The McKinsey Gap: Your Employees Use AI More Than You Think, And That Is Your Problem

Your employees use AI more than you think. McKinsey measured this cleanly in 2025. This is not the feel-good story you were expecting. It is your biggest adoption problem.

Most Mittelstand CEOs we sit with in strategy workshops at Sentient Dynamics tell the same story: "Our workforce is struggling with AI. We are building an adoption roadmap, we are investing in training, we are bringing people along." The intent is honest. The premise is factually wrong. The workforce is not struggling. The workforce is ahead of you. And that is the problem you have to solve before you walk into the next "AI strategy deck" meeting.

What McKinsey actually measured in 2025

McKinsey published two big studies in 2025 that, taken together, paint a clear picture. The first is the annual "State of AI 2025" (November 2025, n=1,993 organizations across 105 countries). The second is "Superagency in the Workplace" (January 2025), a dedicated workplace survey that polled C-suite and employees in parallel. The combination produces the gap that is sitting on your desk.

The core wording from State of AI 2025, quoted as published: "88 percent report regular AI use in at least one business function, compared with 78 percent a year ago." So 88 percent of organizations use AI regularly in at least one business function, up from 78 percent a year earlier. And: "23 percent of respondents report their organizations are scaling an agentic AI system somewhere in their enterprises." So 23 percent are already scaling agentic AI in at least one business function. Not enterprise-wide, in at least one business function. The distinction matters when you cite the number to your board.

The real punch comes from the Superagency study. Original McKinsey wording: "C-suite executives estimate that only four percent of employees use generative AI for at least 30 percent of their daily work, when the number is closer to 13 percent." Plus the headline finding: "Employees are 3 times more likely than leaders expect to be using Gen AI for at least 30% of their daily work."

C-suite executives guess 4 percent. Employees report 13 percent. That is not slightly off, that is a factor of 3.

And that is not the only perception gap. Here is the synthesis we pull from Superagency, State of AI 2025, and the Bitkom AI Study 2026:

Dimension	C-suite / CEO view	Employee view
Share of staff using GenAI for >=30% of work	4% (McKinsey)	13% (McKinsey)
Pace of adoption next 3 years	"We will phase this in carefully"	"Should take over at least 30% of my work"
Training need	"Staff need foundations first"	"We already know what we are doing"
Main blocker	"Employee skepticism and change"	"Leadership hesitates, gives no tools"
Risk of Shadow AI	"Compliance, data loss"	"We use it anyway, it makes us faster"

In addition, two more well-known sources: Microsoft + LinkedIn 2024 Work Trend Index: "78% of AI users are bringing their own tools to work, Bring Your Own AI (BYOAI)" and "75% of global knowledge workers" already use GenAI, in small and medium-sized companies even 80 percent. Slack Workforce Index April 2025 layers on: "Between November of 2024 and April of 2025, AI usage rose from 36% to 60% of desk workers, with 42% relying on it regularly, at least weekly."

And in Germany? Bitkom AI Study 2026, representative of 604 companies with 20-plus employees: the share of companies actively using AI rose from 17 percent (prior year) to 41 percent, another 48 percent plan to use it, 81 percent rank AI as the most important technology. But: only 30 percent see themselves as leaders, 62 percent as laggards. That is not an employee problem, that is a leadership perception problem.

3 consequences for you as a Mittelstand CEO

If you accept the numbers above (and you should, because McKinsey, Microsoft, Slack and Bitkom measured them with consistent methodology), three things follow for you as CEO. None of them is the "we need more training" answer that most strategy decks open with.

Consequence 1: Your "AI strategy slide" is outdated before you show it

The typical Mittelstand AI strategy looks like this: phase 1 pilots, phase 2 train the workforce, phase 3 scale. The hidden assumption: leadership is ahead, the workforce catches up. That assumption is wrong.

What actually happens: when you present your "AI strategy 2026" today, half the room (your works council rep, your HR lead, your CIO) looks at the deck and thinks: "My intern built this in Claude last Wednesday." You lose credibility in the first minute. The workforce is not impatient because they do not understand AI. They are impatient because they use AI and leadership cannot keep up.

What you should do instead: no strategy deck without a clean baseline assessment. Ask the workforce anonymously: who uses what, how often, for which tasks? Only then do you have ground to build a strategy on. Otherwise you are designing a roadmap for a problem you do not have.

Consequence 2: Shadow AI is not the risk you think it is

If the term "Shadow AI" has come up in your last three audit meetings, it was probably framed as "compliance risk" and "data loss". Not wrong, but not the main story. Microsoft 2024 said it clearly: "78% of AI users are bringing their own tools to work, even more common in small and medium-sized companies (80 percent)." 80 percent in SMEs. That was 2024.

The real Shadow AI risk is not the GDPR fine (real, but manageable with a clean tool policy). The real risk is productivity asymmetry. You now have two classes of employees: the ones who quietly use AI and deliver 30 to 60 percent faster, and the ones who do not and fall behind in performance reviews. Those classes are not distributed along hierarchy lines, they are distributed along initiative and tech affinity. That breaks every compensation logic, every career-path assumption, every team composition you built up over the last decade.

What to do: legalize Shadow AI before you try to monitor it. Clear list of allowed tools (with enterprise license, data residency, audit log), clear list of prohibited tools (private accounts, unclear data flows, non-EU hosting for sensitive content). The frustration in your workforce disappears ("finally I can say openly what I do"), and you get the telemetry you need.

Consequence 3: Your training pyramid is built upside down

The standard Mittelstand training pyramid: at the bottom, all employees get an "AI basics course", in the middle, specific roles get use-case training, at the top, leadership gets a "strategy briefing". The McKinsey gap shows this pyramid is inverted.

