Synthetic Panels for Government and Public Services

Jason Duke, Founder, Kronaxis

Tag: Public Sector

A public body cannot simply guess how people will react to a policy, a service change, or a piece of communication. It has a duty to consult, and often a legal one, and that duty falls hardest exactly where consultation is most difficult: the residents who do not answer surveys, the groups a headline sample under counts, the people whose response only shows up once a change has already landed on them. Traditional research is at its weakest and most expensive precisely where a public body most needs to hear. A grounded synthetic panel is a way to widen the room before you commit, not to replace the people in it.

The representation problem is a public sector problem

The commercial world can sometimes live with a research sample that speaks mostly for the comfortable majority. The public sector cannot, because the people a sample misses are frequently the people a policy affects most. This is also, as it happens, the hardest case for the current generation of artificial respondents. A Stanford group found this year that simply using a larger model does not close the gap on underrepresented populations; it stalls there. We wrote about why in Scale Will Not Fix Synthetic Research. The lesson for a public body is blunt: a synthetic panel is only worth having if it was built to represent the people at the edges, not the average in the middle.

What a synthetic panel lets a public body do

Used honestly, a synthetic panel is a fast, cheap way to rehearse. Before a formal consultation, it can surface the questions that will land badly and the groups who will read a proposal differently from how it was written. Before a service change, it can test whether the people who rely on that service most will understand and accept it, or quietly fall through a gap. Before a public communication goes out, it can show where the same words reassure one community and alarm another. None of this replaces real consultation. It makes the real consultation sharper, cheaper to run, and far less likely to discover an avoidable problem after the decision is made.

Grounding, not guesswork

The value of any of this rests entirely on how the panel is built. A thin persona over a general model tends to answer as the internet's average, which is worse than useless for a public body, because it launders a majority default as if it were the voice of everyone. Our panels are grounded. Populations are weighted to census structure, so the mix of people reflects the place, not the platform. Each persona carries a measured disposition, our DYNAMICS-8 profile, and cultural grounding drawn from established social science rather than from whatever a web corpus happened to contain. And because we support cross cultural research across twenty countries, the same rigour applies to a diverse population as to a uniform one. The point is not to invent opinions. It is to represent difference faithfully enough to be useful.

Evidence it changes behaviour, not just opinion

A public body should be sceptical of any tool that claims to model people, and rightly so. So we hold our own work to the standard we would want applied to us. In a choice experiment registered in advance, with 117 real people, matching the character of content to a reader's own disposition raised conversion from nineteen to thirty one per cent, and the signal came from the deeper part of disposition that people never consciously report. The analysis was fixed before any data was collected, so the result could have failed and did not. We explain why that discipline matters in what preregistration actually proves. It is direct evidence that grounded disposition changes what people do, which is the only thing a public body ultimately cares about.

Where it belongs, and where it does not

We do not sell magic, and we would not want a public body to buy it. A synthetic panel is a tool for hypothesis generation, rapid screening, and rehearsal: it narrows the field, surfaces the dynamics behind a reaction, and lets a team test many versions of a proposal in a morning. On the highest stakes decisions it belongs alongside real consultation, statutory duties, and lived experience, never instead of them. Anyone promising a synthetic public that perfectly replaces engaging with real citizens is selling a shortcut that does not exist. What we offer is narrower and real: a grounded way to prepare, so that when you do consult, you consult well.

How to work with us

We build source available software in the United Kingdom, and we are set up to start small. The sensible first step is a scoped pilot on a real question you are already facing, run against a panel built to represent your actual population, with the reasoning behind every synthetic response open to inspection rather than hidden. If it earns its place, it scales from there; if it does not, you have lost a pilot, not a programme. You can read more on our government and defence page, or start a conversation directly.