Report 02 · AI enablement
The AI-Ready Workforce
Almost every company is buying AI and almost none are ready to use it, so this report maps the gap between investment and readiness and lays out a 90-day plan to close it with the people you already employ.
The short version
Everyone is buying AI. Almost no one is ready to use it.The AI conversation in most boardrooms is about tools. The data says the constraint is people, and more precisely leadership’s willingness to train them. McKinsey finds 92% of companies planning to increase AI investment, while 1% describe themselves as mature in how they use it. Meanwhile employees are already three times further ahead than their leaders estimate.
Your workforce is not waiting for permission. The question is whether their experimentation compounds into capability, or stays hidden in browser tabs.
Evidence: 1 McKinsey & Company (2025)
Companies planning to increase AI investment over the next three years.
1 McKinsey & Company, 2025Leaders who say AI is fully integrated into how their organization works.
1 McKinsey & Company, 2025Employees are three times more likely to be running 30% or more of their work on AI than leaders think.
1 McKinsey & Company, 2025Employees ranking training the number one factor for adoption. About half report getting little or none.
1 McKinsey & Company, 2025What is inside
The question is no longer whether to allow AI. It is whether daily experimentation compounds into capability, or stays hidden in browser tabs.
The adoption illusion
Everyone uses it, few gain from it.On paper, adoption is done: 88% of organizations now use AI regularly somewhere in the business. But only about 6% report significant enterprise-wide impact from it. The rest sit in what McKinsey’s researchers call pilot purgatory, experiments that never graduate into how work actually gets done.
Evidence: 2 McKinsey & Company (2025)
Underneath that sits a perception gap that should worry every HR leader. In Writer’s 2025 enterprise survey, 75% of executives said their organization had successfully adopted AI. Only 45% of their employees agreed. Gallup finds roughly seven in ten US employees never use AI at work in any sanctioned way, and of those who do, more than half hide it from their employer. Shadow adoption means the change is happening anyway, just without guardrails, quality control, or shared learning.
Evidence: 3 Writer (2025) / 4 Gallup (2024) / 5 Azumo (2025)
Source: Writer, Enterprise AI Adoption Survey 2025
Evidence: 3 Writer (2025)
Source: BCG, AI at Work 2025, several-times-a-week use
Evidence: 7 Boston Consulting Group (2025)
Across three separate studies, the bottleneck is not employee resistance. McKinsey’s verdict is that employees are ready and leadership is the barrier. The fastest wins are permission, guardrails, and training, in that order.
Evidence: 1 McKinsey & Company (2025)
The bottleneck is not employee resistance. Grant permission, set guardrails, then train.
The training gap is the readiness gap
Five hours of training changes the curve.Ask employees what would make them use AI well and the top answer is not better models. It is training, ranked the single most important adoption factor by 48%, while roughly half report receiving minimal or none. Perceptyx puts structured AI training at just 35% of organizations and finds over 60% of employees lack confidence using AI in their daily work.
Evidence: 1 McKinsey & Company (2025) / 6 Perceptyx (2025)
The dose matters less than you would fear. BCG’s global survey finds regular usage sharply higher among employees who received at least five hours of training, with in-person coaching on top. Five hours. The gap between sending a memo about Copilot and building an AI-capable workforce is measured in an afternoon per person, if the afternoon is designed well and reinforced daily afterwards.
Evidence: 7 Boston Consulting Group (2025)
Five hours of training separates users from spectators. Most organizations still have not scheduled hour one.
Evidence: 7 Boston Consulting Group (2025)
One more BCG finding sharpens the urgency: when employees do not get the AI tools they need, more than half say they will find their own and use them anyway. The choice was never AI or no AI. It is sanctioned capability or shadow usage.
Evidence: 7 Boston Consulting Group (2025)
The choice was never AI or no AI. It is sanctioned capability versus shadow usage.
In the EU, readiness is now a legal duty
Article 4 has applied since February 2025.If your organization operates in the EU, or has staff or contractors using AI on its behalf there, AI literacy stopped being optional on 2 February 2025. Article 4 of the EU AI Act requires every provider and deployer of AI systems to ensure a sufficient level of AI literacy among staff and anyone operating AI on their behalf. There is no headcount carve-out and no risk-tier exemption: a 60-person firm whose marketers use ChatGPT is a deployer.
