Artificial intelligence can help businesses test pricing strategies, but relying on synthetic consumers without checking their responses against real market behaviour could lead to costly decisions, research expert Dr Asma Qureshi has warned.
Qureshi, in an article published by Research World on November 1, said AI-generated estimates of how much consumers are willing to pay should be treated as simulations rather than direct evidence of what customers will spend.
The warning comes as businesses increasingly explore synthetic data to assess product features, compare price points and model consumer choices. Brandspur Brand News reports that Qureshi’s central concern is not whether AI can produce pricing estimates, but whether those estimates reflect how people behave when faced with actual spending decisions.
Research cited in her analysis shows that synthetic models can produce sharply different results for the same product. In one study, estimates for a laptop projector varied from about US$333 to US$1,074 depending on the AI model used, despite no change in the product or consumer scenario.
Other experiments found that AI systems could generate internally consistent consumer choices while substantially overstating the value people placed on hotel features. In one example, synthetic estimates for flexible cancellation were between 3.6 times and more than 13 times the human benchmark.
Qureshi said these gaps expose a risk for companies that interpret large synthetic panels as proof of market demand. Generating more AI respondents may make a result appear statistically precise, but it does not necessarily correct a model’s underlying bias.
She also noted that estimates can improve when models are trained or grounded in human research from the relevant product category. However, such gains may not carry over to other products or accurately reproduce differences between consumer groups.
See Also: Research Expert Warns AI Interviews Cannot Replace Human Conversations In Consumer Research
The article argues that synthetic research should be judged by whether it supports the same business decision as reliable human or behavioural evidence, rather than by a single accuracy score. Measures such as choice agreement, test-retest reliability and statistical fit capture different aspects of model performance and cannot be treated as interchangeable.
For pricing teams, Qureshi recommends keeping a separate human benchmark that is not used to prompt, train or refine the synthetic system. Businesses should first run AI recommendations alongside conventional research, then compare the decisions and their commercial outcomes before allowing the model to influence live prices.
She proposes an operating framework called ANCHOR, which calls on organisations to establish a real-world reference point, define the decision being made, select an appropriate synthetic model, preserve independent test data, evaluate recommendations in parallel and continually reassess performance.
The framework also highlights the need to account for privacy and fairness, particularly when systems use individual purchase records or inferred personal traits to estimate how much a specific customer might tolerate paying.
Qureshi’s conclusion is that AI can help businesses screen ideas, explore scenarios and narrow the range of prices worth testing. But real consumer evidence remains essential for decisions with significant commercial consequences.
The advantage, she argues, will belong not to companies with the biggest synthetic panels, but to those with stronger real-world data against which their AI systems can be tested.
