Methodology · Ethics & oversight
North AI models how populations respond — never who an individual is. These are the commitments that keep the models accountable, each one verifiable without exposing the model itself.
Synthetic
The 100,000 profiles per simulation are statistical agents fitted to aggregate population parameters — no real, identifiable people.
Aggregate-only
Outputs are distributions over segments. North AI cannot build, store, or target a profile of a named person.
Oversight
Calibration, validation, and release require neuroscientist and data-scientist sign-off — not unattended automated tuning.
Fairness
Systematic error is measured across age, gender, and region, and corrected before a calibration ships.
Synthetic audiences
The audiences in a North AI simulation do not correspond to real individuals. Each of the 100,000 profiles is a synthetic agent whose behaviour is drawn from aggregate population parameters — the statistical shape of how groups attend, react, and share.
Because no profile is built from or linked to a specific person, there is nothing to re-identify and no individual record to expose. The audience is a model of a population, not a directory of people.
Aggregate-only by construction
Every result North AI returns is aggregate by design: distributions over audience segments, confidence intervals, and scenario paths. The platform does not produce, store, or sell individual-level profiles, and it cannot be used to identify or target a named person.
This is a structural property, not a setting. Targeting an individual isn't something we choose not to do — it is something the system is not built to do.
Human + scientific oversight
Model behaviour is reviewed by neuroscientists and data scientists. Calibration, validation, and release follow a documented scientific protocol with human sign-off at each gate — the models are not left to tune themselves unattended.
Bias & fairness
Predictions are evaluated for systematic error across age, gender, and region. Our cross-border calibration cohort lets us screen the same algorithm across markets, surface divergence between them, and correct it before release rather than after.
The science underneath these commitments is trained only in the lab with consenting participants — recruited through a vetted research panel with fair-pay minimums and the panel's own ethics framework — and calibrated against a 300,000-person cohort. See how we capture data →
We can walk procurement, legal, and security teams through our data provenance, controls, and documentation — and onboard a custom cohort if you need one.