The $4.8 Million Mistake That Taught Me to Fear the Word “Help”

I once took a $4.8 million loan that was supposed to help my company. Government money, built to support producers like us. On paper, exactly the help we needed.

It took 18 months to close. When we finally did, I had to bring $300,000 of my own to the closing table — just for lawyers and fees, none of it rollable into the project. The terms were so rigid (full collateral, a personal guarantee, a rule for everything) that a profitable, capable business ended up strangled by the very thing meant to save it. My honest verdict: the worst decision I ever made. I’m a little ashamed I was that eager for the “help.”

I watch owners make the same mistake with AI

Not by doing nothing — by taking the big, rigid “help” that looks like support and quietly buries them. The difference between their mistake and mine is mostly the invoice.

The AI version of my closing table

It’s the 18-month enterprise platform. The vendor you can’t leave without a migration project. The all-in-one “AI transformation” that needs a dedicated team to run and a lawyer to escape. It arrives dressed as progress and leaves you slower, poorer, and locked in — paying for capability you never quite use. That isn’t a technology problem. It’s my loan, in a different suit.

The rule I learned the hard way

Cheap, flexible, and reversible beats big and rigid — every single time. The best AI you can buy is the one you could walk away from next month without a lawyer. So start small, keep the exits open, and let resultsdecide what’s next — not a contract you signed before anything worked. That’s the whole idea behind shipping the deployable thing instead of the moonshot, and behind earning one real win before you commit to a strategy.

Real help doesn’t need collateral. If an AI “solution” wants an 18-month commitment before it saves you a dime, it isn’t help — it’s the closing table all over again. And I’ve already paid to learn how that ends.