The Real State of AI Adoption in Companies: What the Research Actually Shows in 2026

If you spend any time on LinkedIn or in the business press, you have seen the various headlines about AI’s impact on companies and jobs.

The tech industry has cut more than 123,000 jobs in the first half of 2026, with AI being the most–cited reason. Oracle attributed 21,000 layoffs to AI in an official SEC filing. CNBC warns of an "AI labor crisis".

And yet, in the same news cycle, you may have read that 88% of companies now use AI, that 95% of corporate AI pilots fail, and that official statistics count barely one in five businesses using AI at all. Meanwhile, researchers at Yale find no visible AI disruption in the labour market.

So who is right? Surprisingly, almost everyone.

For this month’s article, I did something different. Instead of looking at one academic paper, I gathered some of the most-cited studies on AI adoption from the past twelve months and compared them. As someone who teaches digital marketing strategy, I kept running into the same question from students and clients: how can all these numbers be true at the same time?

There is a simple answer, and it changes how you should read every AI headline from now on: these studies do not measure the same reality. And most media—whether social or legacy—tend to misrepresent the findings to generate hype.

Why 88% and 20% Are Both True

Let’s start with the most basic question: how many companies actually use AI?

McKinsey’s State of AI survey (2025) reports that 88% of organisations use AI in at least one business function. Stanford’s AI Index Report (2026) arrives at the same figure.

On the other hand, official statistics tell a different story. The US Census Bureau, which surveys a representative sample of roughly 1.2 million businesses, finds that about 20% of firms used AI in early 2026. Eurostat reports exactly the same number for EU enterprises: 20.0%, up from 13.5% a year earlier.

How can both be true? The answer lies not in the data, but in who gets asked.

McKinsey surveys the kind of organisations that answer McKinsey surveys: mostly large, international, and already invested in technology (38% of respondents work for companies with more than $1 billion in annual revenues).

Whereas the census counts every business, including the bakery, the fiduciary, and the two-person consultancy. It is worth noting that when Census Bureau researchers weight the same data by employment, the U.S. figure rises to 32%, because larger firms are more likely to adopt (Bonney et al., 2026, The Microstructure of AI Diffusion).

The consequence is simple: there is no contradiction here, only different populations (samples). Both surveys are honest. The headlines that quote them rarely are.

The Adoption Funnel: Use, Scale, Impact

The second distinction is what we mean by “adoption”, and it explains the failure headlines.

Using AI is not the same as being transformed by it. Think of it like gym memberships in January: almost everyone has one. Far fewer people train every week. Hardly anyone does a Hyrox.

The research used for this article maps neatly onto this adoption funnel with three stages: use, scale, and impact.

1. Use: 88% of organisations use AI somewhere (McKinsey, 2025).
2. Scale: fewer than 10% have fully scaled AI in any single business function (Stanford AI Index, 2026), and nearly two-thirds have not begun scaling at all.
3. Impact: only about 6% report a meaningful effect on enterprise-level earnings (McKinsey, 2025).

A note on sources here: the scale and impact stages rely on McKinsey and Stanford, because official statistics do not measure profit-and-loss impact. But where government data can look inside firms, it confirms the funnel’s shape. The Census Bureau finds that 57% of AI-using firms deploy it in three or fewer business functions, and 65% limit worker use to three or fewer tasks (Bonney et al., 2026). Adoption is broad. Depth is rare, in every dataset that measures it.

Now place these numbers next to the investment: global corporate AI spending reached 581 billion dollars in 2025, more than double the year before. The gap between what companies spend and what reaches the bottom line is huge.

This is also where the most viral AI statistic of the past year belongs. In August 2025, an MIT-affiliated report claimed that 95% of generative AI pilots fail to deliver measurable returns. This number was widely shared on social media. But what was not (or less) shared was the methodology used by the researchers: a small sample, a narrow definition of failure (no profit-and-loss impact within months of a pilot), and a focus on custom enterprise projects rather than everyday AI use. Some analysts took the study apart within weeks. The finding points in a real direction, but the number itself measured something much smaller than the headline implied.

More robust survey data confirms the direction without the drama. S&P Global’s Voice of the Enterprise survey found that 42% of companies abandoned most of their AI initiatives in 2025, up from 17% a year earlier, citing cost, data privacy, and security as the main obstacles. That sounds alarming. My reading, however, is that the number also reflects something healthier: companies are running far more experiments and killing the bad ones faster. Analysts at BCG and KPMG made a similar point when the data was released, arguing that surfacing and stopping failed pilots is part of a culture of experimentation. Read this way, rising abandonment is partly a sign of maturity, not just failure.

Is AI Taking Jobs? It Depends Where You Look

Here, the research offers a distinction I find more useful than most statistics: the difference between augmentation (AI helps a person do their work) and automation (AI does the work instead).

Brynjolfsson, Chandar and Chen (2025) analysed payroll records covering millions of US workers and found that employment for workers aged 22 to 25 in AI-exposed occupations has seen a relative decline of 13% since late 2022, while employment for older workers in the same roles held steady. The declines concentrate precisely where AI automates tasks rather than augments them. Entry-level jobs, it turns out, look a lot like the tasks AI does well.

