Sunset Date: The Deliverable Nobody Schedules

At go-live, the agreement was simple: run the spreadsheet and AI side by side for two weeks to compare results, then shut down the old system.

Six months later, both systems are still active.

The problem isn't distrust in the technology. It's that nobody scheduled the day to turn off what came before.

The company didn't deliberately go back to the past. It simply never decided on the present — and while that decision remains unmade, two teams are feeding data into different places, two reports arrive with numbers that don't match, and nobody knows for certain which one is the source of truth.

Running two systems in parallel indefinitely isn't caution. It's the most expensive symptom of incomplete adoption: it doubles the work, splits the confidence, and creates two "truths" inside the same team.

After seeing this pattern repeat across company after company, I started including a mandatory deliverable in every AI project I coordinate: the sunset date. The date on which the previous system stops existing — treated with the same weight as the go-live itself, with an owner, a deadline, and a communication plan.

Without it, the go-live doesn't close the transition. It opens a limbo nobody knows how to close.

Does your company have a defined sunset date for its ongoing AI projects? Tell me in the comments — and if not, tell me what's blocking that decision.