Want to Be Part of the <5% of AI Projects That Work? Here's What Two Years of Research Shows

November 19, 2025

Last week, Atheni.ai was at the Gartner Symposium/ITxpo Barcelona at the invitation of Gartner - the conference where 6,000+ CIOs, CTOs, and technology executives gather annually to figure out what to invest in and how to invest it.

We were one of only 30 vendors globally selected for their analyst evaluation programme - evaluated by the analysts who advise Fortune 500 boards on technology strategy. Why us?

The number one question dominating every session, every conversation, every hallway discussion across the four day event:

"How do we actually get value from AI?"

Not "which tools should we buy" or "what's the latest model capability." But the fundamental implementation question that determines whether your multi-million pound AI investment becomes competitive advantage or expensive disappointment.

The research Gartner presented validates everything we've been building at Atheni.ai since we started helping businesses integrate tools like ChatGPT into their workflows two years ago. But it also revealed something critical about why transformation succeeds or fails.

It's not about the tech, per se. It never was.

Inside this issue:

→ Why 71% of CIOs are stuck in AI “vision mode” - and the human-first question that unlocks real ROI

→ The Lenovo experiment that doubled productivity with the same AI tool - the implementation shift that changed everything

→ The hidden bottleneck nobody fixes: middle managers - the layer that determines whether AI fails or scales

→ The language mistake halving AI adoption - and how scarcity framing quietly kills transformation

→ The missing skill in 95% of organisations: translating AI wins into executive-level business value

→ The governance tiering errors almost everyone gets wrong - and how misclassification creates chaos or stagnation

→ The uncomfortable truth: leadership behaviour - not tools - is the floor every transformation stands on

🎧 Here's a NotebookLM overview of the research from this event.

71% of CIOs Are Stuck in Vision Phase with AI

What Gartner's research shows: 71% of CIOs are stuck in "vision phase" - unable to move from AI strategy to actual implementation.

This isn't because they lack budget, or because the technology isn't ready. The reality is that there's a question they need to be asking that they're not asking yet.

Most organisations so far have asked something like: "What can AI do for us?"

But the organisations that have really leveraged AI for measurable value have asked a different question: "What's the role of our humans in an AI-intensive future, and how do we guide them there?"

That shift - from technology-first to human-first - is the difference between the 5% who succeed and the 95% who don't.

When you start with "what can AI do," you get:

When you start with "what's the role of humans," you get:

The difference isn't subtle. It's the difference between 71% stuck and actually moving forward.

The Human Bottleneck: Why $15.7 Trillion Stays Potential Instead of Actual

Imagine a world in which that "potential" - the $15.7 trillion AI could add to the global economy by 2030 - could actually materialise.

For now, it's simply potential. And given our track record of embedding technology into our jobs, roles, and businesses, it will remain largely potential, not actual value.

A clear example: Wandsworth prison - one of the largest prisons in Western Europe - still uses paper files. In 2025. Not because the technology to digitise doesn't exist - it's existed for decades! But because we never built the human adaptation layer to embed it properly in our organisations.

That's not a prison problem. That's a technology adoption problem. And it's everywhere.

But imagine if we could solve that blockage. If the technology's value could flow and actually add that $15.7 trillion. Then we live in a different world entirely.

What's stopping it?

We're missing the human adaptation layer.

The uncomfortable truth: Technology change is rocketing upward whilst human adaptation capability is staying stable - or actually getting slower. Our ability to adapt isn't just slow in comparison to technology change. It's actively declining because the task of embedding technology is becoming more complicated, and we're not rising to the challenge right now.

Think about learning to swim.

The water is powerful, dynamic, constantly changing. You can't control it. But with the right technique - understanding the currents, knowing how to breathe, developing the muscle memory - you can navigate it. Even harness it. It doesn't need to be a fight - it can be a ride if we get the right technique.

Without a good stroke and good technique, it's easy to become tired and overwhelmed - thrashing and exhausting yourself fighting something that you should be flowing with.

AI is the same.

The technology moves faster than any individual can master it. New models every month. New capabilities every week. Features announced faster than anyone can learn the previous version.

We can't keep pace with technological change by learning each new tool individually. It's impossible.

What we need is the human adaptation layer - new ways to build capability that match the speed of technological development.

This is what's missing from 95% of AI transformations: organisations buy tools, expect people to figure it out, wonder why nothing changes and they're not seeing clear ROI.

That's like throwing someone in deep water and wondering why they're not Olympic swimmers yet.

The adaptation layer is the technique that turns overwhelming technology into amplified capability. It's the infrastructure that makes transformation possible.

And right now, almost nobody is building it.

