Demographics from a masked shop floor

01 — The challenge

Where it started.

The retailer’s predictive analysis depended on demographic data the cameras could no longer capture: masks defeated the existing recognition, and customers with carts could not be tracked reliably.

Without that input, the analysis was being run on a shrinking share of the shop floor.

02 — What we built

The work itself.

Detectron 2. Object detection covering customers and carts, so movement through the store can be followed rather than sampled.

A custom CNN for facial features. A model trained on the features still visible when a mask is worn, inferring age and gender group rather than identity.

Demographics per location. Results tied to where in the store they were observed, which is what makes them actionable for merchandising.

03 — Results

What the client has now.

Demographic capture restored under masked conditions. Trends and patterns identified for marketing and product decisions. Shopping experience personalised by age and gender group, per location.