AI for European businesses

By Richard Garner 

Learning about U.S. consumer behaviour is one of the highest costs of entering the American market for European businesses. Digital population twins give them a low-cost way to test their assumptions before introducing new products. 

For a European company hoping to conduct business in the United States, there are many potential advantages. There is also plenty of room for error. 

A consumer product that sells well in Germany may not sell well in California. A brand name that sells well in London, England, may be perceived by U.S. consumers differently than its managers in the U.K. expect. A French food company may learn that its best prospects are in segments that demographic research never pointed to. 

Many people think of the United States as one market, but it is really many separate markets. Tastes and behaviours differ by region, age, income, cultural background, and shopping habits. Even people with similar demographic characteristics can want very different things. 

Understanding target market differences has traditionally required market research and focus groups, as well as testing new products and promotions with real consumers. AI tools now provide ways for companies to test products and promotions on a virtual U.S. population before committing a launch budget.

The expensive process of learning about the United States 

International business involves making significant decisions with little information. Pricing and other marketing choices must be settled before there is any feedback from customers in the new market. 

Market research can provide some of that feedback, but it has drawbacks. Recruiting participants is costly and time-consuming, and results can be skewed by selection and response biases. Respondents also do not always report their own behaviour accurately. 

The way survey respondents are chosen matters. Jon Krosnick, a Stanford University professor and one of the leading authorities on survey methodology, has spent decades studying the differences between random and nonrandom sampling. In an interview for an earlier investigation into the reliability of market research, Krosnick explained that random sampling gives researchers a scientific basis for generalizing findings to a wider population. Nonrandom methods, including opt-in panels, introduce selection effects that are difficult to measure or correct. 

Krosnick does not argue that every nonrandomized result is wrong. He argues that there is no reliable way to tell which ones are right. That uncertainty matters most when research is used to justify a major investment. Misjudging public reaction to a TV ad is inconvenient, but misjudging the market behind a multimillion-dollar U.S. expansion is far more expensive. 

A digital American consumer 

New AI research tools allow companies to learn more about consumers. Instead of surveying a new sample for each project, companies can use a digital representation of the population and test how different segments react to products, prices, and advertising. 

Toronto-based FishDog is creating a digital twin of the U.S. population for this use. The model is built from census and survey microdata and reflects the country’s demographic and behavioural characteristics. According to FishDog, it matches the U.S. Census within about one percentage point on most demographic dimensions and has been checked against several hundred published human survey results. 

That allows companies to run multiple studies with the same digital population without recruiting hundreds of new Americans each time. Questions worth exploring include whether a lower price would change who buys, which positioning works best with different groups, and whether young consumers like the product for the reasons the company assumed. The research may also show that some people who look like the company’s most loyal customers would reject the brand. For a European company considering a U.S. launch, answers like these could decide whether the expansion succeeds. 

Population versus personas 

A disadvantage of using AI tools in marketing research is that a language model can describe consumers in ways that sound believable without being accurate. For instance, if a researcher asks a commercial AI chat tool to describe suburban fathers, university students, and retirees, it will produce a convincing portrait of each. Those portraits may reflect stereotypes the tool absorbed during training rather than actual consumer traits, and the tool may agree too readily with the researcher’s own assumptions. 

FishDog focuses on building a representative underlying population rather than on making individual personas sound convincing. “Representativeness lives in recruitment, not in creation,” says Andreas Duess, CEO of FishDog. 

Its digital consumers are drawn from a structured population, not invented one at a time to fit a researcher’s expectations. The aim is to preserve the range of attitudes found in a real market. People with similar ages, incomes, and household composition may feel very differently about the same product. To be useful, a digital model must reflect that diversity, not reduce consumers to demographic stereotypes. 

This is especially true for European businesses whose assumptions about U.S. customers have been shaped by their experience with European customers. Sometimes the most useful finding is that the customers a company expected to target are not its strongest opportunity. 

Testing the decision before making the investment 

Consider a European consumer products firm looking to expand into the U.S. market. Its top executives expect most American demand for the product to come from high-income, urban consumers aged 25 to 40. 

Typically, these assumptions would guide the research brief, the choice of whom to survey, and the first marketing campaign. With a digital population, the company can test competing assumptions first. It could compare advertising that emphasizes convenience with advertising that emphasizes quality, or check whether household composition predicts price sensitivity better than age. It could also see whether the product would appeal to consumers outside the metropolitan markets it planned to target, or uncover unintended interpretations of its planned advertising. 

Instead of putting the first hypothesis into a single expensive study, business executives can test alternatives and identify which questions need further research with real respondents. 

Cost is a major benefit. Traditional research requires recruiting and surveying a new sample for each study. A digital population removes that step, so companies can test more concepts at the planning stage and compare a wider range of options before investing in the one or two they decide to pursue. 

Evidence concerning the digital twin 

A digital population is useful only if it is accurate, and plausible responses are not enough. It must reproduce key characteristics of the real population, and its results must correspond with independent evidence. 

FishDog validates its model through calibration and benchmarking. Its U.S. digital twin is built to reflect actual American demographic distributions, and the company says it agrees with Census benchmarks within about one percentage point on most demographic dimensions. It has also been tested against several hundred published human survey findings. 

The two tests have different goals. Calibration checks whether the digital population resembles the real U.S. population. Benchmarking checks whether the digital population’s answers reproduce results from independent human studies. Neither test can show that every prediction about a new product or purchasing decision will be correct. Together, however, they provide evidence for evaluating the model, rather than relying on how realistic its responses appear. 

Krosnick’s research points in the same direction. His work on randomized and nonrandomized sampling shows that the quality of a research finding cannot be judged by how credible it appears. That work concerns human survey sampling, not FishDog’s digital population, and applying it to synthetic research is a separate argument. But the general principle, that any method needs evidence its findings reflect the population and behaviour it claims to measure, applies here too. The way to validate a simulated consumer is to compare its findings with independent evidence, not to judge how plausible it sounds. 

The new economics of market entry 

In addition to lowering research costs, digital population studies change when and how businesses can investigate strategic questions. 

A European company can test marketing concepts on a digital population before commissioning a full U.S. market study. Early testing can help it decide which ideas deserve further work with real consumers, and how different segments are likely to read a nationwide advertising campaign before it launches. After launch, sales data and other U.S. results can be compared with the research findings. 

Research in this context becomes an ongoing activity, not a set of isolated projects. Human research, behavioural data, and digital population modelling each provide different evidence and can be combined, with real-world outcomes serving as the final measure of accuracy. 

For FishDog, the opportunity is making detailed population research available earlier and more often, particularly for companies making expensive decisions in unfamiliar markets. For European companies looking to do business in the United States, it offers a practical new resource for understanding the largest consumer economy in the world. 

Few companies can afford to test every assumption before entering a new international market. The more important question is how much they stand to lose by not testing the assumptions that matter most. In international business, the most expensive research finding is often the one discovered after the money has already been spent.

About the Author

Richard GarnerRichard Garner is a Canadian journalist, media executive and communications consultant with more than 30 years of experience across broadcast, print and digital media. His journalism has appeared in publications including The Globe and Mail and Toronto Sun. He is the Co-founder of ONE Network Productions Inc. His current writing explores the intersection of technology, business innovation and emerging market trends.

LEAVE A REPLY

Please enter your comment!
Please enter your name here