Latest / Influencer Marketing with Fexingo: Creators, Sponsorships, and Modern Endorsement Deals / How Influencer Agencies Are Using Synthetic Audiences
Transcript
- Lucas: So there's a term I've been hearing more in agency strategy meetings lately: synthetic audiences. Luna: Synthetic as in fake? Like ai generated people? Lucas: Exactly that. Not real humans, but demographic and behavioral profiles built from aggregated data — age ranges, income brackets, purchase history patterns, media consumption habits — and then run through a model that predicts how that group would respond to a specific influencer's content. Luna: So before a brand ever sends a product to a creator, they're essentially stress-testing the campaign on a digital twin of the audience? Lucas: That's the pitch. And it's picking up steam because a typical mid-tier influencer campaign can run anywhere from fifty thousand to a quarter million dollars. Brands want to know, before they sign the contract, whether the content will actually move the needle. Luna: Right, because the old way is you look at follower count and engagement rate and kind of guess. Lucas: Exactly. And those metrics have known flaws — engagement rate can be juiced, follower count can be bots. Synthetic audience testing is supposed to get closer to actual conversion likelihood by modeling what a real segment of buyers would do. Luna: How does the modeling work? I mean, what data goes in? Lucas: It varies by agency, but typically it's a blend of first-party brand data — past customer purchase records, CRM segments — plus third-party panel data from firms like Nielsen or Comscore, plus the influencer's own audience analytics. The model then generates a synthetic cohort that mirrors that target demographic and runs simulations of ad exposure and click-through. Luna: So it's like a focus group but nobody has to sit in a room with lukewarm coffee. Lucas: Pretty much. And a focus group gives you maybe twelve people's opinions. A synthetic audience can simulate tens of thousands of responses in minutes. Luna: Okay, give me a real example. Who's doing this? Lucas: There's a midsize beauty brand — I'll call them 'Glow & Co.' because they asked not to be named directly — that used synthetic audience testing last quarter. They were considering a partnership with a creator who had two point four million followers on Instagram, mostly Gen Z women. The brand's internal data showed their core buyer was actually millennial women aged thirty to forty-five. Luna: So the audience looked right on paper but was off in practice. Lucas: The synthetic model predicted that only eleven percent of the creator's audience would fall into the brand's high-propensity purchase segment. They ran the campaign anyway as a test, and the actual conversion rate came in at nine percent. They would have spent about two hundred thousand dollars and gotten almost no return. Luna: Ouch. So the model saved them from a six-figure mistake? Lucas: That's what the agency claims. They reallocated the budget to three micro-creators whose synthetic scores were above forty percent, and the overall campaign ROI improved by about forty percent quarter-over-quarter. Luna: So the synthetic test essentially became the new 'green light' gate before contracts get signed. Lucas: That's exactly where it's heading. Some agencies are now building synthetic audience scores into their standard pitch decks. If a creator's audience doesn't score above a certain threshold on the brand's target segment, they won't even present the partnership. Luna: And if today was actually useful to you — especially if you're on the brand side or thinking about creator partnerships — the way these episodes stay ad-free is listener support. It's buy me a coffee dot com slash fexingo. Just a simple way to keep the conversations going without sponsors. Lucas: Yeah, we really appreciate that. And it keeps us free to dig into stuff like synthetic audiences without anyone pulling the strings. Luna: Alright, back to the modeling. What about the ethical side? You're essentially simulating human behavior without real people's consent. Lucas: That's the big question. The data that feeds these models comes from real people — purchase histories, browsing behavior, demographic info — even if it's aggregated and anonymized. There's a difference between statistical modeling and creating a digital replica of a person, but the line gets blurry. Luna: And the FTC hasn't weighed in on synthetic audiences specifically, have they? Lucas: Not yet. But there's growing discussion in advertising law circles about whether synthetic audience testing constitutes a form of profiling that falls under consumer privacy frameworks. The California Privacy Rights Act, for example, gives consumers the right to opt out of the sale of their data. If synthetic modeling is considered a sale, brands could face compliance issues. Luna: So the same technology that makes campaigns more efficient could also create legal exposure. Lucas: Exactly. And there's another layer: what happens when synthetic audiences are used to negotiate creator pay? If an agency says 'our model shows your audience only converts at twelve percent, so we're offering less,' the creator has no way to verify that model. Luna: That feels like a power imbalance. The creator's entire business is their audience, and suddenly a black-box model is devaluing it. Lucas: Right. I talked to a creator lawyer who said they're starting to see contract clauses that reference synthetic audience scores as a factor in performance bonuses. If the model says you underperformed, you don't get the bonus — even if your real engagement was healthy. Luna: So the model becomes a self-fulfilling prophecy. It predicts low conversion, you get paid less, and you have no recourse. Lucas: That's the concern. Some creator advocacy groups are pushing for transparency requirements: agencies should have to disclose when a synthetic audience test was used and share the methodology. Luna: Do you think that'll happen voluntarily, or will it take regulation? Lucas: I think we'll see a mix. Large brands like Unilever and Procter & Gamble have been pushing for more accountability in influencer measurement. They might demand standardization from their agencies. But smaller brands may not have the leverage. Luna: What about the accuracy question? How reliable are these models when the real world is messy? Lucas: Good question. The Glow & Co. example had a two-point spread between predicted and actual conversion. That's pretty tight. But other cases I've seen had wider gaps. One fashion brand ran a synthetic test that predicted a fifteen percent click-through rate, and the actual campaign delivered six percent. Luna: So it's not foolproof. Lucas: Not at all. The models are only as good as the data feeding them, and they can't account for things like trending topics, platform algorithm changes, or a creator having a bad week. They're a tool, not a crystal ball. Luna: Still, if they can prevent even a few big misses, brands will keep using them. Lucas: Absolutely. And I think the next evolution will be real-time synthetic testing — where the model updates during a campaign and recommends adjustments to content or targeting on the fly. Luna: That sounds like it could get really powerful — and also a bit creepy. Simulating audiences in real time as they react? Lucas: Yeah, it's a fine line. But the direction is clear: data-driven decision-making is moving from hindsight to prediction to real-time optimization. Synthetic audiences are just one piece of that. Luna: I wonder how creators will adapt. Will they start building their own synthetic models to push back? Lucas: Some already are. A few top-tier creators have hired data scientists to build audience models that they can present to brands, essentially saying 'here's what my audience is actually worth.' It's a counterweight to the agency models. Luna: So the arms race moves from followers to algorithms. Lucas: Exactly. And that's why this topic matters — not just as a tech story, but as a power story. Who controls the data, who owns the model, and who gets to define value in the creator economy. Luna: That's a solid place to leave it.