Most marketers agree that knowing your customer is everything. Yet real customers are often the hardest to reach.
That paradox shaped a Marketing Meetup session with Peter Weinberg, co-founder of Evidenza, the synthetic research platform revolutionising how insight is gathered. Peter, once part of the legendary B2B Institute at LinkedIn, described why the next era of market research will be lab-grown, not survey-bound.
Below is a summary of his main ideas, written with the help of AI (so please excuse any tiny errors).
Table of Contents
- From Vanilla to Data: The Long History of Synthetic Substitutes
- The Empty Chair: Re-Creating the Voice of the Customer
- Hard to Reach Does Not Mean Hard to Model
- Killing the Guesswork: When AI Outperforms Humans
- Pocket PhDs and the 95 / 5 Rule
- Codified Creativity and Brand Codes
- When AI Needs Humans
- Accessibility and the Cost Curve
- How Synthetic Research Models Work
- Human Bias, Machine Precision
- Conclusion
From Vanilla to Data: The Long History of Synthetic Substitutes
Peter began by reminding the audience that “synthetic” is not a new concept. Long before AI, people found ways to recreate scarce resources in the lab: vanilla, rubber, diamonds and even insulin. Each began as a luxury and became widely available through synthesis.
His argument: synthetic substitutes do not make things worse, they make them better. Now marketing is entering the same transition. For a century, insight has been organic, gathered through focus groups and surveys. The next century will be synthetic, powered by machine-generated audiences and automated analysis.
The Empty Chair: Re-Creating the Voice of the Customer
Peter revisited Jeff Bezos’s famous “empty chair” at Amazon meetings, meant to represent the customer. The problem, he said, is that the chair has always been empty. Few marketers can afford large-scale research, and many B2B audiences are impossible to recruit.
Synthetic research fills that chair. Using Evidenza’s system, teams can create AI-generated respondents who mirror real-world buyers. A synthetic respondent can have thousands of traits , job title, company size, hobbies and demographics , making it statistically comparable to a real sample.
Marketers can now run qualitative interviews or quantitative surveys in minutes, producing results that once took months.
Hard to Reach Does Not Mean Hard to Model
One of Peter’s best-loved lines was: “People confuse hard-to-reach audiences with hard-to-model audiences.”
Fortune 500 CEOs may never attend your focus group, but there is enormous public data about them , interviews, earnings calls, books and articles. AI can model their thinking because it has read everything ever written about them.
The same applies to niche professionals such as cyber-security analysts or mining engineers. Traditional research cannot reach them, but synthetic research can simulate them with surprising accuracy.
Killing the Guesswork: When AI Outperforms Humans
Sceptics often claim AI “hallucinates”. Peter turned the idea on its head: humans hallucinate too. He told a story about a sceptical agency exec who tried to “catch out” synthetic respondents by asking if Germans had heard of Toto toilets. The AI answered correctly; the human was wrong.
Traditional market research, he argued, is riddled with bias, boredom and low-quality responses. People lie on surveys, skip questions or guess. By contrast, large language models can be tested, calibrated and improved.
Evidenza runs evaluation studies comparing synthetic survey answers with human results. Across hundreds of tests with brands like EY, Salesforce and Dentsu, the correlation has been above 75 percent , equivalent to the reliability expected in the hard sciences.
“The hallucination rate of AI is below five percent,” he said. “The hallucination rate of humans is much higher.”
Pocket PhDs and the 95 / 5 Rule
Before Evidenza, Peter helped popularise the 95 / 5 Rule through the LinkedIn B2B Institute: only 5 percent of buyers are in-market at any given time; 95 percent are future buyers.
AI now makes it possible to measure that split for every category. Some B2B sectors, such as cyber security, behave more like B2C markets with frequent purchases, while others, like enterprise software, have decade-long buying cycles.
Synthetic research can uncover these nuances in days, not months, bringing academic principles from the Ehrenberg-Bass Institute into everyday practice.
“Large language models have read How Brands Grow,” Peter joked. “They know it better than most marketers.”
Codified Creativity and Brand Codes
Creativity, Peter said, is not magic; it has structure. Just as Disney films often follow the same “lost and found” story arc, great brands rely on codes , distinctive assets such as logos, taglines and colours.
AI excels at recognising these patterns. Evidenza’s models can map a brand’s visual and verbal codes, assess their distinctiveness, and even recommend new directions that stand out within a category.
That is how Evidenza itself landed on its Italian Renaissance futurism aesthetic. The AI noted that few tech firms used that style, making it memorable by design.
Once AI knows a brand’s codes, it can generate on-brand creative work at scale , reliably, consistently and without rebellion against the brand guidelines.
When AI Needs Humans
Despite his enthusiasm, Peter was clear that AI cannot replace people. It excels at the middle of the process , collecting, analysing and formatting data , but humans are still needed at the beginning and end.
Humans prompt the AI with the right questions and verify the results, judging what matters. “AI does the middle to middle,” he said. “People still decide what to ask and what to do with the answers.”
That division mirrors other technologies: ATMs changed the work of bank tellers but did not eliminate them. Synthetic research will expand, not shrink, the role of insight professionals.
Accessibility and the Cost Curve
Synthetic research is already ten times cheaper than traditional surveys, but that does not make it cheap. A segmentation study that once cost $1 million may now cost $100,000 , still beyond many small firms.
Peter compared it to synthetic diamonds: early buyers were industrial companies using diamonds on drills. Over time, technology improved and costs fell, until lab-grown diamonds became mainstream jewellery. The same pattern, he said, will apply to AI research.
For now, even small teams can experiment using chat-based tools to explore buyer questions or category perceptions , a useful first step before investing in full-scale synthetic studies.
How Synthetic Research Models Work
Peter described Evidenza’s approach: the platform sits on top of multiple large language models. When a user inputs a topic, such as “chronic refractory gout”, the system searches across global data , academic papers, earnings calls, news articles and forums , to understand everything ever said about the subject.
It then builds AI personas that reflect that combined knowledge. Each persona represents a statistically modelled buyer, allowing marketers to run conversations or surveys instantly.
As he put it, “Hard to reach doesn’t mean hard to model.”
Human Bias, Machine Precision
Ultimately, Peter’s point was not that synthetic data is perfect, but that human data is not either. The difference is that AI accuracy can be measured, calibrated and improved. Human bias cannot.
In a world where most marketers still make decisions on “internal vibes”, synthetic research offers something better than intuition , evidence.
Conclusion
Peter Weinberg’s vision of synthetic market research reframes AI as a democratiser, not a disruptor.
By generating reliable customer insight at speed and scale, AI can make marketing more scientific, more accessible and ultimately more creative.
“The future of marketing,” he concluded, “belongs to those who stop guessing and start modelling.”