What the 5% Who Succeed Are Doing Differently

Gartner presented case study after case study. Let's roll through what the research is showing separates success from failure when it comes to AI:

The Lenovo Experiment: Implementation Determines Everything

Lenovo ran a controlled experiment with Microsoft Copilot . Same tool. Same company. But two different implementation approaches.

Group 1: Standard Rollout

Group 2: Systematic Change Management

Same technology. 36% higher adoption. 100% more productivity impact.

The difference wasn't the tool, but instead the human capability infrastructure around it.

This is empirical proof that the £80M transformation problem isn't solved by better software - it's truly solved by better implementation.

The Force Multiplier Layer Most are Missing

Here's the statistic that changes how we might think about transformation: Gartner's Alicia Mullery interviewed 250 companies and found only 3 were doing anything specific for middle managers around AI.

Organisations are starting to see the light and invest in:

And they completely skip the layer that makes or breaks everything: middle management.

Why does this matter so much?

Because employees don't look to C-suite strategy when deciding whether to adopt AI. They look to their direct manager.

When your team is deciding "should I try this AI thing," they're watching:

You can have perfect C-suite strategy. But if middle managers aren't equipped to lead transformation in their teams, adoption dies.

The managers who succeed aren't doing it by accident. They've solved a specific problem most managers don't even know exists.

The Broken Link: Why Good Work Doesn't Translate to Business Value

Managers see ground-level benefits:

Executives want business outcomes:

There's a broken link between these two. And that gap is where transformation dies.

What winning organisations do: Train managers to tell value stories.

Don't say: "Carol saved 6 hours this week"

Say: "With my team of 10, we can now deliver the work of 11.5 people"

Back it up with:

Same underlying win. Completely different conversation.

One is a time stat. The other is a business outcome.

This communication skill - translating ground-level AI benefits into executive-language business value - is what turns pockets of success into organisational transformation.

And almost nobody is teaching it.

How Language Shapes Reality in AI Transformation

Helen Poitevin presented compelling research on how language shapes AI adoption. The patterns reveal something powerful:

Scarcity language creates paralysis:

Result: Zero-sum thinking. Fear. Resistance. Protective behaviour. Transformation stalls.

Abundance language creates momentum:

Result: Expansion thinking. Curiosity. Experimentation. Transformation accelerates.

Around 30% of workers fear AI will take their job (with some studies showing as high as 71% of workers concerned about AI's impact). That's the starting emotional reality in your organisation whether you acknowledge it or not.

Scarcity language confirms their fears. Abundance language transforms them.

Think about how agents work. Right now, you can build an AI agent that comes in every morning with several briefings, each succinctly delivering their findings after five different agents researched the market whilst you slept. They draft communications in your voice for your review. They quality-check work against your defined standards. They monitor workflows and flag opportunities to eliminate wasted time and resource. They review junior employees' work so that by the time you review the quality is already higher. They monitor communication and flag issues before they escalate. For senior salespeople, they identify cross-selling opportunities to review and act upon if sensible.

This is abundance. This is "tackle challenges we couldn't attempt before."

But if your language frames it as "AI doing parts of your job," you've created scarcity. And killed adoption.

The question isn't whether you'll use language around AI transformation. You will. The question is whether that language creates possibility or fear.

The Governance Balance That Makes or Breaks Progress

Most organisations make one of two mistakes - and both are category errors that treat AI as a technology rollout instead of a human challenge:

Too far one way: Govern everything as high-risk Result: Kills experimentation, slows adoption, teams work around official channels. No AI progress at all.

Too far the other way: Treat everything as low-risk Result: Massive compliance liability, data leaks, quality disasters. Potentially damaging consequences.

The organisations that succeed understand use cases exist in tiers with different risk profiles requiring different oversight:

Tier 1: Low Risk, High Learning

Example: Carol in marketing spending one hour learning to automate her social media posting = 20 hours a week saved. That's specific, huge ROI. But the same skill for someone in compliance returns zero. We must get more sophisticated matching people to the right tools, agents, automations, use cases, and skills.

Tier 2: Medium Risk, Clear Oversight

Tier 3: High Risk, Strict Control

The mistake is miscategorising. Treating personal research like regulatory compliance kills momentum. Treating customer data like personal productivity creates liability.

Get the tiers right, and people can experiment safely whilst the organisation manages risk appropriately.

Get them wrong, and either nothing happens or disasters happen.

Even if you only have access to Copilot at work - you can already be building custom GPTs and honing your context skills. The tools exist. The question is whether you're building the capability to use them well.

The Three-Legged Stool (Now With Gartner's Validation)

At Atheni, every engagement starts with what we call the three-legged stool. Gartner's research validates why all three legs are non-negotiable.

LEG 1: SKILLS What are the highest-ROI AI capabilities for each specific role and workflow?

Not "AI in general." What capabilities move the needle for THIS person's work?

