AI Can Play the Buyer, But It Can't Be the Person

A friend of mine who does B2B marketing was venting to me recently.

He said he doesn't bother with real people for customer research anymore. He just feeds the model behavioral data, company info, job titles, industry — and in seconds it generates a crowd of "synthetic buyers" (synthetic audience). You can take these virtual customers and run tests on them: Would this ad copy work? Would they accept this price?

Sounds great, right?

But the more he used them, the more something felt off. The simulated buyers looked impeccable on paper — the right titles, the right industries, the right company sizes. But the moment you asked, "How would this person actually decide?" these fake buyers started getting things badly wrong.

Why?

Think About How You Buy Something Expensive

Let's zoom out and look at you.

The last time you bought a phone, signed up for a course, or purchased a car — what did you rely on when making that decision?

Your job title? Your income? Which city you live in?

All valid. But you know in your gut that the moment you actually pull the trigger, it usually comes down to something else entirely.

Your personality.

Are you the researcher who scans a hundred spec-comparison charts and immediately reaches for their wallet, or the follower who buys whatever their friends are using? Are you the impulsive buyer who places an order the second someone recommends it, or the cautious type who deliberates for three months before finally paying up?

See, that's something job titles and income brackets can't explain.

The Two-Thirds Rule: How Roles and Industries Filter for the Same Personality

There's a study that's pretty fascinating. The author spent seven years studying ten thousand buyers across fifteen industries and discovered something that surprised even him.

Buyer personalities are not randomly distributed.

There's a pattern. And it's remarkably pronounced.

He calls it the Two-Thirds Rule: within a specific role and industry, there are typically two personality types that account for at least two-thirds of the people in that group.

What does that mean in practice?

Take data scientists in the life sciences sector — their personalities are highly concentrated in one type. Airline operations folks? They cluster toward an entirely different personality.

When you think about it, it's not really surprising.

The role itself acts as a filter. A job that requires burying yourself in experiments and wrestling with data day after day naturally retains a certain kind of personality. A job where you're constantly coordinating, racing the clock, shouldering pressure — that retains a different kind entirely.

The role selects the person, and the person selects the role. Over time, personality clusters.

What Today's Synthetic Buyers Are Missing

So why do today's AI synthetic buyers keep making dumb decisions?

Back to my friend. What data did he feed the model?

Where the company is located, how many employees it has, what industry it's in. What this person posted on LinkedIn, what conferences they attend, what their job title is.

Here's what all of this data has in common: it describes the person's "professional identity."

But putting on a uniform doesn't change who you are.

The researcher put it bluntly: every person, when they go to work, plays a "work persona." And whether that work persona is your real self? Not necessarily.

But what truly determines how you behave the moment you're making a purchase — that's the person underneath the uniform.

The polished, suited-up "me" posting industry insights on LinkedIn and the "me" lying on the couch at midnight, agonizing over whether to order a fourth cup of bubble tea — are they the same person?

Yes.

But when it comes to buying expensive things, it's often the second "me" calling the shots.

And AI has only learned from the first one.

How to Make Synthetic Buyers More Human

So what do we do?

The researcher offered three recommendations. After reading them, I realized the logic isn't complicated — the key is that most people simply haven't thought about it from this angle.

First, don't generate personas from demographics and company data alone. Since two-thirds of the people in a given role and industry share highly concentrated personality types, that distribution should become a standard input field — just like company size or industry.

Second, the model needs to work at the intersection of role and industry. Looking at role alone is useless. Looking at industry alone is useless. Personality concentration emerges from both dimensions together.

Third, calibration shouldn't rely solely on demographics. A synthetic buyer can have the right title, the right industry, the right company size — and still react to an ad the wrong way. Because whether they're comfortable taking risks, how they process information, what tone they respond to — all of that is tied to personality, not to company annual reports.

To put it in one sentence:

Add "personality" as an input, and the model goes from "looks the part" to "makes decisions like the real thing."

One Final Judgment of My Own

I'm actually quite bullish on the synthetic buyer direction. B2B marketing customer research is genuinely too slow and too expensive. AI is on the right track here.

But I'm increasingly convinced that what people are really missing when they use AI for this is a dimension of data — not a volume of data.

You think cramming in more company annual reports and LinkedIn activity feeds will make the model more accurate. But you're just stacking layer after layer on the same single dimension. Stack it as high as you want — it's still flat.

Personality is the overlooked dimension that can make the whole thing stand up.

Next time you see a crowd of AI-generated "customers," ask yourself one thing:

Their resumes all check out, their companies all check out. But would they really make the same decision as the real person?

I don't know. But it's a question worth thinking about.