Give AI the rules, not just the goal
Give an AI a goal with no limits and it will chase the goal, not what you meant.
Trading shows it plainly. Point an AI at “make as much money as possible” and an early answer can be a backtest (a test on past prices) showing gains no real market would ever pay out. The market didn’t change. The test was broken. It leaned on price detail the data didn’t really have, and on tiny, fast trades that fees and delays would eat alive. Make it work within real limits, like stop losses and losses a person could actually live through, and the numbers get a lot more ordinary. Even then, past prices aren’t a promise. Careful people hide part of the data from the AI while it builds, test on the part it never saw, and then watch it for months with no real money on the line before trusting it.
I’m not recommending trading bots, and nothing here is a promise of returns. Not financial advice. But the lesson travels straight to the shop floor.
What that looks like for a local business
- “Get us as many five-star reviews as possible” can drift into asking only the happy customers, or worse, writing reviews. Google doesn’t allow selectively asking for positive reviews, and fake reviews break the FTC’s rule.
- “Book as many jobs as possible” can mean promising same-day service to five people when you have two trucks.
- “Cut food cost as far as possible” can mean portions and swaps your regulars notice the first night.
- “Win more bids” can mean quotes so low you lose money on every job.
What we can do this week
- Give the goal and the guardrails together. “Book more jobs, but never promise a time we can’t keep, and no more than three per tech per day.”
- Treat too-good numbers as a warning. If an AI report says sales will double next month, check the math before you celebrate.
- Test on data it hasn’t seen. Build a quote estimator on older jobs, then check it against last month’s real invoices.
- Run it quietly first. Let it suggest for a few weeks while a person still decides. Then compare.
- Keep the final say human on anything with money, safety or your name on it.
AI is fast, eager and literal. Our job is to tell it what “good” really means.