Is That Your Final AI Strategy?


“Would you like to use a lifeline?”

“YES. I will need all three.”

camera zooms in on a man who suddenly regrets attending the AI strategy meeting

There he is, sitting under the bright lights like he is one question away from glory, except this is not really Who Wants to Be a Millionaire? anymore. This is the corporate version, where the prize is not a giant check from Regis Philbin, but the ability to walk back into the next steering committee without finance asking why the AI pilot somehow became a very expensive screensaver.

The question on the screen: Why is your company’s AI implementation not delivering the results you promised?

A: We skipped the boring data prep stuff
B:
Leadership thinks “AI” means magic
C:
We pitched it before we knew what it did
D:
It turns out “automate everything” was a terrible strategy

contestant stares in absolute terror

At first, he thinks it has to be A, because yes, the data was a mess, but everybody knew that, and somehow “we will clean it up later” became an acceptable sentence said by adults in a budget meeting. Then B starts looking suspiciously familiar, because no one ever actually said AI was magic, but there were definitely moments where people acted like the model would crawl through seventeen systems, fix the process, calm down operations, and create business value by Friday.

Then he looks at C and gets that awful feeling in his stomach, because pitching something before fully understanding it is not exactly rare in corporate life. Sometimes the story gets ahead of the substance. Sometimes the ambition is clearer than the operating model. Sometimes the slide is beautiful and the plan is basically “we’ll figure that out after approval,” which, to be fair, is how many bad decisions are born wearing a very nice font.

And then there is D.

contestant quietly remembers approving the “automate everything” slide

This is where Regis would lean in, smile a little, and say, “Are you sure?” Not in a cruel way. More in that deeply unsettling game-show way where the host knows you are about to talk yourself into the wrong answer on national television while your family watches from the audience with the emotional range of a hostage video. The problem is that AI rarely fails for one clean reason. It usually fails because several half-true assumptions get stacked on top of each other until everyone is surprised the thing leans. The data was “available,” but not clean. The use case was “strategic,” but not specific. The workflow was “understood,” but only by the people who did not have to use it. The change management was “included,” which often means someone added a training slide near the end.

audience goes silent and IT whispers, “We tried to tell you.”

So the contestant asks the audience, which is probably what should have happened before the strategy was approved. Ask the people who do the work. Ask the people who know the exceptions. Ask the person who knows why the official process has not matched reality since 2018. Every company has that person. Sometimes it is Rick. Sometimes it is Susan. Sometimes it is a terrifying spreadsheet named “Master_Final_UseThisOne.”

The audience selects E: All of the above.

host checks the card. “That is not an option.”

Exactly! That was the problem.

Because the real lesson is not that AI is bad or overhyped or doomed. AI can be incredibly useful when it is aimed at a real problem, fed by trusted data, placed inside a workflow people actually use, and owned by someone accountable for more than a demo. But when companies skip the unglamorous work and call the ambition a strategy, they end up under the lights trying to choose between four answers that are all uncomfortably true. So yes, use the lifelines. Use 50/50 to eliminate anything that sounds like “AI will just figure it out.” Phone a friend who understands the data. Ask the audience that has to live with the process after the consultants leave.

Because when Regis asks, “Is that your final AI strategy?” you really do not want your answer to be: “We thought the robot would handle it.”


Previous
Previous

Oh, the Data You’ll Need

Next
Next

What Matters Most When Scaling AI in 2026