PRODUCT · SYNTHETIC POPULATIONS
Simulatable environments, not augmented personas.
Populations coherent at the scale of several million individuals, calibrated on public and proprietary data, that you can interview, evolve, and expose to successive scenarios.
THE FUNDAMENTAL SHIFT
A population is no longer a marketing target: it becomes a simulatable environment.
In classic research, a population is an object you segment and measure to extract aggregate indicators: purchase intent, preferences, brand perceptions. That model has produced decades of useful insights, but once described, the population stays frozen. There is no way to make it react to a new scenario, or to make it age.
With synthetic populations at scale, the logic inverts. Each synthetic individual carries social, economic, geographic, behavioral, cultural, professional and attitudinal dimensions. That richness makes it possible to submit the population to a scenario, observe differentiated reactions by typology, identify tipping points, and project trajectories over several months or years.
It is a simulated human system, with its internal diversity, its tensions and its critical windows. That quality transforms the nature of the analysis: you explore a possible future rather than measuring a state of opinion.
HOW THEY ARE BUILT
Rigorous calibration, controlled granularity, structural guarantees.
Calibration on public data
Each synthetic population is calibrated on the public data relevant to the decision: INSEE at IRIS level for territorial populations, sector data published by regulatory authorities, national and European surveys, public studies from research organizations. This calibration ensures the population's statistical coherence with the demographic and behavioral reality of the studied perimeter.
Enrichment with proprietary data
When the client holds relevant proprietary data, customer segments, consumption histories, internal panels, loyalty cards, anonymized receipts, it is added to the calibration to refine the structuring of typologies. This integration strictly respects our confidentiality guarantees, with no transfer of personal records into our system.
Structuring into granular typologies
A synthetic population is not homogeneous. It is structured into typologies crossing several dimensions: typically 10 to 30 typologies per simulation, depending on the case's complexity. This granularity makes it possible to observe differentiated reactions by subgroup and reveal micro-patterns invisible to aggregate analysis. It is what allowed our cases to identify that "4% of the population concentrate 78% of the political risk" or that "3 sensitive typologies concentrate 62% of the false positives generated by a predictive AI model".
An absolute guarantee: no personal records
No personal records ever enter our system, at any point in the process. The proprietary data clients provide is anonymized before integration and used exclusively to refine the statistical calibration of typologies. This guarantee is structural, documented and auditable: it is a precondition for the trust of the risk and legal teams we work with.
WHAT IT CHANGES
Classic panels and synthetic populations do not serve the same decisions.
| Classic panels and samples | Synthetic populations | |
|---|---|---|
| Typical scale | A few hundred to a few thousand individuals interviewed | Several hundred thousand to several million coherent individuals |
| Analytical granularity | Broad segments, typically 3 to 8 typologies | Fine-grained typologies, 10 to 30 crossing several dimensions |
| Temporal dimension | A snapshot at time T, hard to reproduce over time | An evolving population: aging, exposure to successive scenarios |
| Dimensional richness per individual | Declarative attributes limited by the questionnaire | Articulated social, economic, behavioral and cultural dimensions |
| Reusability | New fieldwork for every new question | A preserved population, continuously queryable via API |
HEALTHCARE
HEALTHCARE
Reviving MMR vaccination coverage across twelve lagging territories.
A national health authority was preparing a plan to revive MMR vaccination coverage in its most fragile territories. Our system rebuilt a synthetic population of 2.4 million parents of children aged 12 to 24 months, structured into 18 vaccine-hesitancy typologies crossing care experience, institutional trust, media exposure and territorial political context. This granularity revealed that the firm national campaign initially considered would have worsened the situation in 4 of the 12 territories: an insight out of reach of classic panel studies, and decisive for the final political arbitration.
Read the full case →INFRASTRUCTURE RATHER THAN A DELIVERABLE
The population is not frozen: it lives, evolves, and can be replayed.
In the classic research model, the interviewed population only exists for the duration of the fieldwork. Once the report is delivered, it disappears. Any new question about that same population means new fieldwork, a new budget, a new calendar. This sequential economy constrains the very way organizations ask their questions: you hesitate to ask something you will not be able to extend, you give up exploring scenarios you would nonetheless find useful.
With synthetic populations, that economy changes. Once built, the population can be preserved, exposed to successive scenarios, aged, submitted to new economic or regulatory constraints, compared across several possible futures. It becomes an infrastructure of simulation available for the duration of the decision it informs, and often beyond.
This evolvability transforms the uses. A marketing team can continuously test the evolutions of its offering. An HR team can simulate several configurations of a transformation plan before committing to one. A public authority can project a policy's effects across several time horizons. A risk team can replay a crisis scenario under several configurations. The synthetic population stops being a one-off study object: it becomes an instrument of continuous steering.
A synthetic population adapted to your decision.
Every simulation is calibrated on the precise parameters of the decision it informs. The dynamic agents scope with you the construction of a population adapted to your situation: geographic perimeter, structuring dimensions, relevant typologies, required granularity. From initial brief to the delivery of the inquiry dossier, allow 20 to 30 minutes.
See the case library →Read the guide: how to create a synthetic population →