How to Think About AI Systems

The ROI-first framework for identifying what's actually worth building. Why most AI projects fail to deliver value and how to avoid that from the start.

Most people who start building with AI spend a lot of time building things that don't matter.

Not because they're bad at building. Because they started from the wrong question. They asked "what can I build with AI" instead of "what does this business need solved, and how do I quantify it."

The result is a lot of impressive demos and very little actual value delivered.

"Because most of what we say and do is not essential. Ask yourself at every moment: is this necessary?"

Marcus Aurelius, Meditations

The ROI-first framework

The frame that changes everything is ROI-first thinking. Before writing a single line of code or connecting a single MCP, ask:

• What is this saving? Time, money, headcount, or all three?
• How many times does this happen? Per day, per week, per month?
• What does it cost when it doesn't happen, or happens badly?
• Can I put a number on that?

If you can answer those questions, you can build something that matters and price it properly. If you can't, you're probably building a toy.

Why sales and marketing dominate

The reason so many AI systems are built around outbound, lead research, email writing, and reporting is simple: the ROI is easy to quantify and the volume is always there.

A sales team sending 50 emails a day manually is spending 2–3 hours on something that can run automatically. That's a number you can put on a table and walk a client through in 10 minutes.

That doesn't mean everything should be outbound. It means the pattern is the same wherever you look: find the high-volume, low-ROI-per-task work, document it, and build the system that handles it. The task almost doesn't matter · what matters is that it's repeatable, measurable, and currently done by a person.

The jobs-to-be-done lens

When you look at a business · your own or a client's · you're not looking for places to use AI. You're looking for jobs that need doing that are currently being done inefficiently:

• Sales research
• Writing emails
• Reading and processing invoices
• Updating CRMs after calls
• Generating client reports
• Onboarding new clients
• Scheduling and follow-up

These are jobs. AI is how you get them done better. The best AI systems are invisible · they just make a job happen faster, cheaper, and more consistently than a person could. That's the target.