Customer Behavior

What behaviour analytics measures in a physical store

A door counter answers one question: how many people came in. Behaviour analytics answers the questions that follow from it — which aisles they actually walked, where they stopped, what they picked up and put back, and where the journey ended without a purchase. Those are different measurements, and only the second kind tells you where to change something.

The four signals worth having

• Path. The routes people actually take, as opposed to the route the planogram assumes. The gap between the two is usually where a category is underperforming.
• Dwell. How long people stand in front of a fixture. High dwell with low conversion is the most useful signal in the set: attention without purchase points at price, stock or clarity — not at footfall.
• Interaction. Pick-up and put-back. A product handled often and bought rarely is a different problem from one nobody touches at all.
• Queue state. Where and when people abandon. Abandonment at the till is the one behaviour that is unambiguously costing a recorded sale.

How this works without identifying anyone

Every signal above is a property of a track — an anonymous shape moving through frames — not of a person. No facial recognition is used and no biometric identity is created, which is the whole reason this data can be collected in a normal store under a normal privacy notice. The practical consequence is that a track ends when someone leaves the camera’s view; it is not stitched to a returning visitor next week.

What it cannot tell you

This is the part most vendor pages leave out, and it is what decides whether the data is worth acting on:

• Not who. Anonymous tracking cannot segment by age, sex or income. Systems that claim to are inferring it from appearance, which is both a biometric processing question and a guess.
• Not across visits. Without identity there is no repeat-visitor history. Frequency questions need loyalty or card data, not cameras.
• Not what it means. A dwell spike is not a preference. Someone standing still may be choosing, confused, on the phone, or waiting for somebody. The measurement is real; the interpretation is yours to test.
• Not outside camera coverage. Every conclusion is biased toward the aisles you happen to have cameras on. Coverage gaps look like low activity.

What to have ready before it is useful

Behaviour data answers a question; it does not supply one. Before a deployment is worth running, you want a specific hypothesis (“this end cap underperforms”), a before-state you have already measured, and a change you are actually able to make. Without the third, the data becomes a dashboard nobody opens.

Two things this is not, because a share of the traffic to this page is looking for them: it is not a consumer-data marketplace — nothing here is purchased or sold about individuals — and it is not a market research report on the behaviour-analytics sector. It is instrumentation for your own stores. The nearest adjacent things we publish are retail analytics for the sales-side view and loss prevention for behaviour that costs stock rather than sales.