How to start an AI business

The mistake nearly everyone makes

People start from the technology: they learn what a model can do, then look for someone who might want it. That's backwards, and it produces businesses with impressive demos and no buyers.

Customers don't buy AI. They buy fewer missed calls, a clean set of books, a report that used to take two days. Whether you used AI is your business, not theirs.

Three shapes that actually work

• AI-delivered service: you sell an outcome and AI removes the slow part, so one person can serve clients that used to need a team.
• Narrow vertical tool: one expensive, repetitive task for one industry, sold as a subscription — hardest, highest ceiling.
• Speed advantage in an ordinary business: nothing about your offer mentions AI, you're just faster and cheaper than competitors.

What to do in your first month

Pick one industry you understand. Find the task in it that's repetitive, document-heavy and currently done by someone expensive. Do that task by hand, with AI helping, for one real client at a real price.

You'll learn the edge cases in two weeks that a year of building would have hidden from you — and you'll have revenue while you learn them.

Where it goes wrong

• Competing with the model providers themselves on general-purpose features.
• Pricing on your cost rather than the customer's saving.
• Shipping AI output nobody checks, in a field where being wrong is expensive.
• Assuming AI removes the need to sell. It never does.