Evidence: 8 European Union (2025)
| Date | What it means for you |
|---|---|
| 2 Feb 2025 | The duty applied. If staff use AI tools at work and there is no literacy program, the organization is already out of step with the obligation. |
| Aug 2025 | Civil-liability exposure sharpened. Harm caused by inadequately trained staff using AI is hard to defend without a documented program. |
| Aug 2026 | Enforcement machinery switches on. National market-surveillance authorities begin supervising Article 4, and penalties are set per member state. |
Evidence: 8 European Union (2025)
The Commission’s own guidance sets a floor that maps almost exactly onto this report: a single onboarding video is not sufficient; training must be tailored to role, technical level and context; and everything, meaning who was trained, on what, and when, must be documented so you can evidence it. A November 2025 Digital Omnibus proposal may soften how the duty is enforced, but human-oversight training for higher-risk uses stays either way, and the liability logic does not move.
Evidence: 8 European Union (2025)
The same role-tailored, documented, continuously refreshed literacy program that Article 4 asks for is the one the productivity data asks for. Compliance and capability are, for once, the same purchase. The only question is whether you buy it as a scramble in mid-2026 or as a habit starting now.
Evidence: 8 European Union (2025)
In the EU, AI literacy is law, not a nicety. Document who was trained, on what, and when.
Who needs to learn what
A role-based literacy matrix.Telling people to train everyone on AI fails because it means nothing. Readiness differs by seat. The matrix below is the version we deploy in 50 to 500-person organizations: four audiences, four depths, and only one of them needs to touch a model parameter.
| Audience | Baseline (weeks 1 to 2) | Ongoing practice | Evidence it worked |
|---|---|---|---|
| Everyone (100% of staff) | What the tools do and where they fail, the data red lines, hallucination and bias basics, and the approved-tool list. | 10 minutes daily inside their own tasks (drafting, summarizing, checking) with verification built into every rep. | Passed baseline check, zero policy breaches, and a documented training record (your Article 4 file). |
| Power users (10 to 15%, self-selected) | Advanced prompting, tool chaining, and building shared prompt libraries for their team’s recurring work. | One workflow experiment per month, office hours for their team, and feeding what works into shared playbooks. | Team-level usage and time-saved deltas, plus reusable assets shipped. |
| Process owners | Redesign method: where AI does volume, where humans hold judgment, and how to instrument a before and after. | One process redesign per quarter with published metrics. | Cycle-time and quality delta on a named process. |
| Leadership | Capability and risk literacy: what to expect from the tools, what governance owes the org, and what good looks like in the usage data. | Visible weekly personal use and a monthly review of readiness metrics. | Role-modeling shows in adoption stats, the strongest signal in the high-performer data. |
Four audiences, four depths of training, and only one needs to touch a model parameter.
Evidence: 2 McKinsey & Company (2025)
Train four audiences, not one crowd. Everyone gets literacy and judgment; depth scales by seat.
What AI-ready actually means
Four layers, not one skill.AI readiness is routinely reduced to prompt training. In practice, the organizations getting value operate on four layers at once, and the ones stuck in pilot purgatory are usually strong on exactly one.
| Layer | The question it answers | What good looks like |
|---|---|---|
| Literacy | Can everyone use the tools competently and safely? | Every role has a baseline: what the tools do, where they fail, and what never gets pasted into them. Refreshed continuously, because the tools change monthly. |
| Judgment | Can people evaluate what AI produces? | People treat output as a draft from a fast, overconfident junior: verify claims, check the edge cases, own the result. This is trained, not announced. |
| Workflow | Has the work itself been redesigned? | At least one core process rebuilt around human and AI, with the human doing judgment and the AI doing volume. Bolting AI onto an unchanged process is where ROI goes to die. |
| Governance | Is it clear what is allowed, and who is accountable? | A one-page policy people have actually read: approved tools, data rules, and human sign-off points. Guardrails legitimize use; silence drives it underground. |
Evidence: 2 McKinsey & Company (2025)
Literacy and governance are week-one work. Judgment builds over months of practice. Workflow redesign is where the measurable money is, so pick one process, instrument it, and publish the before and after.
