We invested $10M in AI. Now my board wants ROI. Here's the framework.

This is the single most common conversation I have with executives in 2026. The pattern is identical: a year of pilots, a vendor list two pages long, a few “wins” measured in hours saved, no measurable revenue impact, and a board meeting in three weeks.

The mistake isn’t the spend. The mistake is the question. Most companies asked “what’s our AI strategy?” when they should have asked “where is our biggest constraint, and can AI move it?”

The framework is Theory of Constraints applied to AI. It’s been the playbook in operations management for forty years; AI just gives it new teeth.

Step 1 — Identify the constraint. What is the single biggest bottleneck in your business right now? Not the loudest problem — the bottleneck. The one that, if loosened by 20%, would move revenue, margin, or retention more than anything else. For most SMBs it’s one of: lead generation, sales conversion, customer onboarding, support throughput, content velocity, or hiring throughput. Pick one. Be honest.

Step 2 — Ask whether AI is the right tool to move it. Not every constraint is an AI problem. If your bottleneck is “we have no product-market fit,” AI won’t fix it. If it’s “we can’t reach enough qualified leads,” “we can’t onboard fast enough,” “we can’t draft proposals fast enough,” “we can’t keep up with support” — AI can move it.

Step 3 — Deploy at the constraint, not around it. Most failed AI deployments happen because companies deploy where it’s easy, not where it matters. Marketing automation in a company whose bottleneck is sales conversion is leverage in the wrong place — the leads that exist are still falling through. Move the constraint, not the surface.

Step 4 — Measure the constraint, not the activity. “Hours saved” is the wrong metric. Hours saved tell the board nothing because they don’t roll up to revenue. The right metric is the throughput of the constraint before and after. If your constraint was sales conversion at 12% and AI moves it to 18%, that’s the number for the board. Hours saved is a side-effect.

Step 5 — Repeat. Once you’ve moved the first constraint, the bottleneck moves somewhere else. That’s not failure — that’s how Theory of Constraints works. Re-identify, redeploy.

This framework is what I take into every advisory call. It works because it forces honesty: most “AI strategies” don’t survive Step 1, because the company doesn’t actually know its constraint. That’s diagnostic information. Better to know now than in the board meeting.

For boards: ask your team for the single constraint AI is supposed to move and the before/after throughput. If they can’t answer either, the $10M is theatre — and the next $10M will be too.