Customer research captures what people want to be true about themselves, not what's true about their behavior. Concept testing was a compromise forced by the cost of building. That cost just collapsed. You can vibe code a working prototype in an afternoon. Test behavior, not stated preferences.
Most AI strategy decks start with cost. Automate processes, deploy copilots, consolidate back-office. It's all inward-facing. The question that should be on page one: how does AI change the way your customers find you, evaluate you, and buy from you? That's where the existential risk lives.
Every investment memo and annual plan runs three scenarios: base, upside, downside. Three futures that feel rigorous but aren't. Stochastic simulation fixes this in principle. Now LLMs can parameterize it in practice. The full distribution of outcomes changes what you see and where you look.
AI agents will soon handle $15T+ in B2B purchases, threatening industrial companies' high-margin aftermarket businesses. When procurement goes autonomous, specification lock-in and relationship moats vanish. The winners will own the signal, controlling the triggers that start every transaction.
