Two AI initiatives. One quietly killed after nine months across 11,300 stores. One quietly compounding for five years across nearly 10,000 pieces of equipment. Same technology category. Opposite outcomes.
The difference wasn’t the AI. It was everything around it — and most initiatives, without realizing it, are already collecting the wrong-side habits on this list:
What Works
- Starts with a named problem
- Small pilots before scale
- Measured against a baseline
- Output plugs into an existing workflow
- Durability tracked over time
What Doesn’t
- Starts with a platform
- Full rollout skips the pilot
- “It worked” is a feeling
- Output creates a new verification step
- Success declared at launch
Nobody sets out to build the “what doesn’t work” version. It happens by default — a platform gets bought before the problem is named, a pilot gets skipped because leadership wants a visible win fast, and “it worked” becomes a launch-week impression instead of a measured number. None of those are dramatic failures on their own. They’re just quiet defaults that stack until the initiative is unrecoverable.
That’s the uncomfortable part. You don’t find out which side you’re on by asking “are we using AI.” You find out by checking your own initiative against this list, line by line — and most people haven’t done that yet.
So what — pull up your current AI initiative and check it against both columns right now. How many boxes are on the wrong side? Now what — which one item, if you fixed it this week, would move the whole initiative to the other column?
Leave a comment