AI Decision Making With Business Rules: Money Stays Safe (Simply Explained)

The Day I Decided AI Couldn't Touch the Money

A customer returns an item to my fashion brand in San Diego. Something has to decide what happens next. Do we put it back in stock? Send it back to the customer? Hold it for someone to look at? And do we give them store credit or not?

Every one of those choices is real money.

Put a final-sale item back in stock by mistake, and now I've got something in my inventory I never agreed to take back. Give someone credit twice on a single exchange, and I've paid out double for one piece of clothing. Hand out credit on a custom-made item that was never eligible, and that's cash gone for good.

When I built the system to handle returns, I had one clear thought that shaped everything: I did not want an AI deciding, on its own, where the money goes.

Here's the thing about AI. It's smart, but it's also a little unpredictable. Ask it the same question twice, worded slightly differently, and you might get two different answers. That's fine for writing an email. It's a disaster for accounting.

So I built it backwards from how most people would. Plain rules run first. The AI only steps in for the genuinely weird cases. And when anything is unclear, the system holds instead of paying out.

That one decision has kept money safe across thousands of returns.

Rules First, AI Second

Think of it like a restaurant kitchen. Most orders are standard. A burger is a burger. You don't need your head chef weighing in on every burger. You need a clear recipe anyone can follow the same way every time.

That's what plain rules do. Same situation in, same answer out, every time. When someone asks me "why did the system decide to restock this?" I can point at the exact rule that made the call. No guessing. No "well, the AI probably figured this."

That's what accounting needs. A paper trail you can defend to anyone, including yourself at 11pm when something looks off.

AI is the opposite. It's brilliant at messy, one-of-a-kind problems where no recipe exists. Hand it a confusing situation with conflicting clues and it'll reason through it better than any rule I could write ahead of time. But that same flexibility is exactly what makes it a bad accountant.

So the split is simple. Rules handle the cases I can write down. AI handles the cases I can't see coming. And for anything that moves money, a rule decides. The AI only advises.

Most of Your Decisions Are Just Patterns

Here's what most business owners get wrong. They assume the hard part is the rare, tricky cases, so they reach for the most powerful tool for everything.

But the volume isn't in the tricky cases. It's in the same handful of patterns repeating over and over.

In my returns system, two simple rules handle the overwhelming majority of what comes through:

If the order was already credited earlier, the only option is to restock it. Never pay out twice. That's not a judgment call. It's a fact about the order's history.

If the item is marked no-return (final sale or certain custom pieces), it goes back to the customer. We never agreed to take it back. The tag is right there on the product.

No AI touches either of these. Running them through AI would be slower, cost more, and be less reliable than a simple rule. Why bring in the head chef to make a burger?

Encode the common cases once, and your entire AI budget goes to the genuinely hard five percent. Most decisions in any business aren't edge cases. They're patterns you can name.

One more thing: even on these simple rule-based cases, the system only recommends. A real person makes the final move.

Where the AI Actually Earns Its Keep

So when does the AI get to play?

When the rules run out. There's a slice of returns my rules can't fully figure out. Exchanges where credit may have already moved at some earlier step, and it's not obvious. Weird combinations of conditions I never anticipated when I wrote the system.

That's exactly where the AI is valuable. Writing yet another rigid rule would be clumsy, but the AI can weigh messy, conflicting clues and tell me what's probably going on.

Here's the line I will not cross: the AI weighs in, but it does not act.

It writes a note on the screen. Something like "this looks like an exchange where credit may have already been issued, verify before crediting." It flags the risk in plain language. Then it stops.

The human still makes the call. The AI just thinks out loud about a hard case so the person deciding is faster and sharper. It's never allowed to pull the trigger on the money itself.

When in Doubt, Hold

If I could keep only one rule from this whole system, it'd be this one.

When anything is uncertain, the system holds. Not credit. Hold.

This works because the cost of mistakes isn't equal. Holding a return for someone to review costs a few minutes. Wrongly handing out store credit costs the full value of the item, and good luck getting it back once it's in someone's account.

So the safe move when unsure is to do nothing and wait. The AI is built the same way. When it's unsure, its note argues for holding, never for paying out. Uncertainty always lands on the cheap mistake, never the expensive one.

That's the principle I'd put in front of any business owner. Design your system so the expensive mistake is the one it physically cannot make. Not "trained to avoid it." Structurally blocked from making it.

How to Figure Out What to Fence Off

This is specific to my brand, but the thinking works anywhere. Run any decision through two questions.

First: does it move money or commit something hard to get back? Refunds, discounts, vendor payouts, pricing changes. If yes, a rule owns the decision and the AI only advises. If it's cheap to undo, like drafting an email or sorting a list, give the AI plenty of room.

Second: can you write the logic down plainly? "Already credited means restock only" is a rule. If you can state it clearly, it's a rule, not an AI job. If you genuinely can't write it down, that's the edge case worth handing to the AI to flag.

Here's the honest part. This is more upfront work than just dumping everything on an AI and hoping. You have to actually map your decisions and write the rules. That's not a flaw. That's the value. The discipline of sorting which decisions are rules and which are edge cases is most of what keeps your money safe.

I run 29 automated systems across my brand. The ones that touch money are the most fenced in, not the most independent. My pricing system covers 564 products. Anything that moves cash follows the same playbook: rules first, AI for the edges, a human on the trigger, and default to the cheap mistake.

The owners who get burned by AI aren't the ones who used it. They're the ones who let it pull the trigger on a decision it should only have been allowed to comment on. The failure wasn't the AI. It was the missing fence.

Ready to bring AI leadership into your company?

I work with a small number of companies at a time. If you're serious about AI, apply to work together and I'll review your application personally.

Apply to Work Together