A friend of mine works at a Fortune 500 insurance company. Last year, her team got access to an AI tool that could summarize claims in seconds. She loved it. Then the pilot ended. No explanation, no replacement, no timeline. She went back to copying and pasting between 4 tabs. That story is the entire enterprise AI problem in miniature, and I'm tired of pretending it's just a strategy debate between executives.
The 95% pilot failure rate that MIT documented isn't an abstract number. It's millions of workers and customers who got a taste of something better and then had it yanked away while their company moved on to the next shiny experiment. Enterprises owe those people more than perpetual proof-of-concept theater.
The $124 Million Science Fair
Companies are planning to spend an average of $124 million on AI deployment over the next year, per KPMG. Morgan Stanley projects nearly $3 trillion in global AI infrastructure investment by 2028. That is real money. Serious money. And right now, most of it is funding a corporate science fair where nothing graduates.
About 50% of enterprises are running 31 or more AI pilots. Deloitte expects that number to hit 70% by 2028. More pilots! More experiments! Meanwhile, only 22% of organizations, the ones Microsoft calls "Frontier Firms," have actually scaled AI across multiple business functions. Those companies see 3x the ROI of everyone else. The gap between experimenting and executing is where all the value leaks out.
Harshita Bordiya at Info-Tech Research Group put it perfectly: leaders need to treat AI like an investment portfolio, prioritizing what delivers measurable savings and killing what doesn't. That's not complicated advice. It's what any person with a monthly budget already does. You don't subscribe to 31 streaming services to "experiment with content." You pick 3 and cancel the ones you never open.
People Are the Product That Never Ships
Here's what bugs me most. Only 5% of organizations report substantial gains from AI, according to Reclaim.ai's 2026 data. But companies that invest in upskilling their workforce alongside AI deployment see returns of 106% to 353%. The difference isn't the technology. It's whether anyone bothered to train the humans who use it.
My insurance friend didn't need a better model. She needed her company to commit to the tool she already had, train her team on it, and build it into their actual workflow. Instead, the budget went to 3 new pilots in different departments. None of those shipped either.
Devon Reyes would argue that you can't skip experimentation, that the cybersecurity risks and agentic complexity flagged by 80% and 65% of leaders respectively demand hands-on testing before you scale. Fair point. Some unknowns do require pilots. But running 31 simultaneous experiments isn't careful testing. It's indecision wearing a lab coat. You don't need 31 pilots to learn that your data infrastructure isn't ready. You need 1, and then you need to act on what it tells you.
The 59% of leaders who expect measurable ROI within 12 months are going to be disappointed if they keep spreading their budgets across dozens of half-baked projects. Portfolio discipline means picking your 3 best bets, funding them properly, training the people who'll use them, and shutting down everything else. Boring? Absolutely. Effective? Ask the Frontier Firms pulling 3x returns.
90% of CEOs say AI will redefine their industry by 2028. I believe them. But redefining an industry requires shipping products to real users, not running demos for the board. Every stalled pilot is a broken promise to an employee who was told their job was about to get easier, or a customer who was told their experience was about to get better.
My friend is still copying and pasting between 4 tabs. Her company's AI budget doubled this year. Something about that math should make you angry.
