We Need Action, Not More Studies: The Cost of Waiting on AI

More than 80% of the seafood Americans eat is imported. That number has been embarrassing for years, and NOAA Fisheries’ answer to it has been remarkably consistent: more science. More studies. More assessments of a problem everyone in the industry can already describe in one sentence. I tried to sit down with the agency’s Assistant Administrator to talk about what domestic producers actually need. The meeting was canceled at the last minute, and then nothing — not even a polite “no thanks.” Meanwhile the imports keep arriving on schedule.

I’m not telling that story to pick a fight with an agency. I’m telling it because I now watch business owners do the exact same thing with AI, and it costs them more than they think.

Studying is the safest way to do nothing

It rarely looks like stalling from the inside. It looks responsible. First an assessment. Then a readiness review. Then a small pilot — not to ship anything, but to “inform the roadmap.” Every one of those steps is defensible on its own, and nobody has ever been fired for asking for more information. That’s exactly why the pattern survives.

Here’s the test I use. Ask two questions about the study you’re about to commission: what result would change the decision, and when does the answer arrive? If nobody in the room can answer both, you’re not gathering information. You’re buying time, and you’re paying retail for it.

Price the waiting, not just the project

Every AI proposal gets scrutinized on cost, and it should. Almost nobody puts a number on the alternative. But inaction has a price tag too, and it’s arithmetic you can do this afternoon: the hours a week that broken process burns, times your loaded hourly cost, times the number of weeks you intend to keep studying it. That figure is your cost of waiting.

Now set it next to the cost of the project. You’ve turned a leap of faith into a comparison. Sometimes waiting still wins — the data isn’t there, the process is about to change anyway — and that is a legitimate answer when you reach it honestly. What you can no longer do is treat delay as the free option. It never was. This is the same reason I tell owners they need one win before they need a strategy: a number on the board beats a plan to eventually produce one.

What action actually looks like

Action isn’t a big-bang rollout. That’s just gambling with a deadline attached. It means one workflow, a few weeks, and a number you agreed to measure before you started — the deployable version, not the moonshot. Small enough that being wrong is cheap, real enough that being right shows up in the business.

I didn’t study my way from 15–20% shrimp survival to over 90%. I ran selections, kept what lived, culled what didn’t, and did it again. It took over a million dollars and years of cycles, and every useful thing I learned came from running the next batch — not from the report about the last one. The information was on the other side of the action the whole time.

“We need more studies” is what an organization says when nobody wants to own the outcome. If that’s where your AI effort has been sitting for six months, the fix isn’t another assessment. Pick one process. Set a date. Decide now what number would make it a win. Then go find out.