The ten-day payback rule
If your first AI project can't pay for itself in ten days, you picked the wrong project. The shape of engagements that actually ship.
We made a rule after the fifth engagement in a row that stalled at the "interesting pilot" stage: the first project has to pay for itself in ten working days.
Not ten days to break even over a year. Ten days of real operator-hours reclaimed, deflected costs, or pipeline added. If the math doesn't work at that horizon, the project is the wrong shape.
Why ten days
Because anything longer and the stakeholder forgets. Quarterly business reviews eat roadmap items that don't show receipts. Champions rotate out. New priorities land. The project dies a quiet death, not because it was bad, but because nobody remembered to defend it.
Ten-day payback forces a different kind of selection. You can't do it on strategy slides. You can't do it on greenfield rewrites. You do it on workflows that already exist, have a cost you can name, and a human doing the toil today.
What the right project looks like
- One workflow, not a platform.
- One owner, not a committee.
- One baseline you can name in dollars before the first commit.
- One measurable outcome that leadership cares about.
If you can't write the project in one sentence, it's not ready. If you can't write the before-and-after metric in one number, the project isn't ready either.
The compounding part
Ten-day payback on project one buys you trust for project two. Project two compounds into project three. By month six you're not selling AI internally anymore — ops is pulling you in. That's the only path to meaningful leverage. The slow build is the fast build.