BCG's advice to boards says it plainly: AI priorities must serve the business strategy, not sit beside it.
For a mid-market CEO, that means judging AI the way you judge every other way of growing the business: what it costs, what it returns, who owns it, and what you give up by choosing it.
A separate AI plan answers to different questions than the rest of the business
My rule of thumb is that a separate plan gets its own metrics, and its own metrics get their own meetings. Before long the leadership team is talking about tool adoption while the business is arguing about margin, hiring and the next customer segment. Nobody did anything wrong. The plan just never had to answer the questions everything else answers.
Ask what an AI decision costs to undo
BCG's most useful idea is to look for one-way doors, the decisions that effectively lock you into a partner, a platform or a way of working. For each one, ask what other options would remain and what it would take to reopen them.
I keep powder dry on purpose. I don't spend everything on one idea, however confident I am, because most of what decides the outcome is outside my control. A reversible bet can be wrong cheaply, while a one-way door has to be right, so it deserves a slower decision and a harder look at the numbers behind it.
When someone brings me an AI business case, I do what I learned to do early in finance, which is to check whether the number was built forward from the facts or backward from an answer someone had already picked. That check works on a one-way door as well as it does on a forecast, and you can usually tell quickly.
Cut any AI project that can't name its line in the strategy
Three questions will sort most of your list.
Which line in the business strategy does this serve? If you can't point to one, it is a side project, and it should be sized like one.
What does it cost to undo? Contract length, data you can't take with you, and people who now depend on a tool are all part of the price.
Who answers for the result in business terms, such as revenue, margin or cycle time? Someone who answers for adoption alone will keep the work beside the strategy.
A separate AI effort is right for a small experiment with a stop date
A small experiment with a hard stop date doesn't need to sit inside your strategy, and a company learning what's possible may want room to try things. The mistake is letting the experiment quietly become the plan. If it is working, bring it inside and hold it to the same standard as everything else.
An hour with your list will tell you more than another dashboard
Take your current AI initiatives and put each one next to a line in your strategy. Cut or resize anything that can't find a line. For the rest, mark which ones are one-way doors and slow those decisions down.
I help mid-market leaders make these calls, with AI as one lever among several, in a way the board can follow and the CFO can defend. If you'd like to work through your list together, I'm glad to talk.
John Wyatt · John Wyatt's page · All posts · RSS