Back to Guides
Enterprise Software 15 min read

The State of AI enablement in 2026

The State of AI enablement in 2026

The state of AI enablement in 2026 is easy to state and hard to accept: access grew, skill did not, and the people who authorize the spend have run out of patience with the gap. Sanctioned tool access widened by half in a year, while daily use among the people holding it barely moved. Every large survey published this year names worker skill, not technology, as the thing in the way. What follows is the data, where it comes from, and what it means if you run a company of 50 to 500 people rather than a Fortune 500 one.

The numbers that define the year

Four published figures carry most of the weight. They measure different things, which matters more than any one of them.

What it measures 2026 figure Source and date
US firms reporting AI use in any business function 19.8 percent nationally; 37 percent at 250-plus employees, 32 percent at 100 to 249 Census Bureau BTOS, as of 3 May 2026
Workers equipped with sanctioned AI tools Around 60 percent, up from fewer than 40 percent a year earlier Deloitte, 3,235 leaders in 24 countries, fielded Aug–Sep 2025
Employees using AI in their role 15 percent daily, 30 percent weekly or more, 52 percent a few times a year or more Gallup, May 2026 update
Employees whose employer communicated a clear AI plan 25 percent Gallup, same update

Read the first and last rows together. One company in five reports using AI at all. One employee in four has been told the plan. Between those facts sits the reason “AI enablement” started appearing in budget lines this year.

The Census figures deserve a second look, because they are the only numbers here that are nationally representative, non-commercial, and broken out by company size. Use rose over the period among firms with 20 or more employees and changed little below that. If you sit in the 50 to 500 band, your peer group is roughly a third adopted, not nine-tenths, whatever the vendor deck says.

One caution on trend reading. The Census revised the question in November 2025, from AI used “in producing goods or services” to AI used “in any business function.” Comparisons across that date are not comparing like with like.

Why the adoption figures disagree with each other

Pages covering this market tend to pick whichever number flatters the argument. The figures are not in conflict. They count different units.

Census counts firms. One response per business, asking whether the business uses AI anywhere. A 400-person company where the marketing team uses ChatGPT counts as a yes.

Deloitte counts access. Its 3,235 senior leaders report what share of their workforce has approved tools. Access is a buying fact, not a behavior.

Gallup counts frequency. It asks employees how often they use AI for their own job, which makes it the only one of the three that measures what a person does.

Stack them and the shape of the year appears. Most firms have not started. Among those that have, most workers now hold a license. Among those workers, fewer than 60 percent use the tool in daily work, which Deloitte reports as largely unchanged from the year before. Access went up. Habit did not.

This is also why internal dashboards mislead. A console that counts licensed users who opened a feature once in a 30-day window reports a healthy number for a team that has learned nothing, the mechanism behind license-usage dashboards overstating real capability.

The constraint leaders named themselves

The most useful finding of 2026 is not a gap statistic. It is that senior leaders, asked to name the biggest barrier to putting AI into existing workflows, named weak worker skills.

That comes from Deloitte’s own reporting of its 3,235-leader survey. The two most common responses were teaching the broader workforce to raise AI fluency, at 53 percent, and designing upskilling strategies, at 48 percent. Far fewer firms reported redrawing roles, workflows, and career paths.

Two things follow. The enablement argument no longer needs making to the people who sign budgets; they made it themselves. And the stated answer is heavy on teaching and light on work redesign, which is the half that makes teaching stick. A company that trains without changing what anyone is expected to produce on Monday has bought a pleasant afternoon.

For the shortest statement of what the word covers and what it leaves out, the definition and its edges hold up better than the vendor glossaries.

What the 95 percent statistic rests on

One number has done more work in this category than any other: that 95 percent of generative AI pilots produce no measurable business return. It appears in Fortune, Forbes, a dozen trade outlets, and most vendor decks written since late 2025. It comes from MIT’s Project NANDA research. Its basis is 52 structured interviews, 153 senior leaders surveyed at four industry conferences, and a review of more than 300 public AI initiatives, collected between January and June 2025.

That is sound qualitative work and a weak footing for a precise claim about a whole market. Treat 95 percent as a direction, not a measurement. Anyone quoting it without the sample size either has not read it or is counting on you not having.

The same study carries a finding quoted far less and worth more here: employees run what it calls a shadow AI economy, using consumer tools without approval. That is not a technology failure. It is what happens when people need the skill faster than the company supplies it, and it is the best argument for approved tools plus real training over a longer policy document.

