Most of this won’t work.
On the discipline of killing AI initiatives before they eat your quarter.
A CEO asked me last month how many AI projects her company should be running. “Seven,” I said. She wrote it down. “How many should we finish?” That’s a better question. The answer is two or three.
If you run a mid-market company and you read the business press, you are being told two contradictory things. The first is that AI will reshape your industry faster than anything has since the internet. The second is that your peers are all running pilots. Both are true. Neither is useful. What matters is which of those pilots are going to land, and who has the discipline to stop the others on time.
The pilot graveyard is the product
In a healthy AI portfolio, most pilots don’t reach production. That is not a failure mode. It is what a functioning evaluation process looks like. The failure mode is a pilot that didn’t work, didn’t get killed, got renamed, and reappeared in next quarter’s deck.
If your company has twelve AI initiatives listed on a status tracker and all twelve are green, that tracker is a decorative object. It is not reporting anything. Somewhere in that list are two or three real bets. The other nine or ten are either busywork, vendor theater, or someone’s personal moat. Finding which is which is the entire job.
Three tests to run before your next initiative review
- What number on the P&L moves if this works? If nobody can answer this question in one sentence, the initiative is not ready for the tracker. Put it back in discovery or stop it.
- What does the negative outcome look like? If the honest answer is “we’ll learn something,” that’s a research project, not a bet. Call it that and budget for it accordingly.
- Who decides to stop? If the answer is “the sponsor,” and the sponsor’s bonus is tied to the initiative shipping, you don’t have a stop condition. You have a vanity project with a ship date.
The best AI portfolios I’ve worked on have had an unofficial ninth metric on the tracker: What gets killed this quarter? Some teams compete to get on that line. They are the teams that eventually ship something people use.
What to do on Monday
Open your initiative list. For each item, write one sentence that answers the first two tests. If you can’t, mark it with a question mark and schedule a conversation with the sponsor. You do not need a new framework. You need to have one harder meeting this quarter than you had last quarter. That’s most of it.
Most of this won’t work. Knowing that before you start is the advantage.