In the world of customer listening, CX appears to be stuck between a rock and a hard place: The customer survey is dead. Insights into how people think and behave – and what this means for CX – remain difficult to extract. Behavioral data collected passively can point to trends and preferences, but falters on depth. The idea of turning to synthetic data to fill the gaps gives some practitioners anxiety.
What if another option existed?
A ground-truth layer built from real behavioral observations captured over almost a decade, at scale, for gauging how creative ideas will land.
A little under 10 years ago, when unstructured VoC data was messy and unusable, creative agency Episode Four embarked on a longitudinal research project built around an open-ended, million-dollar question: What would you do with your free time and extra money?
"We asked that in a few different ways, but we asked in an open-ended way," says Mark Himmelsbach, founding partner of Episode Four and RYA and a former consultant at the US Secret Service. "Because if I gave you one wish – and you couldn't wish for more wishes – you would likely ask for more time in life to do stuff, or more money."
Himmelsbach and his business partner, CCO and co-founder Teddy Lynn, started surveying 1,000 US-based consumers every week… and have continued to do so for more than nine years.
Today, their dataset comprises nine billion data points and is still growing. It covers:
- 180 genres, such as video games, sports, arts, television, film;
- 20 actions, for example "I want to spend time with friends", or "I want to go on a trip";
- 30 demographic questions.
Those data points power RYA, which is increasingly functioning as a verification layer for AI data pipelines: a way to ask whether synthetic audience models truly reflect how real people behave.
Following the launch of RYA's API, CX Network spoke to Himmelsbach to find out more.
This interview covers:
- How RYA ensures its insights reflect real-world consumer sentiment and avoid regressing to the mean.
- How RYA's data inspired a muscle car giveaway at the 2026 PGA TOUR in Fort Worth, Texas – and what it says about puppets.
- Measuring the ROI of synthetic data and using it to solve current challenges around consumer engagement.
What would you do with your free time and extra money?
For Episode Four, the shift from creative agency to data and AI company occurred "a bit by happenstance".
"We're very intentional now, but we kind of stumbled our way into it," Himmelsbach explains.
In the mid-2010s, very few people were talking about synthetic data. But Himmelsbach and Lynn had a problem that was best solved with data.
"The whole origin story of how we started collecting data was 'how do we get brands to take bigger creative leaps?'," Himmelsbach says. Although he and Lynn had names such as BBDO, IPG and Ogilvy on their CVs, "two middle-aged guys walking into any brand saying 'trust us this creative idea is cool' wasn't going to fly."
The result was more than a stronger pitch. Initially, it elevated creative work by grounding it in data-driven consumer insights. As the side quest gained pace, it reshaped Episode Four into a data- and AI-led operation, which would pave the way for a whole new venture, RYA, short for radical yet acceptable.
Few examples so clearly embody how AI should amplify rather than replace human creativity.
RYA officially launched in November 2024 as a creative AI tool that generates unique cultural insights and novel creative ideas almost instantly. This year the capabilities were made available through an API.
"We had interest from a handful of brands and now agencies and consultancies to bring our data into their platform," Himmelsbach says. "It's a novel data set, so why limit other people using it?"
The target market is mature Fortune 500 and larger agency, holding companies and consultancies that have built their own platform. One banking client uses RYA to specifically research trends among Asian American small business owners in Los Angeles. Adobe uses it to find creative professionals in different industries who happen to be diverse in background.
How to create a trustworthy synthetic dataset from billions of real data points
RYA's nine-billion-plus data set is balanced to reflect the entire US population and accounts for different take-up rates among different populations.
"These are real actual people," Himmelsbach says. "We can proxy nearly any audience for a brand and these real people will tell them where they would spend their free time and extra money, both open-ended and closed-ended questions."
Then the synthetic data takes over.
On the technicalities, Himmelsbach explains: "It's more machine learning on the raw data side but then to manufacture those outputs we've built an AI harness. We now orchestrate eight different LLMs, all of which excel at a different job. Then there are a couple of hundred workflows, all custom built with our domain expertise around marketing to make sure the output is the best answer possible."
"By asking it and chatting with the AI, it takes those real answers and turns it into proxies for what that audience would say," he adds.