Readiness is four layers (literacy, judgment, workflow, governance), not a single prompt-writing skill.
The 90-day readiness plan
Legitimize, train, then redesign.The plan below is deliberately narrow: legitimize what is already happening, train past the point where usage sticks, then prove the value on one redesigned workflow. It is built for a People team of one to five, not a transformation office.
The next 90 days, phase by phase
Days 1 to 30, legitimize and baseline
Publish the one-page policy (approved tools, data red lines, sign-off points) and declare an amnesty for shadow users, because you want their head start, not their resignation letters. Survey actual usage anonymously and baseline confidence and skill by team, so you can measure the delta at day 90.
Days 31 to 60, train past the five-hour line
Give the recruiter, the controller and the support lead their own role-specific ten-minute daily reps, not one generic webinar. Name a coach per team, since BCG pairs training with in-person coaching for the biggest usage lift. Make judgment part of every exercise: find the error in this AI output, then fix it.
Days 61 to 90, redesign one workflow and publish the math
Pick one process with volume and pain, rebuild it around human and AI, and instrument time, quality, and satisfaction. Re-run the day-1 baseline for confidence, usage and skill deltas, then take the before and after to leadership with the next two candidate workflows. This is how pilots escape purgatory.
Evidence: 7 Boston Consulting Group (2025)
Legitimize, train past five hours, then redesign one workflow and publish the before and after.
Is your workforce AI-ready?
Eight questions, then the one-page version.Check every statement that is true today.
Evidence: 2 McKinsey & Company (2025)
| Score | Band | What it means |
|---|---|---|
| 0 to 2 | Watching | AI is happening to your organization, in the shadows. Start with policy and amnesty this month. |
| 3 to 4 | Piloting | Pockets of use, no compounding. Get everyone past the five-hour line. |
| 5 to 6 | Building | Capability is forming. Redesign one workflow and publish the before and after. |
| 7 to 8 | Compounding | You are ahead of roughly 94% of organizations. Scale what works and keep measuring. |
Evidence: 2 McKinsey & Company (2025)
Takeaways, the one-page version
This week: publish the one-page policy and open an anonymous usage survey. This quarter: cross the five-hour line and take one redesigned workflow’s before and after to leadership.
Sources and method
Every external numeric claim in this report points to one of these 2024 to 2026 sources. Forecasts and self-reported surveys are labelled so they are not mistaken for causal proof.
Superagency in the Workplace
Employee and C-suite surveys on genAI use, training and maturity: 92% raising AI spend, 1% mature, employees three times further ahead than leaders estimate, 48% rank training first.
mckinsey.comThe State of AI
1,993 respondents across 105 countries. 88% use AI regularly but only about 6% report significant enterprise-wide impact; high-performer practices include leadership role-modeling and workflow redesign.
mckinsey.comEnterprise AI Adoption Survey
Executive versus employee perception gap: 75% of executives versus 45% of employees say their organization has successfully adopted AI; 72% develop AI in silos.
writer.comA People-First Approach to AI Adoption
Roughly seven in ten US employees never use AI at work in any sanctioned way.
gallup.comShadow AI in the Workplace
More than half of employees who use AI at work conceal that use from their employer.
azumo.comEmployee AI Confidence and Training
Only 35% of organizations provide structured AI training; over 60% of employees are not confident using AI tools in daily work.
perceptyx.comAI at Work 2025: Momentum Builds, but Gaps Remain
Global survey. Regular use above 75% for leaders and managers versus 51% frontline; usage sharply higher with at least five hours of training plus in-person coaching; over half will use unsanctioned tools if not provided.
bcg.comRegulation (EU) 2024/1689 (AI Act), Article 4, and European Commission AI Literacy Q&A
Duty applicable 2 February 2025 to all providers and deployers, any size, any risk tier; national market-surveillance enforcement from August 2026; measures must be role-tailored and documented; a single onboarding video is not sufficient. November 2025 Digital Omnibus proposal noted. Not legal advice.
eur-lex.europa.eu