What the AI enablement market is selling in 2026

Four distinct offers now go to market as AI enablement services, one label over four different price tags.

  • Platform work. Data readiness, integration layers, search. Priced like an IT program, sold by systems integrators.
  • Operating-model consulting. Strategy, governance, centers of excellence. Sold by the large firms and the mid-tier ones now ranking for the term.
  • Course access at scale. Seat-based libraries from the learning platforms. Cheap per head, measured in completions.
  • Facilitated capability work. Hands-on sessions on a team’s own documents, with a measurement before and after. The smallest of the four by vendor count.

The fourth is where the 2026 evidence points, and where published data is thinnest, because the firms doing it are small. Correlation One’s State of AI Enablement 2026 puts the ratio at two to three times more AI licenses than trained active users across its client base, with monthly active use often settling between 25 and 50 percent and concentrated among enthusiasts.

Hold onto that ratio with a caveat attached. The report gives no sample size or method, its clients are unnamed, and its examples run from 1,400 to 16,000 seats. The direction matches Deloitte and Gallup; the precision cannot be checked, and nothing in it was written about a 200-person company.

A word on agents, since every 2026 forecast leads with them. At this size they change the training question less than it seems: a team that cannot brief a model, check its output, and judge which tasks to hand over will not supervise an agent well either.

What a program costs against what you already spend

Nobody ranking for this term publishes a number, so these are market ranges rather than a quote. Corporate AI enablement is also priced by component rather than as one product. Facilitated LLM fluency work runs roughly $2,000 to $15,000 depending on headcount and session count. Custom automation projects, a separate purchase after skill exists, run roughly $25,000 to $150,000.

Set that against what a company of this size already spends on training. Training magazine’s 2025 Training Industry Report puts average spend at $874 per learner, with midsize companies at $782 and an average midsize training budget near $1.6 million. Training hours per employee came in at 40, down from 47 the year before.

The arithmetic is awkward for anyone calling enablement unaffordable. A program in the middle of the workshop range is a rounding error against that budget, and smaller than the license line it exists to justify. The real block is that the AI line sits in software and the skills work sits in L&D, so neither owner holds both halves of the argument. Which is why the easiest moment to fund it is the six weeks before a renewal, the whole case for funding enablement out of the renewal itself.

The regulatory year: one deadline moved, the training duty did not

2026 produced one real regulatory change and much confused commentary about it.

The EU’s Digital Omnibus on AI deferred the high-risk duties under the AI Act. Standalone high-risk systems under Annex III move to 2 December 2027, and AI built into products covered by EU sectoral law under Annex I moves to 2 August 2028, per the published timeline. Transparency duties under Article 50 held at 2 August 2026, and content watermarking applies from 2 December 2026.

The training-relevant part did not move. Article 4 of the AI Act, which requires providers and deployers to take measures supporting AI literacy among staff and others operating AI systems on their behalf, has applied since 2 February 2025. It names no curriculum and promises no individual standard. For a US company with EU staff or EU customers, it turns documented training effort from a good idea into a duty, and the high-risk deferral changed nothing about that.

There is still no US federal training requirement, and state activity keeps shifting. Neither is a reason to run a program, but both make the paperwork worth keeping.

Who owns AI enablement in 2026

The clearest market signal this year is not a statistic. It is a job title. AI enablement lead roles are now posted by universities, industrials, and investment firms, and the postings describe change management: find the workflows worth changing, build the training, report what improved.

Companies of 50 to 500 people will not hire that role, and mostly should not. The work lands on a COO, a VP of Operations, a Chief of Staff, a Head of L&D, or a CFO who got tired of the license line. It is a second job on top of a first, and the failure mode is reliable: ownership with no protected calendar time produces a month four that looks like month one.

Two names and an honest hours estimate beat a steering committee. Where a company sits on that question is most of what the maturity levels measure. The sequence from rollout to a reportable number is set out in the program playbook.

The year amounts to this. The case for buying AI access is settled, and the case for building skill on top of it is now made by the buyers themselves, the argument behind why buying licenses is not the same as building. What remains unsettled is whether anyone will measure it.

For a read on where your team sits against these numbers, we run a free AI-readiness call that works through your license data, your current usage, and what a defensible first program looks like at your headcount.