If we zoom out, however, the picture is less dramatic, and this is where the second half of our distinction comes in. Most companies today deploy AI to augment rather than to automate: the US Census Bureau finds that 66% of AI-using firms rely on it solely to support their workers' existing tasks (Bonney et al., 2026).

And augmentation, so far, changes work without destroying it. The Budget Lab at Yale tracks the overall labour market monthly and finds stability, not disruption: no clear relationship between AI exposure and unemployment. In the most rigorous study of the augmentation side to date, Humlum and Vestergaard (2025) linked surveys of 25,000 Danish workers to employer records and found that AI chatbots, tools that help workers do their jobs rather than do the jobs for them, saved about 3% of working time, with no measurable effect on earnings or hours. Not 30%. Three.

In other words: where AI automates, entry-level jobs are quietly disappearing. Where AI augments, which is most places, jobs are quietly getting a little easier. Two different mechanisms, two different headlines, one technology.

So what explains the 123,000 job cuts? Partly the real, localised effects on entry-level roles. But researchers point to something else as well. Gartner, after tracking 1.4 million layoffs in 2025, estimates that less than 1% were directly tied to AI productivity gains, and Yale researchers have raised the question of "AI-washing": for a company correcting years of over-hiring, "we are restructuring for AI" simply sounds better to investors than "we hired too many people." Many of those cuts came not from AI, but from balance sheets.

Why This Matters for Marketers and SMB Leaders

Understanding these headlines is important for business leaders because you make real decisions based on these narratives and how you compare to your competitors or sector: how much budget you should allocate to AI next year, which tools to buy, whether to hire that junior marketer, etc.

Overestimate the revolution, and you chase enterprise-grade transformation projects that even large corporations abandon at a rate of 42%. Underestimate it, and you dismiss a technology that is measurably reshaping entry-level work and quietly compounding small efficiencies.

The next time a study lands in your feed, three questions will tell you more than the headline:

1. What did it measure (use, scale, or impact – automation vs. augmentation)?
2. Who was sampled (McKinsey clients or every business in your country)?
3. Over what timeframe?

The MIT study was not wrong. The way it was quoted was. A finding about custom enterprise pilots, judged on profit-and-loss impact within only a few months, was turned into a verdict on all corporate AI: a narrow definition of failure, quietly generalised to everyone. Watch for these signals whenever a number looks too quotable to be true.

What Adoption Looks Like When It Works

So what does the realistic path look like, especially for smaller organisations?

Swiss data offers a grounded picture. According to AXA’s annual labour market study, the share of Swiss SMEs using AI rose from 22% to 34% in a single year. The top applications are not moonshots: translation (52%) and correspondence (47%). Unglamorous tasks, repeated daily, where small time savings compound.

The OECD’s 2026 D4SME survey adds a useful maturity scale: among SMEs that use AI, 76% are AI novices using basic tools for isolated tasks, followed by optimisers (15.3%) and explorers (5%), while only 3.6% qualify as AI champions with deeply integrated systems (Figure 7 in the report). If your company is in the novice category, you are not behind. You are the norm.

What about the minority that does capture real value? The studies converge on three behaviours:

1. they buy specialised tools rather than building their own (purchased solutions succeed roughly three times as often as internal builds),
2. they redesign workflows instead of adding AI on top of old processes,
3. and they let the people who do the work drive adoption rather than a central innovation team.

Actionable Takeaways

So, what do you do with all of this on Monday morning? You cannot control the headlines, but you can control how your company reads them and acts on them. Here are three moves, all within reach of a small team and a normal budget:

1. Benchmark against official statistics, not vendor headlines. If your business uses AI productively in two or three functions, you are ahead of roughly 80% of companies in Europe and the US, whatever the conference keynotes or the 20-something AI consultants on LinkedIn suggest.
2. Climb one rung, not the whole ladder. The evidence favours augmentation in narrow, repeated tasks (drafting, translation, analysis, customer requests) over transformation projects. Pick the task your team repeats most often and improve that first.
3. Make quitting cheap. The 42% abandonment rate hides a lesson: successful adopters treat pilots as experiments with predefined kill criteria. Decide before you start what success looks like in 90 days, and walk away without regret if you do not see it. This reminds me of a great post from Kevin Kuhn at Gopf: Build an AI Lab instead of AI Lighthouse.

A Clearer Picture, If a Less Quotable One

Despite the contradictory headlines, the current state of AI adoption is surprisingly coherent. Adoption is real and accelerating on every continent. Transformation remains rare, even among the corporations spending the most. Job effects are real but concentrated, and smaller than the layoff announcements suggest. The technology is neither the revolution its vendors promise nor the disappointment its critics enjoy.

That is good news. You do not need to be in the 3.6% of AI champions to benefit. You need a clear view of where you stand, one well-chosen task, and the discipline to measure what actually changes. That is not just achievable for an SMB. It is exactly where SMBs have the advantage: shorter distances between the person who buys the tool and the person who uses it.