One RYA client uses Claude to plug all their data into a Model Context Protocol (MCP) that allows broad access to the insights. Another company, in the consumer packaged goods industry, uses it to supplement their sales data with cultural and interest-related insights.
"There's a really interesting gulf between large brands that have the bandwidth and expertise to build their own platform internally and then others who need to kind of string together other ones for their employees," Himmelsbach says.
Why synthetic customer panels produce boring answers
Although breadth may be the goal, particularly when it comes to synthetic data, a danger lurks in insights that have been flattened out to reflect the "average consumer".
According to Himmelsbach most synthetic data sets quickly regress to the mean, "and that's problematic in a couple ways", not least because the output is "pretty predictable, boring responses".
First, average eliminates nuance and "for creativity we want the ends of the bell curve as well as the middle", Himmelsbach says.
The second trap is that diversity is ignored. "Synthetic data sets smooth out the edges, the extremes and the diversity. You lose a lot of the nuance that makes having data – and using that data to form creative ideas – special," he continues.
Proving his point, RYA's data has uncovered a number of consumer trends that are anything but boring. In fact, Himmelsbach says "there's always something surprising".
Among those surprises are the music and art tastes of high net worth individuals (HNWI) and the prevalence of puppets.
In researching HNWI trends for a financial services firm, RYA served up data that confirmed those with more than US$10 million to invest love art, but hate art museums. Instead, they want to visit galleries and exhibits where they can buy what's on display.
"Do you want to guess their favorite musical genre? It's EDM," Himmelsbach reveals. "People think that result makes sense because these individuals made their money in Silicon Valley, or they're the New York, new money individuals. But that's not true. The music they're most passionate about is trap or southern hip-hop and that blows people's minds," he adds.
When Episode Four pitched the idea to award a restored 1982 Schwab Scrambler to the winner of the 2026 Charles Schwab Challenge at the PGA TOUR in Fort Worth, Texas, RYA's data confirmed it would be a hit. "When we first pitched that idea to Charles Schwab, the CEO was like 'why in the heck would we ever do that?' But the data said it would work and he was sold," Himmelsbach recalls.
"By having the data brands have permission to take a big leap. It's kind of de-risking creativity with data," he explains.
Another genre that pops up frequently is puppets. "It turns out there's puppets on Broadway and a lot of people knit puppets! There's always something surprising the data," Himmelsbach says. "When applying it to creative ideas, it's great to have that that nuance and surprise."
Justifying an investment – and proving a return – in synthetic data
Basing operational and CX decisions on external, population-scale datasets requires a leap of faith from practitioners. They need insights that can hold up to scrutiny, retain nuance and reflect diversity. They also need to be able to show ROI.
But synthetic data is splitting the insights industry.
In background research, CX Network has heard many reservations, from verifying representation claims, to trust and accuracy issues and concerns around how quickly it can respond to real-world events; everything from interest rate rises to competitor product launches.
"Where the world is going is everyone is offering this data and no one can actually source it, back it up, or verify that it's right," Himmelsbach says. "People still need the insight, but being able to trust the insights immediately is a huge gulf, and I think it will be the next big topic of conversation."
Another huge topic of conversation in CX is ROI. CX Network's 2026 research into the state of CX found 52 percent of practitioners say the pressure to prove ROI is increasing.
Himmelsbach says one client evaluates time saved and use cases to understand how RYA allows human workers to re-focus their time on creative work. AB testing is also valuable for indicating the performance of ideas validated by RYA vs those that are not.
While RYA was originally a litmus test for creative ideas, in its current form it has the potential to guide brands through some of the most difficult questions in CX right now: How can we communicate, connect and engage customers in a meaningful and memorable way?
The established tools – passive listening methods such as behavior analysis, social media monitoring and sentiment analysis – have already been updated with AI capabilities, but what if they're the wrong AI capabilities? The data overwhelm, inertia and indecision that practitioners report from these insights indicates that CX needs a new approach.
"I absolutely love the creative industry and when using technology to apply creativity, humans are essential in that mix," Himmelsbach says. "No machine could ever have the taste, the nuance or craft of humans. But this helps us get there faster and helps humans break through quicker, which I find really inspiring."