Frequently asked questions

What is the state of AI enablement in 2026?

Access expanded and capability did not. Deloitte reports that the share of workers with sanctioned AI tools rose from fewer than 40 percent to around 60 percent in a year, while daily use among them stayed below 60 percent. Leaders in the same survey named weak worker skills as the top barrier to putting AI into real workflows.

What percentage of companies actually use AI in 2026?

By the Census Bureau’s Business Trends and Outlook Survey, 19.8 percent of US firms reported using AI in any business function as of 3 May 2026. Adoption rises with size: 37 percent at firms with at least 250 employees, 32 percent at 100 to 249. Another 20 to 23 percent expected to adopt within six months.

Why do the AI adoption numbers disagree with each other?

They count different things. Census counts firms that use AI anywhere. Deloitte counts the share of workers with approved tools. Gallup counts how often employees use AI for their own job. A company can be a yes for Census, give 80 percent of staff licenses for Deloitte, and still show 15 percent daily users by Gallup’s measure.

Is it true that 95 percent of AI projects fail?

The figure comes from MIT’s Project NANDA research and rests on 52 structured interviews, 153 surveyed senior leaders, and a review of more than 300 public AI initiatives, collected in the first half of 2025. That supports a direction, not a market-wide failure rate. Read it as a warning about unmeasured pilots, and be wary of anyone quoting it without the sample.

What is the AI activation gap?

The distance between licenses bought and people who can use them well. Correlation One’s 2026 report puts it at two to three times more licenses than trained active users across its client base, with monthly active use often settling between 25 and 50 percent. The report gives no sample size, so treat the ratio as directional.

How much does AI enablement cost in 2026?

Facilitated LLM fluency work runs roughly $2,000 to $15,000 in the market, depending on headcount and session count. Custom automation projects are a separate purchase after skill exists and run roughly $25,000 to $150,000. For scale, midsize US companies spend about $782 per learner per year on training overall.

Who owns AI enablement when nobody’s title says AI?

At 50 to 500 people, whoever authorizes the license spend, with a named day-to-day owner under them who has protected hours. Enablement lead roles now exist at large employers, but hiring one at this size is rarely the right first move. Both names and an hours estimate belong in the plan.

Did AI regulation change in 2026 in a way that affects training?

The EU’s Digital Omnibus on AI deferred high-risk duties to 2 December 2027 for standalone systems and 2 August 2028 for AI built into regulated products. The AI literacy duty under Article 4 was not deferred and has applied since 2 February 2025. If you have EU staff or customers, documented training effort is still owed.

Do AI agents change what we should train people on?

Not at this size, not yet. Agents raise the value of the same core skills: briefing a task precisely, checking output, and judging which work to hand over at all. A team that cannot do those three things with a chat tool will not supervise an agent well.

What should a mid-market company do first in 2026?

Measure before buying anything else. A scored baseline on real tasks, taken from the people who already hold licenses, tells you whether the problem is access, skill, or workflow. Most find it is the second, and the number gives you something to compare against eight to twelve weeks later.

Key takeaways

  • Access widened and habit did not. Sanctioned tool access rose from fewer than 40 percent to around 60 percent of workers in a year, while daily use among those with access stayed below 60 percent and barely moved.
  • Census data, the only nationally representative source in this set, puts US firm-level AI use at 19.8 percent as of May 2026, with 37 percent above 250 employees and 32 percent at 100 to 249. Your peer group is about a third adopted, not nine-tenths.
  • The headline figures only look contradictory because they count firms, access, and frequency. Ask what the unit is before quoting any of them.
  • Senior leaders now name worker skill as the top barrier themselves, and their stated answer leans on teaching while skipping work redesign, which is the half that makes training hold.
  • The 95 percent failure statistic rests on 52 interviews, 153 survey responses, and a review of 300-plus public initiatives from the first half of 2025. Direction, not measurement.
  • Money is rarely the obstacle. Workshop work at market rates is a rounding error against a midsize training budget. Ownership with protected calendar time is the scarce input.

Last Updated: Nov 17, 2026

AW

Arthur Wandzel

SFAI Labs helps companies build AI-powered products that work. We focus on practical solutions, not hype.

Make your team fluent in AI — then automate what proves out

  • Hands-on training applied to your own documents and workflows
  • Often reimbursable through state small-business training programs
  • Built for 5–50 person firms with no IT department

Related articles