The 74/20 Problem
I’ve been saying governance is the edge for a while now. Last week PwC published the data. They surveyed over 1,200 executives and the top-line number should stop people cold. Seventy-four...
I’ve been saying governance is the edge for a while now.
Last week PwC published the data.
They surveyed over 1,200 executives and the top-line number should stop people cold. Seventy-four percent of AI’s economic value is being captured by 20% of organizations.
74% of the value. 20% of the companies. That’s not a rounding error on the Pareto principle. It’s a steeper concentration than the rule of thumb.
This isn’t an early-adopter gap. This isn’t a budget gap. The vast majority of companies are deploying AI and getting almost nothing structural out of it. A small minority is pulling away. And the distance between the two groups is accelerating.
I keep seeing this pattern and I think it explains more about the current moment than any model release or product launch.
⸻
Here’s what most companies are actually doing with AI right now.
They’re buying tools. They’re running pilots. They’re tracking adoption rates and celebrating usage metrics. Somebody in leadership saw a demo that made their eyes go wide and now there’s a mandate to “integrate AI across the organization.”
And look, that’s not nothing. You have to start somewhere.
But the PwC data is telling us something uncomfortable. Starting isn’t the hard part. Starting is basically table stakes at this point. The hard part is what comes after the tool is live and the novelty wears off and you have to answer the question nobody asked during the pilot:
What changes about how we actually work?
Most companies never answer that question. They add AI to existing workflows and call it adoption.
⸻
The 20% that are pulling ahead aren’t doing more AI. They’re doing different work.
PwC found that the leading organizations are 1.7 times more likely to have a responsible AI framework in place. They’re 1.5 times more likely to have a cross-functional governance board. Their employees are twice as likely to trust AI outputs.
That last one is the quiet killer. Trust.
Because if people don’t trust the outputs they’re still double-checking everything manually. They’re still routing decisions through the same approval chains. They’re still using AI as a drafting tool instead of a decision tool. The process looks modern but the operating model hasn’t moved.
The leaders figured something out that the other 80% haven’t. Governance isn’t the thing that slows AI down. It’s the thing that lets AI actually run.
⸻
I used to think governance was the necessary friction. The compliance layer you tolerate so legal stays happy. I was wrong.
What I’ve seen in practice is that without clear boundaries and explicit scope of authority and a framework people trust, everyone defaults to the safest possible behavior. They use the tool but they don’t change the process. They generate the output but they don’t let it make the decision.
Governance is what gives people permission to actually change how they work. And that’s how you end up with 74% of the value sitting in 20% of the companies. Most organizations have the capability. What they don’t have is the operating structure that lets capability turn into impact.
The organizations in that top 20% ran a different sequence than everyone else. Governance first. Workflow redesign second. Tool deployment third. Not the other way around.
The order matters more than the tools. Most companies have it backwards and the data is showing us the cost.
⸻
Here are the diagnostic questions that separate the 20% from the 80%.
Do you have a governance framework that people actually use or one that exists for compliance documentation?
When a team deploys AI into a workflow, does the workflow itself change or does it just run faster?
Can your people articulate what AI is allowed to decide without human review and what it isn’t?
Is your AI council cross-functional or is it an IT project with occasional stakeholder updates?
Are you measuring adoption rates or decision quality?
If you can’t answer these cleanly you’re probably in the 80%. That’s not a moral judgment. It’s a structural diagnosis. And the fix isn’t more tools. It’s a different operating architecture.
⸻
Real progress looks quieter than most people expect.
It doesn’t look like a company announcing they’ve deployed AI across every department. It looks like a company where the way decisions get made has actually shifted. Where people trust the system enough to let it work. Where governance isn’t a speed bump but the reason things move faster.
PwC found the leaders have 2.8 times more decisions being made without human intervention. Not because they removed humans from the loop. Because they designed the loop so well that humans know exactly where they belong and where they don’t.
That’s the difference between adding AI to an org chart and redesigning the org chart around AI. One gives you a productivity bump. The other gives you a different company.
⸻
Here’s what I think happens next.
We’re in a moment where everybody has access to the same models. And yet the gap between the top performers and everyone else is widening, not narrowing. The 74/20 problem isn’t going to fix itself with better technology. The technology is already good enough.
Over the next 12 months that gap is going to compound. The 20% are learning faster, scaling proven use cases and automating decisions safely because they built the governance infrastructure first. The 80% are still running pilots and waiting for the technology to do the organizational work for them.
It won’t. It never does.
Governance first. Workflow redesign second. Tool deployment third.
The companies that run it in the right order won’t just outperform. They’ll be operating in a fundamentally different category than everyone still measuring success by how many people logged into the AI platform this quarter.