The cost of specificity

I keep meeting marketing teams who describe their use of AI the same way: someone opens a laptop, produces six versions of an email in the time it used to take to get one past a second pair of eyes, and everyone in the room feels the relief of visible progress, because output going up always reads that way. What nobody in that room tends to ask is whether the thing driving the email was worth writing about in the first place — usually it's a year-old assumption about the buyer, written into a positioning document during a workshop nobody's revisited since, and six faster emails built on the same wrong assumption are just six wrong emails that arrive sooner than they used to.

I think it's worth asking why this keeps happening, because the answer isn't stupidity. Revisiting an assumption about the buyer is slow and slightly humiliating, in the way admitting you were wrong about something for two years tends to be, and it doesn't produce anything you can show a manager in the meantime. Writing another draft does. So the tool gets pointed at the part of the job that was already fast, and the part that was actually expensive — finding out you were wrong, and then doing something differently for the specific person who needed you to — stays exactly as expensive as it always was.

The two jobs

Those were always two separate jobs, and the cost is the whole story. Finding out you were wrong about a buyer meant somebody had to read forty sales calls, or sit through a research project that took a quarter to report back, which is why most companies did it once or twice a year at most and treated whatever they concluded as settled fact until the next review. Acting on it was expensive for a completely different reason: even once you knew, say, that the person signing off on security cared about something the person signing off on the budget had never mentioned, writing something specifically for the security lead, and something else specifically for the budget holder, and something else again for whoever inside the account was quietly trying to convince their own boss, took real hours from a real person, on top of whatever else that person was meant to be doing. Most companies made the sensible trade under that constraint, which was to write one document good enough to survive being read by everyone on a buying committee and genuinely aimed at none of them.

What I think is actually new here isn't that a model can write. It's that both of those jobs — finding out, and then doing something different for each person who needed you to — have become cheap enough that a company doesn't have to ration either of them any more, and the interesting question is what happens once specificity stops being something you have to save for the one deal big enough to justify a salesperson doing it by hand.

Where specificity lived

Because that, historically, is where specificity actually lived — with the salesperson, in the room, tailoring the pitch to whoever was in front of them, because that was the one point in the whole process where somebody was already being paid to pay that much attention to one buyer. Everything upstream of that conversation was written for an average person who didn't exist, because writing for the actual person hadn't been worth the hours until a human being with a commission on the line was doing it face to face.

What generative AI moves is that attention earlier, into whatever a buyer reads before anyone from the company has said a word to them. The security lead's question about evidence, or the person quietly trying to justify all this to their own boss, aren't new — every buying committee has had a version of them, and any salesperson worth their commission learned to expect them years ago. What's changed is that the answer now exists before that salesperson is in the room.

I've caught myself doing the lazier version of this on my own work, more than once — sending out a sharper draft of something whose underlying premise I hadn't actually checked in months, because the sharper draft felt like the progress that mattered, when the actual work was reconsidering whether the premise still held. That's the trap this shift makes easier to fall into, not harder: it's now entirely possible to get faster at producing specific answers to the wrong questions, which is a worse position than being slow and roughly right, because at least being slow gives you time to notice.

The real change isn't that marketing can produce more material. It's that the reasoning behind a purchase no longer has to be reconstructed from scratch inside every account. For the first time, the work a salesperson used to do one conversation at a time can begin before the conversation exists.