Connecting video analytics to your POS: what to know before integrating

What combining footfall and receipts tells you

A video counter tells you how many people walked into a store. The till tells you how many receipts were issued and for how much. On their own, each figure tells half the story: a busy day with few receipts raises very different questions from a quiet day with excellent conversion.

Bringing the two together lets you calculate combined indicators, the best known being the conversion rate: the share of visits that end in a purchase, estimated by dividing the number of receipts by the number of entries. Other combinations are possible too: average basket relative to footfall, sales per visitor, or the number of people queuing compared with the number of open checkouts.

These indicators are useful for comparing time slots, days or stores, and for checking the effect of a change: a new layout, extra staff, a local campaign. They remain estimates, and their quality depends on both sources and on how they are matched.

Exports, APIs and sync frequency

The first technical question is simple: how do you get the till data out? The answer depends entirely on the software in place, which is why every integration is assessed individually. The most common situations are:

• an API provided by the POS vendor, allowing receipts or aggregates to be read automatically;
• a file export (CSV, Excel), scheduled or manual, dropped at regular intervals;
• access to a database or an existing reporting tool, where the vendor or provider allows it;
• no usable interface at all, in which case manual matching remains possible during a pilot but is not viable at scale.

Choosing the right level of detail

To calculate a conversion rate, you rarely need item-level detail or customer identities. An hourly aggregate (number of receipts, revenue) is usually enough. It is also the leanest choice from a data protection standpoint: the less you collect, the less you have to protect.

How often to sync depends on the use. A daily refresh is fine for trend analysis; near real-time matching only makes sense if an immediate decision depends on it, which is rarely the case for this kind of indicator.

Finally, be clear about who governs access: who holds the API credentials or manages the export, what your contract with the POS vendor allows, and who gets notified if the data feed stops. These may sound like administrative details, but they decide whether the indicator keeps working over time.

Time alignment and scope

This is where most errors hide. Two reliable sources can produce a misleading indicator if they do not measure the same thing at the same time. Points to check:

• clocks: the POS system and the video system must be in sync, ideally via a shared time server;
• natural lag: a customer walks in at 11:50 and pays at 12:10; over short time slots this skews the ratio, so it is better to work with wider slots or whole days;
• staff: employees using the customer entrance inflate counts; depending on the setup, certain time windows or doors can be excluded, but individuals are not identified;
• groups: a family of four creates four entries and one receipt, which mechanically lowers the apparent rate;
• which tills count: click-and-collect orders, returns, checkouts in an annexe; you need to decide what to include.

Combined indicators and source labelling

On a dashboard, every indicator should make clear where it comes from. We use three categories: video-only indicators (entries, occupancy, queue length), indicators from a third-party system (receipts, revenue) and combined indicators that rely on both.

This labelling is not a detail. If the conversion rate falls, the cause could be a genuine drop in purchases, a rise in footfall that did not convert, a camera outage or an incomplete till export. Showing the sources side by side lets you diagnose the issue quickly instead of reacting to a number whose origin you cannot trace.

It also helps to keep a data quality flag for each period: camera offline, POS file missing, partial day. An incomplete day should be flagged rather than quietly folded into the averages.

When reading these indicators, trends tell you more than a single day: comparing a Saturday with previous Saturdays, or a store with itself before and after a change, is usually far more meaningful than a raw figure measured against a theoretical target.

Feasibility study and limits

Before integrating, a feasibility study checks three things: how reliable video counting is at your entrances, validated against manual counts; whether till data is available and of good quality; and whether the combined indicators actually support your decisions. It is often wise to start with a few weeks of manual matching before automating anything. That phase shows whether the combined figures raise useful questions for store managers, and whether the effort of a full integration is justified.

A few limits to keep in mind: the conversion rate is a ratio sensitive to both sources, it says nothing about why people did not buy, and it cannot be compared directly between stores with different entrances, staff flows or formats. On the data protection side, the approach relies on aggregated counts, with no facial recognition and no attempt to link a filmed person to a receipt.

To find out more, see our retail industry page, along with our Solutions and Custom Solutions pages, which explain how we assess integrations with existing systems.