key findings: US firms with 250 or more employees now use AI in operations at nearly twice the national business rate (37% vs 19.8%), while adoption among the smallest firms has stopped growing entirely [1]. In the mid-market, 82% of companies report AI in production somewhere, but only 26% have it scaled and governed [2], and roughly 95% of generative AI pilots still show no measurable P&L impact [3]. Across every dataset, the recurring blockers are ownership, governance, and change management, not model capability. The mid-market stall is a leadership gap, not a technology gap.

How wide is the gap between enterprise and mid-market AI adoption?

Start with the strictest measurement in the country. The Census Bureau's Business Trends and Outlook Survey asks businesses a hard question: did you use AI to produce goods or services in the last two weeks. Not "do you have an AI strategy," not "has anyone on your team tried a chatbot." Actual AI in actual production, right now.

By that standard, 19.8% of US businesses were using AI as of May 3, 2026, with overall use hovering between 17% and 20% from December 2025 through May 2026 [1]. Now cut it by size. Firms with at least 250 employees: 37%. Firms with 100 to 249 employees: 32%. And here's the line that should bother every mid-market owner: AI use increased among firms with at least 20 employees over that stretch, but didn't change significantly among firms with fewer than 20 [1]. The big companies are still climbing. The small ones have flatlined.

The Federal Reserve looked at the same landscape in April 2026 and added the number that tells you where this is heading. Firm-level adoption stood at about 18% at the end of 2025, but 78% of the US labor force works at firms that have adopted AI, and about 41% of workers report using generative AI for work [4]. Read those together. Adoption is concentrating where the headcount is. Enterprises are absorbing AI into how work gets done while most smaller firms watch.

Meanwhile the self-report surveys tell a sunnier story, and the difference matters. The US Chamber of Commerce found 58% of small businesses saying they use generative AI in 2025, up from 40% in 2024 and 23% in 2023 [5]. That's not a contradiction, it's a different question. "We use generative AI" usually means somebody in the shop has a chatbot open. The Census question means AI is doing production work the business depends on. Both things are true at once: most small and mid-size companies now touch AI, and very few have AI doing accountable work in operations. Honestly, the whole story of 2026 lives in that gap.

At the top of the market, Stanford's 2026 AI Index reports that 70% of organizations now use generative AI in at least one business function and 88% use AI in some form, while AI agent deployment remains in the single digits across nearly all business functions [6]. Even the leaders are early. Which means the gap is still closeable, for now.

What do the 2026 numbers actually say?

One note before the list. These surveys ask different questions of different populations, which is why a 19.8% and an 88% can both be correct. I've kept each number attached to what was actually asked.

Adoption and the size gap

Production, value, and stall rates

Leadership, governance, and talent

Where does the AI money go, and where does the value come from?

Here's the finding I'd frame and hang on the wall. The MIT NANDA research found that more than half of generative AI budgets go to sales and marketing tools, while the better ROI shows up in back-office automation [3]. Companies are spending where AI is visible and getting paid where it's boring.

Stanford's data backs that up from the other direction. The documented productivity gains cluster in structured, measurable work: 14% to 15% in customer support, 26% in software development, 50% in marketing content output [6]. AI pays where the work has clear inputs, clear outputs, and somebody counting.

And notice what Deloitte found: nearly three-quarters of organizations say their most advanced initiative meets or beats ROI expectations [8], even while most of their experiments never scale. The technology works when it ships against a measured workflow. Most of it just never ships.

That matches what I've seen inside my own companies. The AI that earned its keep was never the impressive demo. It was document handling, scheduling, follow-ups that used to slip, quotes that used to take days, the unglamorous paper-moving layer every operations-heavy business runs on. Nobody screenshots that stuff for LinkedIn. It's also where the money is.

One more number from that MIT research worth taking personally: purchased tools and partnerships succeeded about 67% of the time, while internal builds succeeded about one-third as often [3]. Mid-market companies that try to build like a Fortune 500 R&D lab are choosing the failure path with the least room for error.

Is the stall a technology problem or a leadership problem?

Look at the failure reasons the analysts actually recorded. Gartner's list: poor data quality, inadequate risk controls, escalating costs, unclear business value [7]. The executive benchmark survey: 93.2% say culture and change management, not technology [10]. Grant Thornton: three in four boards approved the money, but almost half never set governance expectations for it [9]. Not one item on any of those lists is a model limitation. Every one of them is something an accountable owner fixes.

Then there's the readiness delusion inside the leadership team itself. CIOs and CTOs are five times more likely than COOs to say the workforce is ready for AI, 44% versus 9% [9]. The people buying the technology and the people running the operation are describing two different companies. When your technology chief thinks the team is ready and your operations chief knows it isn't, pilots get launched into workflows nobody prepared, and then everyone blames the AI.

The mid-market version of this is sharper. Netrio's survey of companies between 200 and 5,000 employees found 82% with AI in production somewhere, and only 26% with it scaled and governed. Just 42% have formal AI policies with enforced controls, and 42% confirmed an AI-related security incident in the past twelve months [2]. That's not a picture of companies that lack technology. That's a picture of technology running around without an owner.

The way I think about it: enterprises solved the ownership problem by paying for it. Even so, only 38.5% of Fortune 1000 companies have a Chief AI Officer or equivalent [10], and those are the organizations that can fund the seat without blinking. Below the Fortune 1000, the seat mostly doesn't exist at all. AI lands on the IT manager's desk as a side project, or on a committee that meets monthly and owns nothing. Then the pilot stalls, and the post-mortem blames the tool.

