Imagine all the imaginary people
If you needed proof we’ve collectively lost our minds, the startup Aaru is a pretty good exhibit.
Aaru is allegedly worth a $1bn. It uses AI agents (“bots”) to simulate human responses for market research: focus groups, surveys, concept tests, even political polling.
So instead of paying real humans to tell you what they think of your new burger, you pay an AI model to pretend to be the humans who might buy it.
What the actual fuck?
Isn't the entire point of market research that humans are weird. Contradictory. Contextual. Mood-dependent. Influenced by what they had for lunch and whether they slept and whether the person running the session reminds them of their ex.
And our big AI innovation here is: let's forget all of that.
It’s like trying to write a travel guide without visiting the place. Just ask Chat GPT to read the other travel guides and summarise it for you.
You could do that, of course.
You might even make some money, I guess.
But at some point you have to ask: what are we doing?
What Aaru actually does (and why that sounds tempting)
The pitch is simple:
- Create thousands of “synthetic respondents” that match a target demographic/psychographic profile
- Run questions, ad concepts, price tests, messaging, whatever
- Get results instantly, cheaply, at huge scale
This is attractive for obvious reasons.
Traditional research is slow. Recruiting is painful. People don’t turn up. Or they do turn up and tell you what they think you want to hear. Or your sample is skewed because the only people with time to do surveys for vouchers are… a very special kind of person.
So the idea of “press button, get insights” will make every marketing team and every product lead start salivating.
And Aaru already claims big-name usage.
I get it.
I also think it’s a perfect symbol of where the AI era is heading: we’re replacing the hard, messy parts of reality with a clean simulation and then congratulating ourselves on the efficiency.
Vibes with spreadsheets?
Here’s my problem.
Market research isn’t a maths problem. It’s a human problem.
And human problems have a nasty habit of breaking when you remove the human.
A synthetic respondent can only do two things:
- Reproduce patterns it has learned from past data
- Generate plausible-sounding explanations for those patterns
That might be useful for some questions.
But it’s also exactly how you end up with products that tests brilliantly in theory but flop in real life.
Because real customers don’t behave like averaged-out datasets.
They behave like:
- tired parents buying whatever stops the screaming
- teenagers signalling identity
- people impulse-buying because they’re stressed
- consumers saying they want “healthy” and then ordering chips
A focus group isn’t valuable because it’s statistically perfect. It’s valuable because it exposes the strange little truths people didn’t know they were thinking until you asked them.
A bot can simulate an answer.
It can’t simulate the moment of recognition when someone says something and then surprises themselves.
The biggest risk
This is the part that really worries me.
Once synthetic respondents become normal, companies will start using them not just to test reality, but to replace reality.
And then you get a feedback loop:
- Train on historic behaviour
- Simulate future behaviour
- Build products based on the simulation
- Collect new data from the products built on the simulation
- Train again
You’re no longer learning from people.
You’re learning from the shadow of people, reflected through your own assumptions.
At that point, “research” becomes performance art. A ritual. A slide deck generator. Something that looks rigorous enough to sign off a decision.
And in corporate life, that’s often the real job: not actually being right, but having enough data to back up why you made such a shit decision.
Selling comfort
Let me be even more annoying for a second.
I suspect a huge chunk of AI adoption isn’t about truth.
It’s about emotional relief.
Real research creates accountability. If you talk to customers, you might hear something inconvenient. You might have to change the roadmap. You might have to admit your idea isn’t as brilliant as you think it is.
Synthetic research offers:
- speed
- certainty-shaped outputs
- volume
- charts
- authority
And most importantly: distance.
If a bot says the concept will work, and it doesn’t, you can shrug and say “the model was wrong.” No one has to own the judgement call.
It’s the same reason people love consultants.
Not because consultants are always right.
Because they give you someone to blame.
Where this might be genuinely useful?!?
To be fair, there might be some sensible uses here I guess. But only if everyone remembers a basic rule:
Simulation is not evidence. It’s a hypothesis.
The danger is that a $1bn valuation encourages the opposite belief: that the simulation is the truth, just cheaper.
And that’s when we go full AI-mad.
The deeper irony
Aaru uses tech to do the one thing tech can’t do: replicate the human experience.
And that obsession says something bigger about this moment.
We are desperate to make the world simple.
We want the mess reduced to inputs and outputs.
We want to believe that if we model it hard enough, we can control it.
But people aren’t controllable. They’re not even consistent.
That’s why marketing is hard.
That’s why leadership is hard.
That’s why life is hard.
But it’s also why it’s interesting.
The AI boom keeps selling the fantasy that we can bypass the human parts.
I keep coming back to the same thought I’ve written about before: the industry needs a story, and AI is the story, even when the product is basically a thin wrapper around probability and confidence.
Aaru is part of that story, distilled.
A billion dollars’ worth of: “what if we could remove people from understanding people?”
Because the moment we stop doing that, we’re not building products for people anymore.
We’re building products for ghosts.
Until next time,
Jon "Just a messy human" Kaye
Choosing to live more deliberately