If you like this kind of evidence-based insight, subscribe to my newsletter, where I break down the latest marketing research and translate it into practical, actionable frameworks.

I’m curious how this matches your experience: Where is your company on the novice-to-champion ladder? And have you seen AI genuinely change a workflow in your business, or mostly add another tool to the stack? Share your perspective in the comments.

Thanks for reading

Marc

Sources

AXA (2025) AI gains ground among Swiss SMEs. Bern: State Secretariat for Economic Affairs SECO. Available at: https://www.kmu.admin.ch/kmu/en/home/new/news/2025/ai-gains-ground-swiss-smes.html (Accessed: 4 July 2026).

Bonney, K., Breaux, C., Dinlersoz, E., Foster, L., Haltiwanger, J. and Pande, A. (2026) The microstructure of AI diffusion: evidence from firms, business functions, and worker tasks. CES Working Paper 26-25. Washington, DC: US Census Bureau. Available at: https://www.census.gov/library/working-papers/2026/adrm/CES-WP-26-25.html (Accessed: 4 July 2026).

Brynjolfsson, E., Chandar, B. and Chen, R. (2025) Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence. Stanford, CA: Stanford Digital Economy Lab. Available at: https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/ (Accessed: 4 July 2026).

Challapally, A., Pease, C. and Raskar, R. (2025) The GenAI divide: state of AI in business 2025. Cambridge, MA: MIT NANDA. (Original report no longer hosted online; for coverage see Fortune, 2025.)

CNBC (2026) '20,000 job cuts at Meta, Microsoft raise concern that AI-driven labor crisis is here', 24 April. Available at: https://www.cnbc.com/2026/04/24/20k-job-cuts-at-meta-microsoft-raise-concern-of-ai-labor-crisis-.html (Accessed: 4 July 2026).

Eurostat (2025) Use of artificial intelligence in enterprises. Statistics Explained. Luxembourg: Eurostat. Available at: https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises (Accessed: 4 July 2026).

Fortune (2025) 'MIT report: 95% of generative AI pilots at companies are failing', 18 August. Available at: https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/ (Accessed: 4 July 2026).

Fortune (2026) 'If AI is roiling the job market, the data isn't showing it, Yale Budget Lab report says', 2 February. Available at: https://fortune.com/2026/02/02/ai-labor-market-yale-budget-lab-ai-washing/ (Accessed: 4 July 2026).

Gartner (2026) Gartner says autonomous business and AI layoffs may create budget room, but do not deliver returns. Press release, 5 May. Stamford, CT: Gartner. Available at: https://www.gartner.com/en/newsroom/press-releases/2026-05-05-gartner-says-autonomous-business-and-artificial-intelligence-layoffs-may-create-budget-room-but-do-not-deliver-returns (Accessed: 4 July 2026).

Humlum, A. and Vestergaard, E. (2025) Large language models, small labor market effects. BFI Working Paper 2025-56. Chicago, IL: Becker Friedman Institute, University of Chicago. Available at: https://bfi.uchicago.edu/working-papers/large-language-models-small-labor-market-effects/ (Accessed: 4 July 2026).

Marketing AI Institute (2025) 'That viral MIT study claiming 95% of AI pilots fail? Don't believe the hype'. Available at: https://www.marketingaiinstitute.com/blog/mit-study-ai-pilots (Accessed: 4 July 2026).

McKinsey & Company (2025) The state of AI in 2025: agents, innovation, and transformation. Available at: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai (Accessed: 4 July 2026).

OECD (2026) Empowering SMEs in the age of AI: the 2026 OECD D4SME survey. Paris: OECD Publishing. Available at: https://www.oecd.org/en/publications/empowering-smes-in-the-age-of-ai_bf5a9816-en.html (Accessed: 4 July 2026).

Roeloffs, M. (2026) 'AI cost 21,000 jobs at Oracle this year, and more layoffs could be coming', Forbes, 4 June. Available at: https://www.forbes.com/sites/maryroeloffs/2026/06/04/tech-industry-loses-123000-jobs-this-year-ai-is-the-most-cited-reason-for-layoffs/ (Accessed: 4 July 2026).

Stanford HAI (2026) The 2026 AI index report. Stanford, CA: Stanford Institute for Human-Centered Artificial Intelligence. Available at: https://hai.stanford.edu/ai-index/2026-ai-index-report (Accessed: 4 July 2026).

The Budget Lab at Yale (2026) Evaluating the impact of AI on the labor market: November/December CPS update. New Haven, CT: Yale University. Available at: https://budgetlab.yale.edu/research/evaluating-impact-ai-labor-market-novemberdecember-cps-update (Accessed: 4 July 2026).

US Census Bureau (2026) 'AI use at U.S. businesses', America Counts, May. Available at: https://www.census.gov/library/stories/2026/05/ai-use-businesses.html (Accessed: 4 July 2026).

Wilkinson, L. (2025) 'AI project failure rates are on the rise: report', CIO Dive, 14 March. Available at: https://www.ciodive.com/news/AI-project-fail-data-SPGlobal/742590/ (Accessed: 4 July 2026).