Why can't mid-market companies just hire their way out?

Because the market won't let them. ManpowerGroup's 2026 survey of 39,063 employers across 41 countries found AI skills leading the global ranking of hard-to-find skills, with 72% of employers reporting hiring difficulty overall [11]. And the size band that hurts most is upper mid-market: companies with 1,000 to 4,999 employees report the worst shortage of any size group at 75% [11].

Run the operator math. You're competing for the scarcest skill set in the world against enterprises that pay more, offer bigger platforms, and can absorb a bad hire. You probably can't evaluate the candidate, because nobody inside the building can tell a deployment operator from a confident strategist. And the search takes long enough that the technology shifts underneath you while the seat sits empty.

So the accountability gap persists even at companies that genuinely want to close it. The traditional answers are a full-time executive most mid-market P&Ls can't justify, an agency that owns your system on retainer, or a DIY effort led by whoever had the least ability to say no. I've written a separate piece comparing those paths honestly, because each one is right for somebody. But the data here explains why so many companies pick none of them and quietly stall instead.

What should mid-market operators do about it?

Six moves, each one aimed at a specific failure mode in the data above.

  1. Name one owner. A single person accountable for AI outcomes, with authority to change workflows, not a committee and not IT as a side duty. Grant Thornton's numbers say boards write checks without assigning ownership [9]. Do the opposite: no owner, no budget.
  2. Write the finish line in dollars before anything starts. Gartner's number one abandonment reason is unclear business value [7]. If a pilot can't state the metric it moves and what that's worth annually, it isn't a pilot, it's a hobby.
  3. Start in the back office. The budget flows to sales and marketing tools while the returns hide in operations [3]. Pick two or three paper-heavy, repeatable workflows where output is countable. Stanford's productivity data says that's exactly where the measured gains live [6].
  4. Buy and adapt before you build. Vendor tools and partnerships succeed at roughly twice the rate of internal builds, 67% versus about a third of that [3]. Your engineering pride is not worth the failure-rate difference.
  5. Put governance in writing while the footprint is small. Less than half of mid-market companies have enforced AI policies, and 42% already ate a security incident [2]. A two-page policy on data, access, and review beats a 40-page one that never ships.
  6. Work in 90-day increments. Deloitte's data says most experiments won't scale on their current trajectory [8]. Scope something deployable in a quarter, measure it against the dollar figure from move two, then decide again. Eighteen-month AI roadmaps are how mid-market companies fund stall.

None of this requires enterprise budgets. The Chamber data shows small businesses that get AI working tend to grow headcount, not cut it [5]. The technology is the same technology the enterprises use, running on your own enterprise AI plan, inside systems you own. What the mid-market is missing is the accountable seat, and that's a solvable problem.

This is the problem I work on every day at Meet Caddy: putting a working AI operation inside mid-market companies in 90 days, with the owner's name on the outcomes. If you want to talk through where your company sits against these numbers, book a free discovery call. And if you'd rather just run the six moves yourself, that's a fine outcome too. The point is that the stall is optional.

Sources

  1. US Census Bureau, "Large Firms With at Least 20 Employees Biggest AI Users," America Counts, May 2026. census.gov
  2. Netrio, "Netrio Survey Finds Mid-Market AI Adoption Is Widespread, but Readiness and Governance Gaps Remain," PR Newswire, June 2026. prnewswire.com
  3. Fortune, "MIT report: 95% of generative AI pilots at companies are failing," reporting MIT NANDA's "The GenAI Divide: State of AI in Business 2025," August 2025. fortune.com
  4. Federal Reserve Board, "Monitoring AI Adoption in the U.S. Economy," FEDS Notes, April 2026. federalreserve.gov
  5. US Chamber of Commerce, "Empowering Small Business: The Impact of Technology on U.S. Small Business," August 2025. uschamber.com
  6. Stanford HAI, "The 2026 AI Index Report," Chapter 4: Economy, 2026. hai.stanford.edu
  7. Gartner, "Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025," press release, July 2024. gartner.com
  8. Deloitte, "State of Generative AI in the Enterprise" quarterly report press release (survey of 2,773 director- to C-suite leaders across 14 countries, fielded July to September 2024), January 2025. deloitte.com
  9. Grant Thornton, "2026 AI Impact Survey" (950 business leaders across 10 industries, fielded February 23 to March 18, 2026), 2026. grantthornton.com
  10. National Law Review, "The State of AI: Key Insights from the 2026 Leadership Survey," reporting the 15th Annual AI & Data Leadership Executive Benchmark Survey of roughly 110 Fortune 1000 companies, January 2026. natlawreview.com
  11. ManpowerGroup, "Global Talent Shortage Reaches Turning Point as AI Skills Claim Top Spot" (survey of 39,063 employers in 41 countries, fielded October 2025), 2026. manpowergroup.com

Read next

AI executive, AI agency, or figure it out yourself: how mid-market companies should decide A fair decision guide for mid-market owners: when to hire an AI executive, when to use an agency, when to DIY, and the questions to ask before signing. The 21-metric framework: how to measure what AI is actually worth in your business 21 metrics, five buckets, and a worked example: how mid-market operators put a defensible dollar figure on AI before buying anything.

Get the next playbook

One useful piece on putting digital labor to work, straight to your inbox. No spam, unsubscribe anytime.

Book a Call