Synthetic respondents are the problem, not synthetic research

Synthetic research is spreading quickly through asset management marketing, but there is little agreement on which parts of the research process it should touch. Under pressure for speed and efficiency, some teams skip data gathering altogether and ask a general-purpose AI tool to build investor personas from scratch. It is fast and cheap, but it replaces real responses with assumptions.

At Fundamental Group, our view is that the question is not whether to use synthetic research, but what you choose to make synthetic. Applied to respondents, it is risky. Applied to the design and analysis of research built on real investor data, it is a genuine accelerator.

Where synthetic research fails: synthetic respondents

Synthetic respondents replace human respondents with AI-generated primary data. An AI model can only recombine what it has already seen, so it smooths away the outliers and edge cases, which is often where the most valuable insight sits. Two examples show the limits of this approach.

Failure mode 1: the ESG turn

Respondents built synthetically on the consistently positive ESG sentiment of 2021 and 2022 would have projected continued adoption. What followed was more complicated: political backlash in the US, regulatory recalibration in Europe through the SFDR overhaul and fund-naming rules, and a more selective investor. A model anchored to the earlier pattern had no basis to anticipate that shift. Continuous tracking of real investors shows a turn as it forms, rather than after it has been baked into your assumptions.

Failure mode 2: a product with no history

Synthetic respondents cannot replicate responses to something that has not happened before. Germany's Altersvorsorgedepot, a new ETF-based pension scheme taking effect on 1 January 2027, is a clear example. Opening retirement investing to first-time savers through brokers, banks, and insurers, it creates an entirely new product category. There is no behavioural history to train on, so no dataset can predict adoption or sentiment. Only real signals, gathered from the moment the product launches, will show how people actually behave.

Where synthetic research works

Survey design and dry runs

This is the least controversial use and the most underrated. Synthetic respondents can pressure-test a questionnaire before it reaches real people, flagging ambiguous wording, leading questions, broken routing and fatigue points. Nothing here depends on the synthetic answers being accurate, because you are testing the instrument, not the audience. It saves fieldwork budget and protects scarce human respondents, such as senior intermediaries and institutional buyers.

Analysis

Synthetic analysis makes it faster to extract, segment, interpret and stress-test data, but only when that data is anchored in human responses. Aureum, Fundamental's research division, has spent more than two decades building relationships across over 12 global markets to gather survey data from financial intermediaries and institutional investors. That feedback is the foundation. We then apply synthetic analysis to surface patterns faster.

Dynamic personas

Personas built from generic AI assumptions are static documents that date quickly. Personas built from human evidence can become living models, updated as new research arrives. They can be cut into specific groups, such as a UK independent financial adviser with an active ETF tilt, and used to test messaging before campaign spend is committed.

Two modes matter here. When queried about what is known, a profile answers only from verified survey data, with no creative gap-filling. When asked how it might react to something new, the output is labelled as an inference. Every persona should also show how many real respondents sit behind it, so users can judge how much weight it can bear. Results then feed back into the model.

Synthetic focus groups

Because each group's psychographic preferences are already established from real data, asset managers can simulate reactions to new messaging, branding or product concepts ahead of a live campaign. This gives marketers a faster, lower-cost way to pressure-test creative thinking. Outputs should be flagged as synthetic and reviewed by a researcher before they shape a final decision, and the strongest ideas should still go to real investors before launch.

Which parts of investor research should be synthetic?

Five questions to ask any research provider

1. Where does the underlying data come from, and how recent is it?
2. How many real respondents sit behind each segment?
3. How has the output been validated against real fieldwork?
4. How are synthetic outputs labelled in client and internal materials?
5. How does the approach handle data protection and the EU AI Act?

Conclusion

Synthetic research is here to stay for asset management marketers, but not every form deserves your trust. Used to manufacture respondents from scratch, it strips out the behavioural shifts that only real investors reveal. Used to design, analyse and extend data gathered from investors themselves, it sharpens and speeds up research and gives marketers answers they can act on with confidence.