Methodology · Data provenance

How we capture the data behind the model.

From the consented people who provide the signal to the synthetic audiences you simulate against, every stage of our data pipeline has a guardrail. Here is the full path — with nothing about the model itself exposed.

01 · Source

Consented data, documented at source

Every neurophysiological and behavioural signal is collected under informed, opt-in consent. Participants are briefed on use and retain the right to withdraw, and we record provenance for each data stream.

No data without consent

02 · Train

The model is trained only in the lab

The neuroscience model is trained exclusively with participants who consented to provide the data, and studied across Latin, English, and US populations so it is not fitted to a single market.

Supervised by scientists

03 · Calibrate

Calibrated against a 300,000-person cohort

Calibration runs through a tracking remote across a cohort of 300,000 people who agreed to take part. The scale and spread let us cross-border screen the algorithms and catch regional divergence before anything ships.

Cross-border screening

04 · Validate

Checked for bias across markets

Predictions are evaluated for systematic error across age, gender, and region. Where a market diverges, the discrepancy is corrected before the calibration is promoted.

Fairness checks before release

05 · Generalise

Translated into synthetic audiences

The calibrated parameters set the rules that 100,000 synthetic audience profiles follow in every simulation — statistical agents fitted to population behaviour, never records of real, identifiable people.

No real individuals modelled

How participants are recruited, consented & compensated

Recruitment

Sourced through a vetted research panel

Lab and calibration participants are recruited through a vetted external research-participant platform — a large, pre-screened pool where people opt in to take part in studies. We do not cold-source, scrape, or repurpose customer data to build the cohort, and participants are matched to the demographics a study requires.

Consent & ethics

Voluntary, informed, and ethics-reviewed

Participation is voluntary and based on informed, opt-in consent: people see what a study involves before joining, are told their neurophysiological and behavioural responses are used to research and calibrate North AI's audience models, and can withdraw at any time. The platform operates an ethics framework for online research, and studies follow its participant-protection standards.

Compensation

Paid fairly for their time

Participants are paid for their time at or above the platform's enforced fair-pay minimum. Compensation is never contingent on the responses they give, so there is no incentive to answer in a particular way.

Cross-border by design

The 300,000-person calibration cohort spans Latin, English, and US populations, so we can run the same algorithm across markets and detect divergence rather than assume one region generalises to another.

Bring your own cohort

If your audience isn't fully represented, we can onboard an additional cohort for your use case and fold it into calibration — so the simulation is screened against the people you actually care about.

Representativeness you can inspect

Cohort availability isn't a black box. Our live audience map shows how many consented participants are available in each country, broken down by age and gender — so you can see the real composition behind a simulation, and where a market is thin, before you rely on it. Explore the audience map →

Reviewing North AI for your organisation?

We can walk procurement, legal, and security teams through our data provenance, controls, and documentation — and onboard a custom cohort if you need one.