Gartner's research on job redesign shows 20x bigger impact from redesigning how work gets done versus hiring or firing. That's not about more people or different people - it's much more about building new capabilities in the people you have.

LEG 2: TOOLS What technology matches those capabilities?

Start with skills, THEN match tools. Not the other way around.

The Lenovo case study proves this. Same tool, different results based on how implementation was approached. Tool selection matters, but it's downstream from capability development.

LEG 3: USE CASES Where can people practise safely and build real fluency?

This is the tiering framework. Real work scenarios with appropriate governance by risk level.

Gartner emphasised: Organisations that reach high AI maturity achieve 2x success rates, 4x more business unit trust, and 3x faster scaling. Maturity isn't about technology sophistication - it's about systematic capability building through real use cases.

Missing any one of these three legs means the stool falls over. That's the 95% failure rate right there.

But Even Perfect Stools Need Level Floors

You can build the perfect three-legged stool. But if the floor isn't level, it still topples.

Leadership is the floor.

Gartner's research was unambiguous: AI transformation succeeds or fails based on leadership behaviour, not technology sophistication.

The managers at Lenovo who drove 2x better results weren't technical experts. They were change leaders who:

Led by Example Used AI daily, shared what they were learning, made mistakes visibly
Created Safe-to-Fail Environments Made experimentation normal, celebrated intelligent failures, focused on learning
Communicated in Abundance Language Framed AI as capability amplifier, connected to mission, showed expanded possibilities
Built Organisational Scaffolding Created peer cohorts for shared learning, linked AI use to performance reviews, made sharing automatic

This is what levels the floor. Without it, even perfect individual capability can't scale.

The Nissan Transformation: Getting Communication Right

Nissan implemented a new customer data platform. They didn't just train people and hope for adoption. They stratified communication by stakeholder group.

Executive leaders got their own communication:

Collaborators (people directly using the system) got their own communication:

Observers (people affected but not direct users) got their own communication:

One platform. Three completely different communication strategies.

Result: High adoption, sustained usage, business value achieved.

The mistake most organisations make: treating communication as one-size-fits-all. "Here's the new AI strategy" sent to everyone creates confusion, not clarity.

Different stakeholders need different information at different altitudes. Your CFO doesn't need the same detail as your operations manager. Your board doesn't need the same message as your frontline team.

Get this wrong, and you've communicated but not connected. Get this right, and adoption becomes inevitable.

The Opportunity in Front of You

If you're reading this and thinking "we need this infrastructure," here's what matters:

The window where AI capability creates competitive advantage is closing.

Right now, "AI-fluent professional" is a differentiator. In 18-24 months, it's table stakes. The organisations moving now are building advantages that compound. The ones waiting for "clarity" are falling further behind daily.

But moving fast without moving smart just means failing faster.

The question isn't "are we using AI?" The question is "are we building the human adaptation layer that turns AI access into sustainable competitive advantage?"

That requires:

This isn't simple. But it's systematic. And the research is now crystal clear about what works.

Gartner studied hundreds of organisations. The patterns are undeniable. The 5% who succeed aren't smarter or luckier. They're building the human capability infrastructure that makes transformation possible.

The 95% who fail are still treating this as a technology problem.

Where To Go From Here

If you want individual capability: Sign up at Atheni.ai. Answer our questions, get matched to your customised capability pathway. Your personal adaptation layer, built for your workflow.

If you're leading transformation: Book an Ignition Session with George Barnes (george@atheni.ai). Two hours where we diagnose your capability gaps, identify your highest-ROI opportunities, spot your adoption barriers, and map your 30-day quick-win plan to build confidence and learn what's possible.

If you're building leadership capability: LAILA Programme (mackenzie@atheni.ai) - leadership development co-created with expert trainer Bazil Arden. Because if leaders can't lead through uncertainty and model capability building, transformation doesn't happen.

If you're redesigning what entry-level looks like: We're helping clients reimagine graduate programmes when junior roles shift from grunt work to automation architecture. Let's talk about what that means for your organisation.

For education: Zephyx.ai - systematic AI literacy infrastructure that's safe, secure, and actually helps students learn better. www.zephyx.ai

The research is clear. The methodology is proven. The question is whether you'll build the human adaptation layer before your competitors do.

Because the technology will keep racing forward; the organisations that win will be the ones who solved for how humans keep pace.

That's the opportunity. That's where competitive advantage lives. And that's what Gartner just spent three days validating for 6,000 technology leaders.

Where are you on your journey? If you're interested to see what's possible for you and your team - get in touch.

Mackenzie Howe Co-founder & CEO, Atheni.ai & Zephyx.ai | MIT Applied Generative AI | Masters Human Capital Economics | Author of upcoming Aevitas-backed book on AI transformation

Forward this to someone trying to turn AI potential into actual results.

Quick Links: