The research was never the hard part

There’s never been more good advice sitting around, and it’s never been cheaper to get. Ask a model for a pricing strategy and you’ll have twelve of them before your coffee gets cold, each one completely coherent, each one probably fine. The kind of market research that used to cost a massive consulting engagement is basically free now.

A lot of founders are quietly panicking that the machines are coming for the thinking. That fear is mostly nonsense. What the models actually came for was the looking-things-up. And in a $10M to $50M business, looking things up was never the bottleneck. It never even made the top ten.

Watch what actually happens when a company stalls. The information is almost always sitting right in front of them. What’s missing is the judgment to know which lever is actually yours this quarter, somebody with skin in the game who owns the number, and an outside eye willing to say the quiet part out loud. A model just agrees with your premise. That’s how it’s wired. You wrote the prompt, so it hands you back your own assumptions dressed in executive bullet points.

The number was sitting right there

Take marketing attribution, because that’s where this usually blows up first. A $20M skincare brand I worked with had an outside agency reporting an 11.7× return on ad spend. On an executive dashboard, that’s an absolute banger of a number. The entire leadership team believed it, and the CEO was getting ready to pour fuel on the fire and put the whole ad budget on steroids. Nobody in the room was starved for data. The charts were right there.

So I went under the hood and reconciled that winning campaign against the money that actually landed in the bank account. It had generated $918 in net sales against $600 in spend. Call it 1.5×. The 11.7× on the slide was dead wrong.

And what happened next was the real tell. The post-purchase email flows that should’ve caught the buyers who bounced had been accidentally switched off during a platform migration two years earlier, and nobody had ever bothered to turn them back on. The company was hemorrhaging cash while celebrating a dashboard. Finding that didn’t take magic prompt engineering. It took someone willing to distrust the number everyone loved and follow the ledger down to the penny.

So when people ask if I’m worried about generative models replacing operators, the answer is no. If anything, the math runs the other way. The more the market gets flooded with tools handing out the exact same hundred strategic frameworks for free, the more valuable real discernment becomes. Because generating options is cheap now. Knowing which one won’t get you screwed is the only thing that matters. I could be wrong, but I don’t think I am.