retail analytics

Aggregate demographic analytics, with the uncertainty stated plainly

In short

Ayonix gender and age analysis estimates apparent age band and gender from appearance for audience analysis. These are estimates, not verified identity or self-identified gender. This site publishes the capability only in aggregate form, with uncertainty reported, sensitive-attribute inference excluded and an anonymous counting alternative offered.

Request a pilot Send a sample video

Conceptual illustration of shop-floor movement used for heatmap analysis. AI-generated conceptual image, not a customer deployment.

Scope

What this analytic does, and what it does not

What it detects

Estimated apparent age band, reported in bands rather than a single age
Estimated apparent gender, reported with an explicit uncertain category
Aggregate composition of an audience over a reporting interval
Change in aggregate composition between comparable periods

What it does not guarantee

A person’s actual age
A person’s gender identity — the estimate is from appearance only
Any accuracy guarantee for an individual observation
Any inference about race, ethnicity, religion, health, disability, sexuality or any other sensitive characteristic — these are not produced and must not be requested

The problem

What customers are actually dealing with

These are the situations that lead teams to look at this analytic in the first place.

An audience report is needed for a campaign and the only alternative is a clipboard survey with a tiny sample.
Vendors present age and gender estimates as if they were facts, and buyers cannot tell how uncertain the figures are.
Legal and privacy teams reject demographic analytics outright because the request arrives with no safeguards attached.
Nobody has asked whether an anonymous count would answer the same business question.

How it works

Detection workflow

Every event carries the rule that produced it, so an operator can see why they were alerted.

1

Detect

People are detected in the camera view; no template or identifier is stored.
2

Estimate

Apparent age band and gender are estimated with a confidence value for each observation.
3

Discard

Low-confidence observations are assigned to an explicit uncertain category rather than forced into a class.
4

Aggregate

Observations are aggregated per interval; intervals below a minimum count are suppressed.
5

Publish

Aggregate composition with uncertainty bands is written to the report. Per-person records are not retained.

Aggregate demographic analytics workflow diagram

Configuration options

Age bands used for reporting, aligned to the business question rather than to a false precision
Confidence threshold below which an observation is classed uncertain
Minimum observations per interval before any figure is published
Reporting interval length
Whether the anonymous counting-only mode is used instead of estimation
Explicit exclusion of any sensitive attribute from configuration and output

Alert quality

How irrelevant and duplicate alerts are reduced

No video analytic eliminates false alerts. These are the mechanisms that reduce them, and the residual rate is measured on your own cameras.

An explicit uncertain category is required; forcing every observation into a class manufactures false confidence.
Minimum-observation suppression prevents small samples being read as findings and reduces the chance of singling anyone out.
Reporting bands with confidence intervals communicates what the method can actually support.
Aggregation before storage means no individual demographic record exists to be misused later.
Estimation accuracy varies with camera angle, lighting and the population observed, and must be characterised on your own site.

Prerequisites

Camera requirements and environmental limits

Camera suitability decides more of the outcome than any software setting. These are assessed per camera before commitment.

Camera requirements

A near-frontal view of faces at adequate resolution — oblique and overhead views are unsuitable for estimation
Consistent, even lighting on the face; strong side-light or backlight degrades estimation sharply
Sufficient resolution across the observation zone, which is usually a much shorter range than for counting
A camera position that does not capture areas where estimation would be inappropriate, such as washroom approaches
Stable mounting so the observation zone remains consistent across the reporting period

Environmental limitations

Estimation accuracy differs across populations, ages and presentations, and error is not evenly distributed.
Appearance-based gender estimation cannot capture gender identity and will misclassify some people.
Age estimates are bands, and boundary cases are inherently uncertain.
Face coverings, hats, glasses and viewing angle all reduce reliability.
Results from one site do not transfer to another without re-characterisation.

Delivery

Alerts, evidence and where they land

Event and evidence fields

Observation zone and camera
Interval start and end
Aggregate composition by band
Proportion classed uncertain
Observation count for the interval
Suppression flag where below minimum

VMS integration

Aggregate audience composition is a reporting output and is not delivered as a VMS alarm. It is consumed in campaign and audience reporting tools.

See compatibility states →

Deployment options

Edge processing so estimation happens locally and only aggregate composition leaves the site
On-premise processing where signage or entrance cameras already aggregate locally
Anonymous counting-only mode where the business question can be answered without estimation

Compare architectures →

Measurement

Dashboard metrics and pilot acceptance criteria

Acceptance thresholds are agreed with you before the pilot starts. This site publishes no benchmark figures, because they do not transfer between sites.

Dashboard metrics

Aggregate composition per interval with confidence bands
Proportion of observations classed uncertain
Observation volume per interval
Suppressed intervals count
Composition change against a baseline period
Camera availability for the observation zone

Measurable pilot criteria

Agreement between aggregate estimated composition and an independent manual sample on your own site
Proportion of observations classed uncertain, reported openly rather than minimised
Behaviour of minimum-observation suppression at quiet times
Confirmation that no per-person record is written at any stage
Documented review by your privacy or legal function before any production use

Governance

Privacy and governance

Outputs are aggregate only. No individual demographic record, face template or identifier is retained.
Race, ethnicity, religion, health status, disability and other sensitive characteristics are never inferred or reported.
Estimated age and gender are not identity data and must not be used as if they were.
In many jurisdictions this processing requires a documented lawful basis, a data protection impact assessment and clear signage. Confirm requirements for each site before deployment.
An anonymous counting alternative is available and should be preferred wherever it answers the question.
Estimation must not be used to make decisions about an individual, such as service refusal or differential treatment.

Frequently asked questions

How accurate are age and gender estimates?

They are estimates from appearance and their accuracy varies with camera angle, lighting and the population observed. Error is not evenly distributed across groups. This site does not publish a headline accuracy figure because a single number would misrepresent how the method behaves; instead, agreement with an independent sample is measured on your own site.

Can it detect ethnicity or other sensitive characteristics?

No. Inference of race, ethnicity, religion, health condition, disability or any similar characteristic is not produced, not configurable and not available on request. This is a firm limit, not a current gap.

Is this identity data?

No. Estimated age band and apparent gender are not identity attributes and cannot be used to identify anyone. They must not be treated as verified personal data or used to make decisions about an individual.

What if we would rather not estimate at all?

Then use the anonymous counting mode. It reports audience volume and dwell without any demographic estimation, and for many campaign and placement questions it is sufficient. It is the recommended default where the business question allows it.

Keep reading

A controlled pilot establishes what this analytic actually does in your environment, against acceptance criteria we agree before it starts. Send a sample video first if you would rather see an assessment before committing to a pilot.