The Synthe Suite
The Synthe Suite applies a single approach across five domains: construct a population of reasoning agents calibrated to a real demographic, put a question to them, and read both the distribution of responses and the reasoning behind each one.
The engine
Most tools built on large language models ask the model for a summary judgement. The Synthe Suite does something different. Each agent holds a structured set of beliefs, desires and intentions, calibrated from demographic and cultural data, and deliberates over them before responding. What comes back is not a single averaged answer but a population of individual positions, each with a reasoning trace attached.
The practical consequence: when a result surprises you, you can open it and see why a particular segment reached it. A result you cannot interrogate is a result you cannot act on.
Synthe Suite platforms are designed to work with expert judgement, not around it.
A traditional expert panel gives you depth, accountability and domain intuition that no simulation reproduces. What it does not give you easily is fifteen segments tested in an afternoon, or the same question re-run after you change one variable, or coverage of a demographic you have no panel access to. That is the gap these platforms fill.
The intended pattern is straightforward: use the simulation to explore the space quickly and narrow it, then commit fieldwork budget to the two or three questions that survived. The output is an input to expert deliberation.
Eight studies. 4,800 participants.
Platform output has been assessed against published empirical research across eight validation studies covering 4,800 participants in total. The criterion is directional convergence — whether the simulation identifies the same direction of effect as the validated study, and ranks options in the same order.
We report it that way deliberately. Directional convergence is what the method supports and what the studies measured. Where a study showed divergence, that is part of the corpus too; knowing the boundary of a method is a condition of using it responsibly.
Sovereignty
For public-sector work, regulated industries and any organisation with a data-residency requirement, this is usually the first question asked and the last one that gets a satisfying answer. Ours is on the record.
The five platforms
Digital twins of real demographics, calibrated on Eurostat and World Values Survey data across 71 geographic profiles in 11 countries. Virtual focus groups, message tests and policy simulations in days rather than months, then take the two or three questions that survive into fieldwork.
BDI profiles extracted from psychologist interviews and coach diaries, held in a knowledge graph of more than 1,400 nodes. Coaches can rehearse a difficult conversation against a simulation of the athlete before having it for real.
Virtual voter panels for testing how a message lands across segments before it is committed to, and for mapping a political landscape where conventional polling is too slow or too coarse.
Competency and disposition profiling applied to team formation: which combinations of people are likely to work, where the friction will sit, and what a group is missing.
Profiling support for jury selection and trial strategy, modelling how different juror compositions are likely to receive an argument.
Across the five platforms the same pattern recurs: a decision where the population that matters is larger, more varied or less reachable than the research budget allows you to consult directly. Consumer and voter segments, teams that do not yet exist, jury pools, athlete psychology. In each case the simulation widens what you can examine, and the expert keeps the decision.
iB3