McKinsey's numbers on AI adoption tell a strange story if you read them closely. 88% of organizations report using AI in some form. Almost two-thirds haven't gotten it past the pilot stage, and only 39% can point to a measurable bottom-line impact. That's not a technology problem. The models most of these companies are using are, in most cases, the exact same models everyone else has access to. The gap is somewhere else entirely.
I've spent the last year running AI agents across four different companies I'm personally involved in: an executive recruiting firm, a robotics maintenance company, a security and executive protection firm where I serve as COO, and a home services business. Some of what runs day to day now: lead sourcing, social content, calendar syncing, sales outreach, inbox management, LinkedIn research. None of it started as "let's roll out AI." It started as "this specific task is eating my week, let's fix that one thing."
That's the difference I keep seeing between what works and what stalls. Companies that treat AI as a platform rollout, something IT deploys and everyone gets trained on in a single afternoon, get exactly what a platform rollout gets: adoption numbers that look good on a slide and almost no change in how the work actually gets done. The tool gets bolted onto the old workflow instead of the workflow getting rebuilt around what the tool can actually do.
What's worked for me is closer to what 30-plus years of operations leadership already taught me long before AI was part of the picture: fix the process, not the org chart. Find the actual bottleneck, not the one that's easiest to point at in a meeting. When I built an automation for a LinkedIn outreach inbox on one of my ventures recently, it wasn't because "AI should manage inboxes" as a category. It was because I watched exactly what was eating time in that specific inbox, and asked what a genuinely good assistant would do about it. Same discipline I've used running P&L for three decades, just pointed at a different set of tools.
There's a leadership piece here too, one that gets less attention than the technical piece. Every one of my companies runs differently, with different people, different risk tolerance, different comfort level with automation. The workflows I've built for a security and protection firm look nothing like the ones running for a robotics maintenance company, because the businesses aren't the same and the people in them aren't the same. AI strategy that ignores that and hands every team the identical tool the identical way is exactly why two-thirds of companies are stuck in pilot purgatory. You can't lead four different rooms the same way, and you can't automate them the same way either.
Four hundred-plus episodes of interviewing founders and executives on my podcast has reinforced this from a different angle. The guests who talk about AI as a genuine advantage almost never describe a company-wide initiative. They describe one annoying, specific problem they got obsessive about solving. The scale came later, once the first fix actually worked and people trusted it enough to hand over the next one.
If there's one thing I'd tell another operator sitting where I sat a year ago: don't start with the technology. Start with the bottleneck you already know is there, the one you've been complaining about for months, and build toward that. The AI part turns out to be the easy part. The operations discipline to actually redesign the workflow around it is the part almost nobody wants to do the unglamorous work of getting right.
Steve Urban is the Founder and Chairman of Riderflex and Robo Reliance, COO at MPS Security & Executive Protection, and host of the Riderflex Podcast. More at steve-urban.com.