People counting has answered the same question for three decades.It measures how many individuals cross a defined boundary within a selected period. That boundary is usually the entrance to a store, shopping centre, terminal or public building.

The practice travels under several names: footfall counting, visitor counting, traffic counting, door counting and, in aviation, passenger or PAX counting. Whatever it is called, the unit is a body crossing a line. The output is a number per hour, day or week.

Footfall analysis is the layer above the count. It interprets that number, usually by combining it with data the business already holds. For example:

• Dividing sales transactions by entries produces a conversion rate.

• Charting entries across the day identifies peak hours and informs staffing.

• Comparing entries with passing traffic produces a capture or drop-in rate.

• Comparing this week with last week shows a trend.

Every one of these measures comes from the same input. That is the source of both the discipline's usefulness and its ceiling.

Better counters still produce a count

People counting instruments have improved steadily.

Manual clickers gave way in the 1990s to infrared break-beam sensors. Thermal imaging followed, using heat signatures to detect people from overhead. Video analytics repurposed camera footage for counting, while stereo vision used two lenses to resolve depth and height.

Each generation improved counting under particular conditions. Each also had characteristic limitations.

• Break-beam sensors can undercount groups walking through together because a beam that is already broken cannot break twice.

• Thermal sensors can lose separation when heat signatures overlap or ambient temperatures approach body temperature.

• Simple video systems can struggle to distinguish people from objects.

• Any single-vantage system can lose people to occlusion when one shopper passes in front of another.

Lin, Seah and Li observed the same issue with simple sensors, noting that single-beam infrared counters can show high error rates when several people cross a monitored area at the same time.

This history explains something otherwise puzzling. Retailers, landlords and airport operators have bought successive generations of counting hardware for thirty years, yet the strategic question they can answer has not moved. Better instruments produced a more accurate version of the same number.

The count was never the limitation. The boundary was.

The ceiling of footfall analysis is structural rather than technical. It is not simply an accuracy problem.

A count measures a boundary. It records that a body passed and holds nothing else: no history before the crossing, no future after it and no way to distinguish two crossings by one person from crossings by two people.

Every metric derived from the count inherits that limitation. A conversion rate built from entries and receipts is not, by itself, an explanation of behaviour. It is one number from the door counter divided into another number from the point-of-sale system. The explanation sits in the gap between them.

The consequences are familiar to anyone who has run a store using this data.

A busy Saturday can depress conversion even when the store sold more to the customers it was built for, because every additional passer-by who stepped inside and left enlarged the denominator.

A campaign that increased entries can look successful whether it attracted buyers or bystanders.

A drop between the shop floor and the fitting room remains invisible because nothing between the two areas was measured.

The number tells you the store was busy. It falls silent on everything that determines whether being busy was valuable.

This is not a criticism of counting. Counting can perform the job it was designed for accurately. The problem is that the job was always narrower than the questions being asked of it.

The visit changes what can be asked

The unit of measurement is now the visit, also called the visitor journey: the path one person took through a location, held as a single anonymous record from entry to exit. Every reported number is a question asked of that record.

Two boundaries come with it, and both matter commercially.

First, a visit stands alone. If a person leaves and returns later, that creates a second visit. Nothing is linked between visits or between locations, and no measure of return frequency is created.

Second, a zone is an area the operator defines and measures. The record holds entry to the zone, exit from it and the time between. It does not observe what a person does inside the zone between those moments.

This is the honest boundary of the measurement, and it should be stated before its capabilities.

Anyone familiar with web analytics already has a useful mental model:

The comparison is not exact, but the structure is useful. Online, it gave businesses a way to understand progression rather than arrival alone. The same structure can now be applied to physical space.

The change does not create one new metric. It means the denominator can describe visits rather than crossings, a drop-off can have an address instead of being inferred from the sales line, and the profile of who entered can be compared with the profile of who bought.

Benchmarking answers a question one location cannot

A figure on its own rarely shows whether it is good.

A fitting room converting at 9 percent to a till has limited meaning in isolation. It becomes useful when compared with equivalent fitting rooms operating between 6 and 14 percent. Benchmarking supplies that context, and it is the capability no operator can build from its own locations alone.

A useful benchmark has three characteristics:

1. It compares shares rather than totals. This reduces the effect of location size and allows differently sized spaces to be compared on equivalent measures.

2. It compares like with like. A fitting room is compared with fitting rooms and a fashion space with fashion spaces, not with a location-wide average.

3. It compares the relevant part of the location. The decision usually concerns which area is underperforming, not whether the whole store is above or below average.

It must also protect the operators represented in the comparison:

• An operator sees a position rather than another operator's value.

• Nobody learns who else is in the comparison group.

• If too few comparable operators exist to produce a safe position, no benchmark is shown.

For landlords and airport operators, this allows tenant performance to be assessed against the wider market rather than only against an estate average. For retailers, it turns an internal figure into market context that can support a board-level decision.

The questions a visit record makes answerable

The purpose of a measurement system is the set of questions it allows an operator to ask.

A door count supports two basic questions: how many came, and was that more than last week?

A visit record supports a wider set:

• Are we moving in the right direction, and on which measure?

• Who is coming, and how many are the visitors that matter?

• What do visitors do after entering?

• At which point do we lose the visitors we wanted?

• How much of the available value are we capturing?

• When is demand present, and is it the right demand?

• How do our locations and spaces compare on value rather than volume?

• What is working, what is not and what should change next?

• When something changes, did it work?

Every one of these questions can be asked of the same record. That is the structural change underneath the new measurement model.

The questions can expand without replacing the instrument because they are queries against one object rather than separate systems bolted together. This reverses the pattern the category followed for thirty years, when every new question appeared to require new hardware.

How a filter can manufacture a finding

A richer record creates a risk that a door count did not have. With one number, there is little to slice. With a visit record, there is a great deal, and careless filtering can produce confident answers that are not true.

The trap is filtering on an outcome and then interpreting the relationship as causal.

For example, a retailer might filter for the busiest hours, observe lower conversion and conclude that crowding caused it. But the visitor mix may also be different at peak times. Traffic, time and sales can move together, so questions involving them may require controlled analysis rather than a dashboard filter.

The same discipline applies to labelling.

• A figure that is derived rather than measured should say so wherever it appears.

• A period in which the instrument was not verified should appear as absent rather than as a clean number.

Neither practice is glamorous. Both separate a figure a commercial team can defend in a rent review or tender from one that collapses under the first challenge.

The language has to move with the instrument

The practical obstacle is smaller and more ordinary than the technology. It is vocabulary.

A team that has worked with door counts for a decade already shares a set of familiar words. Most survive when measurement moves to the visit, but some require more precise definitions.

Conversion changes when the denominator describes visits rather than crossings. Dwell changes when time belongs to a visit within a zone rather than being inferred from activity in an area. Accuracy changes when it is supported by evidence from the operator's own conditions rather than a percentage in a brochure.

Agreeing on those definitions before the first report arrives is one of the simplest steps an operator can take.

The companion guide, From door to visit: The vocabulary of a measured location, explains the terms and how they relate to one another.

The account of the development of counting hardware, from manual clickers through infrared break-beam sensors, thermal imaging, video analytics and stereo vision, and the characteristic limitations of each, reflects general industry practice rather than a single source. Accuracy figures from competing vendors vary widely and are not reproduced here.