Retail Analytics

What retail analytics built on cameras and POS can actually answer

Most retail analytics runs on transaction data alone. That tells you what was bought, by whom and when — everything that happened at the till. It cannot tell you anything about the people who did not buy, because they never appear in the data. Adding camera data changes the questions available, not just the volume of them.

Questions a till-only system cannot answer

• How many people came in versus how many bought. Conversion rate needs a denominator, and the denominator is not in your POS.
• Where people stopped and then left. An aisle with high dwell and low purchase is a specific, fixable problem — usually price, stock or signage — but a sales report shows it only as an absence.
• Whether a queue cost you a sale. Walkouts leave no transaction. They are invisible to every system that starts at the scan.
• Whether the shelf was actually full. A line that stops selling looks identical in sales data whether demand fell or the facing was empty.

A footfall counter is not this

Door counters answer “how many”. They do not distinguish a customer from a delivery driver or a member of staff stepping out, they cannot say which part of the store someone reached, and they cannot connect an entry to a transaction. If the question is “how busy were we”, a counter is sufficient and much cheaper. If the question is “which aisle is losing people”, it structurally cannot help.

What the numbers do not tell you

• Correlation, not cause. Dwell time rising after a layout change may mean the display works, or that people cannot find what they came for. The data narrows the question; it does not answer it.
• Seasonality will outrun most interventions. Comparing a week to the week before, rather than to the same week last year, produces confident conclusions about nothing.
• Store layout changes invalidate historical comparison. Zones are defined against a floor plan. Move the fixtures and the series breaks — which is worth knowing before you build a year-on-year report on it.

What it needs to work

• Cameras with a usable view of the floor, not only of the entrance and the till.
• POS and camera clocks in sync. Joining a visit to a transaction is a temporal join; drift breaks it silently rather than loudly.
• A defined zone map that matches the actual floor plan, and gets updated when the floor changes.
• A question you want answered. Dashboards that exist to be looked at get looked at twice. The ones that survive answer a decision somebody has to make weekly.

Comparing stores, departments and shifts

The comparison is usually more useful than the absolute number. One store’s conversion rate means little on its own; the same figure against eleven sibling stores, or against the same department last quarter, is a question worth acting on. The caveats are the same ones that apply to any comparison: stores differ in catchment, size and staffing, so a gap is a prompt to go and look rather than a verdict on a manager. Comparing a department against itself over time is usually safer than comparing two stores against each other.

Privacy

Counting and movement analysis is behaviour-based and does not identify individuals — no facial recognition, no biometric matching. Analytics on shoppers still carries signage and disclosure obligations in most jurisdictions, and that is a question for your own counsel rather than a vendor.

If loss rather than conversion is the reason you are here, the same camera and POS data feeds loss prevention, and our shrink calculator works out what your inventory variance actually is before any system is involved.