The base does not need a basics course, it needs permission, tooling and a clear frame of what is already allowed. The 20 percent in the middle (power users in every team) need multiplier roles, with time budget and mandate. The leaders at the top need what they get least of: a practical hands-on workshop, because otherwise they cannot model adoption. If your CFO does not personally use AI, he cannot decide what will happen in his team in 2027.

We call this the workforce pyramid inversion, and we lay it out in detail in the workforce AI training pyramid post with the Bitkom 2026 stage distribution and a 30-60-90 plan. For now, the principle: do not plan the training pyramid the classic way, plan it on the actual baseline distribution. Otherwise you are training the wrong people on the wrong content at the wrong time.

What you must do NOW (3 immediate actions)

Not in six months. Not after the next strategy offsite. This week.

Action 1: McKinsey gap audit inside your leadership team. Block 90 minutes with your board or management team. Each person writes on paper: "What percentage of our employees use AI for at least 30 percent of their work?" Anonymous. Then reveal. You will see: estimates land between 2 and 15 percent, the McKinsey number is 13. The discussion about the spread alone shows you where the blind spots sit.

Action 2: Anonymous workforce survey in 48 hours. Three questions, no more. (1) Which AI tools do you use today for your work, professionally or privately? (2) For which tasks? (3) What is keeping you from using more? Microsoft Forms is fine. Anonymous. You do not need a consulting firm's evaluation, you need 60 percent response in 48 hours. That gives you the numbers for action 3.

Action 3: Tool policy on one page. Which tools are allowed (with enterprise license, EU data residency, audit log)? Which are off-limits (private accounts with company data, unclear data flows, US hosting for sensitive content)? Which data can go in, which not? One page. Distributed to everyone. That takes the pressure off the workforce that has been building up for twelve months.

These three actions cost you no software license and no external consultant. They cost leadership time and the courage for an honest diagnosis. The data is on your desk in fourteen days.

How the McKinsey gap will evolve in 2026

The gap is widening, not closing. Three reasons.

First, the tools get better and more invisible. ChatGPT in 2024 still ran in its own browser tab. In 2026, Claude sits in your developer's IDE, Microsoft Copilot in Outlook and Teams, Perplexity in the browser sidebar. Usage becomes more measurable for IT and at the same time more naturalized for the employee. They will not ask whether they may use AI, they will just use it.

Second, agentic workflows graduate from pilot. State of AI 2025 says 23 percent are already scaling agentic AI in at least one business function. In twelve months that is 40 to 50 percent. The difference to chat usage: agentic systems take over workflows, not just answers. Your employees will automate workflows you as leadership never saw, because they ran below your radar.

Third, the EU AI Act shifts the conversation. From August 2026 onward, you as an employer must demonstrate that your staff have AI literacy (Art. 4). That will not reduce AI usage, it will force you as CEO to document what they are doing. If you do not know today what your workforce does, you will not know in a year what to document.

So the McKinsey gap is not a transient finding. It is the default state for the next two to three years. The question is not whether you close it, but whether you steer it actively or suffer it passively.

FAQ

Is the McKinsey gap specific to US firms or does it apply to the DACH Mittelstand?

The Superagency study measured globally, with European data points. The Bitkom AI Study 2026 confirms the pattern for Germany: 41 percent of companies use AI actively, but only 30 percent see themselves as leaders. If 81 percent name AI as the most important technology yet only 30 percent feel ahead, that is the same perception gap. In Mittelstand, on top of that, decentralized structures slow official adoption while staff experiment faster. So if anything, larger, not smaller.

Should I not clear compliance risks first before legalizing Shadow AI?

Wrong sequence. As long as you do not know what is happening, you cannot evaluate the risks. A 48-hour survey gives you telemetry. A one-page tool policy takes the most harmful practices off the table (private accounts with customer data, US hosting for sensitive content). Proper vendor selection comes next. Starting the other way around gives you six months of debate and zero telemetry.

What if my workforce says "we do not use AI at all"?

Then you either have a very honest industry (rare) or a trust problem. The Microsoft figure of 80 percent BYOAI in SMEs is methodologically clean. If your survey says "0 to 5 percent", your staff is afraid to say so. Tool policy first, then survey again. Otherwise you are measuring something different from what you wanted to measure.

How do I tell Shadow AI apart from a legitimate power user?

You do not tell them apart by behavior, you tell them apart by tools. If the power user runs inside your enterprise ChatGPT license and the shadow user runs in a private account, the difference is licensing context. Move both onto enterprise tooling, then you have power users and no shadow users. That is the most important consolidation for 2026.

Sources and next step

The numbers above come from:

• McKinsey, "The state of AI in 2025: Agents, innovation, and transformation", November 2025, n=1,993 across 105 countries
• McKinsey, "Superagency in the workplace: Empowering people to unlock AI's full potential at work", January 2025
• Microsoft + LinkedIn, "2024 Work Trend Index Annual Report", May 2024
• Slack Workforce Index, April 2025 (Salesforce press release, Daily AI Workforce Use Growth)
• Bitkom, "AI Study 2026: Artificial Intelligence in Germany", February 2026, n=604

If you want to go deeper on the Mittelstand pattern, read in parallel:

• AI Leaders vs Laggards: why the margin gap is 47 percent
• Bitkom 2026: 89 percent of German companies engage with AI
• The 5 leadership beliefs blocking your AI adoption
• What is Agentic AI: executive crash course
• AI pilot graveyard: why pilots never reach production
• AI Literacy Mandate August 2026: executive checklist
• AI Maturity Check in 15 minutes
• AI Talent Crisis: why your Berlin senior was just bought by Tesla

We run a McKinsey gap stress test with your leadership team. 1 day, all the relevant CEO/employee perception gaps on the table, with Bitkom and McKinsey benchmarks plus a 90-day plan. Book a slot.