It helps to separate three very different kinds of system:

  • Anonymous counting and attention measurement: Software counts how many people pass or look towards a screen, without identifying anyone. This is the most common and least intrusive approach, though it can still be subject to privacy law depending on how images are processed.
  • Estimated attributes: Some software estimates age range or gender from a face. These estimates are often inaccurate, can be biased across different groups, and are regulated or restricted in some places.
  • Identification: Facial recognition or fingerprints to recognise specific people. This is highly sensitive personal data, heavily regulated and generally inappropriate for advertising.

Claims that signage can reliably detect mood or emotion, or infer language preference from a face, are not well supported. Emotion-recognition is widely criticised as unreliable and is restricted in some jurisdictions. More reliable ways to personalise content include:

Approach Example
Time of day Breakfast offers in the morning, dinner offers in the evening
Location context Different content near the entrance and at the checkout
Viewer choice A language button on an interactive kiosk