The Owner's AI Audit

Kris Van Meter · July 2026 · Guide

Ten questions. They take fifteen minutes, and they will tell you more about your AI spend than any vendor demo, because every one of them points at the same underlying issue: what is the spend actually optimizing for?

There are only two honest answers. Optics, meaning the work looks impressive and everyone feels current. Or outcomes, meaning revenue up, expense down, margin wider. Tools can't tell the difference. Owners can. Answer these out loud and don't grade on a curve.

The Questions

1. What number does your AI spend report to?

If the answer is a usage dashboard, a seat count, or "the team loves it," the spend reports to nobody. Spend that isn't accountable to revenue, expense, or margin drifts toward whatever photographs well.

2. Can you name one task AI eliminated last month, in dollars or hours?

Named, specific, measured. "We use it for lots of things" means nobody measured anything. A single honest "the weekly report went from three hours to ten minutes" beats a hundred vague wins.

3. Who reviews the output, and against what standard?

Unreviewed output is a liability with good formatting. If review happens, ask what it's checked against. A written standard makes review real. A vibe check makes it theater.

4. Does anything run without a human touching it?

If every AI task starts with someone sitting down and typing, you own a faster typewriter. Valuable, but the compounding starts when work happens on a schedule instead of on demand.

5. When the machine gets something wrong, where does the correction go?

This is the question that separates level two from level five. If corrections live in one employee's chat history, you fix the same mistake forever. If they get written into shared instructions that load every time, the mistake dies. Corrections that persist are assets. Corrections that evaporate are payroll.

6. Is anyone in your company rewarded for using AI more?

Careful with this one. Usage targets produce usage, and usage is the easiest number in the building to inflate. Reward eliminated hours and captured margin instead, and let usage find its own level.

7. Would the output survive your best client reading it unedited?

If the answer is no, that's fine, as long as somebody's edit is in the loop and the standard is written down. If the answer is "we wouldn't dare," the tool is a draft engine being sold internally as a finished-work engine.

8. If every AI subscription vanished tomorrow, what would break?

The uncomfortable version of the value question. If the honest answer is "nothing, we'd just be slower for a while," the spend is decoration. Something should break. The goal is to build things whose absence hurts.

9. Who owns the instructions?

Your prompts, templates, standards, and rules are the part of the system that's actually yours. The models are rented and getting cheaper by the quarter. If the instructions live in personal accounts and individual heads, the asset walks out the door with whoever built it.

10. Is your expensive AI work building something or repeating something?

Spending on the best model to design a template, a checklist, or a system is capital: you pay once and the result works every day after. Spending premium money to do the same task over and over is a subscription to your own inefficiency. Build expensive. Run cheap.

Scoring

Eight or more answers pointing at outcomes: you're operating ahead of most funded companies. The next question is scale.
Four to seven: normal, and fixable. Pick the two worst answers and fix only those this quarter.
Three or fewer: your AI spend is optics. The good news: the gap between optics and outcomes is structure, and structure can be built.

Companies rarely fail this audit because the technology fell short. They fail it because nobody in the building was paid to ask these questions. The owner is the exception. You're the one person whose incentives actually match it; everyone else can hide behind a usage chart.

If you'd rather walk through your answers with someone who builds these systems for a living, that's what Discovery is for. Bring the two worst answers. That's usually where the money is.