Your customers are lying to you, and they’re lying to themselves too

The article warns that businesses may make costly decisions using consumer research distorted by people’s self-reporting and social pressures. ESOMAR estimated the insights industry at $150 billion, including $56 billion spent on market research; the author argues that AI-generated synthetic audiences trained on flawed data can accelerate inaccurate conclusions. As an example, co-founder Alfred Malmros’s advice-call app saw usage peak at 25,000 before falling to 1,000 active users, despite encouraging survey responses.

Analysis

The investable issue is not whether synthetic audiences work in principle; it is whether their outputs are validated against observed behavior. If AI makes low-quality consumer research cheaper and faster, the likely near-term effect is more false confidence—not automatically better marketing ROI. The downside can flow beyond research vendors: brands may misallocate media, launch products against stated rather than revealed demand, and carry excess inventory. That risk is most consequential for consumer businesses with costly launches or long planning cycles, but the article provides no evidence to quantify exposure at any named company.

The potential winners are providers able to demonstrate calibration against purchases, repeat usage, or controlled experiments—not vendors selling synthetic respondents alone. Established research firms could defend value if they can improve data quality and prove predictive lift; commoditized survey work faces pricing pressure. This is a structural, 6–18 month differentiation story, not a clear immediate sector catalyst. The commentary offers a plausible failure mechanism, but no independent performance data or evidence of broad adoption, so avoid treating it as proof of imminent earnings damage.

Contrarian angle: synthetic audiences may be useful for screening concepts and generating hypotheses, provided real-world tests remain the decision gate. The market may overreact to “synthetic” as a category; the meaningful signal is whether model predictions consistently fail behavioral validation. Reassess if vendors publish credible holdout-test results or consumer companies disclose material changes to research spending or launch outcomes.

Key Decisions for Investors

• No directional trade on this article alone. Put research and consumer-exposed holdings on watch for evidence that marketing ROI, launch hit rates, or inventory outcomes are deteriorating; the article supplies no company-level data.
• For the next 1–3 months, favor diligence over a sector position: ask research and marketing-technology vendors for out-of-sample validation against purchases or usage, not just survey agreement or simulated-consumer consistency.
• Avoid paying an AI premium for synthetic-audience claims without independent behavioral validation. Revisit if a vendor demonstrates repeatable predictive lift on held-out real-world outcomes; otherwise treat the feature as vulnerable to commoditization.
• Falsify the concern if controlled pilots show synthetic-audience predictions tracking observed conversion and repeat purchase, or if consumer companies report stable launch performance and marketing efficiency despite adoption.