Before putting a new lunchbox on store shelves, convenience store CU decided to ask artificial intelligence (AI) for a reaction first instead of real consumers. BGF Retail(282330), which operates CU, recently teamed up with AI startup Intellisia to introduce an "AI synthetic consumer" technology that predicts reactions to new products, prices, and promotions without surveying people, applying it to product planning and store operations.
A fledgling startup founded only a year and a half ago is drawing attention in the retail industry. Intellisia, launched in Feb. last year, is a company that fronts "AI synthetic consumer." Instead of surveying people, it creates virtual consumers with AI to predict reactions to new products.
Chief executive Baek Seung-guk, who leads the company, is a co-founder of the personalized content and ad recommendation company Dable. In Nov. 2021, he was the key figure who sold Dable to Nol Universe at a corporate value of 200 billion won. When a three-year noncompete expired, he jumped back into founding within a week. If Dable was a company with technology that read preferences from texts people read to recommend articles, this time he turned the direction the other way. Baek explained, "We create virtual consumers from data people leave behind and work backward to find their preferences—how they would react to a new snack or marketing copy."
On the 15th, we met Baek at Intellisia's headquarters in Songpa-gu, Seoul. He said, "Surveys are only the first steppingstone for what we need to do," adding, "Ultimately, the goal is to reduce corporations' decision-making failures and the resulting expense through various methods."
◇ Synthetic consumers chosen by CU and CJ, 90% similar to human choices
Intellisia has dug deepest into food and consumer goods. In the first half alone, it conducted 100 proofs of concept (PoC), and reproducibility (how closely it matches people's response rankings and ratios) rose to 89% in the first quarter and 91% in the second quarter. About 50 of the corporations that ran these projects converted into paying clients. They include BGF Retail, CJ CheilJedang(097950), Pulmuone(017810), Fursys Group, LG Uplus(032640), CJ Foodville, CJ Freshway(051500), Kyowon, 대상, Sangmidang Holdings (formerly SPC), and Kurly.
Intellisia also played a major role in the domestic market push of "Scrub Daddy," also called the "America's national scrubber." In this company's new product concept study, the design and differentiation indicators came out high, but purchase intent was conspicuously low. Baek said, "We immediately brought back 20 synthetic consumers who only showed low purchase intent and interviewed them, and the answer was 'it's too expensive,'" adding, "By changing the price and asking, we pinpointed the threshold where 'I would buy up to 11,000 won, but above that it drops sharply.'" The client then said, "Still find a way to sell it at this price." Additional research yielded the insight that "Scrub Daddy is not an everyday item but a kitchen objet." The idea is to use a pretty Scrub Daddy where guests can see it and a regular scrubber where they can't. Baek said, "No one online mentions a regular scrubber brand and Scrub Daddy together, but AI can calculate the distance between words in data to find intersections that even people don't know."
◇ Survey 500 people? Do it more precisely in three days
The core competitiveness is time and expense. Traditional research takes two months to survey 500 people and costs about 20 million won. To create a synthetic consumer group of the same size, it takes three days and 5 million to 6 million won. Above all, you can immediately bring back the same respondents and keep digging by asking, "Why did you answer that way?" As with the Scrub Daddy study, it is difficult with human surveys for one study to flow smoothly into cause interviews and price sensitivity research. Baek explained, "With new product trends speeding up, there are many cases where you have to launch within three to four weeks, and a two-month study misses the timing—that's why corporations are paying attention to AI consumers."
Another strength is "candor." Because people are not actually opening their wallets, they tend to answer generously in surveys, whereas synthetic consumers have less of that "inflation." Baek cited a survey for launching a 150,000-won whiskey. "If you ask people, they say 'I'll buy it every week,' and if you forecast demand based on that, it gets inflated by dozens of times," he said. "By contrast, a synthetic consumer answers coolly, saying something like, 'I already drink 30,000 to 50,000 won whiskey, so I'd buy this only once a year on a special day.'" On items where consumers can answer emotionally—such as purchase frequency or purchase intent—AI can be closer to reality.
Its strengths also stand out in global expansion. Recently, food and consumer goods corporations often target domestic and overseas markets together from the outset when launching new products, but surveying people makes the expense soar exponentially. Baek said, "If you survey 500 consumers in the United States, the expense jumps to about 1 billion won," adding, "It takes several times more time and money just to validate whether the question translations are right and to find local panels." Synthetic consumers can quickly validate multiple countries at once without that burden. Intellisia has already built panels in 12 countries and is conducting more than 10% of all studies as overseas research.
There is, however, a limitation in that this technology is based on online "utterance volume (mention volume)" data. Baek said, "We got all the way from first to seventh place right for favorite vegetables, but if you ask about disliked vegetables, we don't do well." That's because people don't buy what they don't like in the first place and don't leave reviews, so there is no data to learn from. Categories that sell in bulk and no one mentions, like 5-kilogram flour sacks, are the same. Baek said, "In these areas, we aim to bolster and train the data to raise reproducibility within two years."
◇ Beyond surveys to stores… AI consumers head to the "purchase touchpoint"
Intellisia does not use synthetic consumers only for surveys. ParaStore, unveiled in May, builds a virtual replica of a real store and brings synthetic consumers directly into it. It simulates how purchase reactions change depending on placement and promotions (buy 1 get 1, posters, number of shelf facings) by displaying new products on virtual convenience store shelves. Baek explained, "We moved the entire purchase touchpoint into a virtual space."
What CU recently adopted is this very ParaStore. The two companies plan to reproduce actual CU stores in virtual space to simulate product displays, traffic flow, and MD configurations in advance.
It is also working with SpaceVisionAI, which started as an in-house venture at Dable and is now independent. The company has analyzed consumer traffic and dwell time and preferences by display location by installing cameras in stores, building numerous cases overseas, including in South Africa and Switzerland. Baek said, "We will combine this data accumulated in real stores with ParaStore."
Baek does not define Intellisia as a research company. He said, "When corporations launch 100 new products, 90 disappear," adding, "The goal is to reduce these decision-making failures and the social expense." He continued, "Surveys are only the first steppingstone; we want to become a global simulation company that supports various corporate decisions."