Strategy4 min
The Organizational 95%
95% of AI pilots fail to deliver ROI. The gap isn't technology. It's the organization.
95% of AI pilots fail to deliver ROI. The gap isn't technology. It's the organization.
88% of enterprises have adopted AI. 61% report zero financial impact.
McKinsey surveyed nearly 2,000 organizations across 105 countries and found what I'd call the widest gap in enterprise technology history: almost everyone is buying AI, and almost nobody is getting results from it.
I've worked with 30+ mid-market companies on AI transformation this past year. The pattern is painfully consistent. The technology works. The organization doesn't.
TL;DR:
Here's how it plays out. Almost every time.
Anthropic's own research found 94% of technical tasks are feasible for AI. Only 33% show actual adoption.
The gap isn't capability.
MIT's Project NANDA confirmed it from a different angle: 95% of AI pilots fail to deliver ROI. Not because the technology broke. Because the organization wasn't ready for what the technology requires.
I call it the organizational 95%. Same failure, almost every time: nobody invested in the organization.
Only 21% of organizations have fundamentally redesigned their workflows around AI. That's where the value lives. Everyone else is bolting AI onto broken processes and wondering why it's not working.
Here's the thing most AI strategies miss entirely:
AI doesn't just change what people do. It changes what people are.
The deal analyst who spent two days building a CIM summary now gets a first draft in ten minutes. The content strategist who spent a week on competitor analysis now gets a first draft in twenty minutes. She's better at evaluating that output than anyone on her team. But nobody told her that's her new job.
That's not a workflow change. That's a career identity shift. You're asking someone to redefine how they add value, and nobody prepared them for that.
This is why "training" fails. A two-hour demo of ChatGPT doesn't address the real question running through everyone's head: Am I being replaced?
The answer is almost always no.
Anthropic analyzed a million AI-assisted conversations and found that only 0-20% of work is fully delegatable to AI, even among their own engineers who use Claude all day. 60% of work involves AI. But the human is still essential.
The problem is that nobody's explaining this. Leaders are sending mixed signals. "AI will transform everything" on Monday, "your job is safe" on Tuesday.
And people are frozen.
The resistance isn't laziness or technophobia. It's closer to grief. The person you're asking to change is watching AI do the thing they spent a decade mastering. That's not a training problem. It's an identity threat.
Software engineering is the clearest example. Senior engineers who've reframed their role, from writing code to directing AI agents and reviewing their output, are producing at multiples they couldn't before. The skill didn't disappear. It evolved. But nobody hands you the new job description. You have to write it yourself.
Most companies approach AI adoption by analyzing what's going wrong.
Wrong approach.
Instead of studying failures, study successes. In every organization I work with, there's one team or one person who's already figured it out. They're using AI in ways nobody taught them. They've built their own prompts, their own workflows, their own shortcuts.
They're usually invisible to leadership. Sometimes they're hiding it because they're not sure if it's "allowed."
Find those people. Study what they're doing. Replicate it. That's your adoption playbook. Start with a real person doing real work with real results, not a strategy deck built on assumptions. The strategy comes after. Once you know what's actually working, you can build a roadmap worth following.
One mid-market services firm, about 200 people. Their CEO heard a competitor announce an "AI-powered" client offering and called us in a panic. "We need AI too."
I asked one question: "What's the most painful workflow your team deals with every week?" He didn't know. That was the real problem.
Before we evaluated a single tool, we spent two weeks mapping eight workflows end-to-end. Not the AI-enhanced version. The current one.
Three of those workflows had 60%+ manual work that AI could handle. The other five didn't. We trained people in cohorts on those three. Client onboarding dropped from three hours to forty minutes.
We didn't touch the other five. That restraint was the point.
Six months later, the team asked us to look at workflow number four. They were ready. Not because we pushed. Because the first three worked.
That firm started exactly where you are. Different industry, same pattern.
The gap between frontier AI users and everyone else is already 6x and widening (OpenAI enterprise data). And execution is getting exponentially cheaper, which means every organization is buying more AI while the hard work of changing how people work hasn't gotten cheaper at all.
So here's your move: pick one workflow. Map it end-to-end this week. Not the AI-enhanced version. The current one, warts and all. Pick the one your team complains about most. That's where the energy is. You can't redesign what you haven't documented.
Start with the 95%. The technology isn't what's holding you back.