You're Using AI to Run a Cron Job

I walked into a 20-person accounting firm last month and found a $3,000-a-month AI platform running... scheduled email reminders.

Not complex analysis. Not document understanding. Not predictive forecasting. Email reminders. The kind of thing a $20 Zapier integration handles in twelve minutes.

The founder was proud of it. "We're an AI-first firm now," he told me. His margins didn't agree.

The Enterprise Shadow

Small businesses have learned the wrong lessons from big companies. Enterprises buy Salesforce when a spreadsheet would do. They deploy SAP for processes that barely exist. They hire consultants to implement software that replaces three people's judgment with forty people's configuration.

And now they're doing it with AI.

The pattern is identical: identify a problem, assume it needs a sophisticated solution, buy the most impressive platform available, then force the team to bend around it. The only difference is the price tag went up.

Why This Happens

Part of it is marketing. AI vendors are exceptionally good at making simple automation sound like machine learning. If you wrap a basic rule engine in enough neural network language, it sounds like you need a data scientist to operate it.

Part of it is fear of missing out. When your competitor announces they're "leveraging large language models," you don't want to admit your solution is... slightly smarter mail merge.

But the biggest part is diagnostic failure. Most businesses skip the step where they actually understand the problem they're solving. They see a symptom — "we're slow at follow-ups" — and jump straight to the most exciting treatment.

The Honest Audit

Here's a framework I've used with a dozen SMBs. It takes twenty minutes and has saved every single one of them money:

Step 1: Write down the exact task you want AI to handle. Not the category. The actual task. "Answer customer questions about shipping" not "improve customer experience."

Step 2: Estimate how many distinct decisions the task requires. If it's fewer than ten, you probably don't need AI. You need better rules.

Step 3: Ask whether a human doing this task would need judgment, context, or creativity. If the answer is no, you're looking at automation, not intelligence.

Step 4: Price the simplest solution that handles the task. Not the AI solution. The simplest solution. Then price the AI solution. If the gap is more than 5x, you need a very good reason.

I've seen companies spend $15,000 on an AI chatbot that handles twelve unique questions. I've seen others spend $200 on a well-structured FAQ and a contact form that covers the same ground with higher satisfaction scores.

The Real Cost

The money is annoying but recoverable. The real damage is opportunity cost. Every dollar and every hour you spend over-engineering a simple problem is a dollar and an hour not spent on a problem that actually needs intelligence.

The inventory forecasting that could save you from a stockout. The pattern in your customer churn that nobody's spotted. The contract clause that's cost you money three times this year.

Those problems exist in your business right now. But you won't find them while you're busy teaching a neural network to send reminder emails.

The Question to Ask This Week

Pick the AI tool you're most excited about right now. The one you're considering buying, or the one you just deployed.

Now describe the actual task it performs using only words a ten-year-old would understand. No jargon. No platform names. Just what it does.

If that description sounds simple, it probably is. And simple problems deserve simple solutions.