How to count store foot traffic by hand and work out the conversion rate

The conversion rate formula

Conversion rate is the number of sales transactions divided by the number of visitors in the same period, times 100. In a 2012 study of an apparel retailer's stores, Olga Perdikaki, Saravanan Kesavan and Jayashankar Swaminathan define it as "the ratio of number of transactions to traffic" (Manufacturing & Service Operations Management 14(1)).

Conversion rate (%) = transactions ÷ visitors × 100

Three words in it carry most of the errors. Transactions means receipts, not items sold or revenue: a customer buying four things is one transaction. Visitors means people who came in, counted by a rule you have written down. The same period means the hours in which the visitors were counted, not the whole day's sales against two hours of counting.

Conversion is one link in a chain. Lam, Vandenbosch, Hulland and Pearce split store sales into "front traffic, store-entry ratio, closing ratio, and average spending" (Marketing Science 20(2), 2001): people passing, the share who come in, the share of those who buy, and what they spend. The closing ratio is the conversion rate.

Each link has its own fix. A low entry ratio is a window and doorway question; low conversion is about stock, service, queues and price; a low average value is about range and selling. Sales alone cannot tell you which one moved, which is the case for counting.

Who counts as a visitor

Count each person who crosses the threshold coming in, and decide in advance, in writing, what to do about staff, children, groups and people coming back in. No standard settles these choices. What makes the figures usable is applying the same rule every time, so a change in conversion is a change in the shop and not in the counting.

The physical rule matters as much as the list. The headcount guide sets one trigger, a leading foot crossing a marked line, and that works as well at a shop door as at an event. Watch the line, not faces, and a group of three is three taps.

Entries, exits and the divide-by-two trap

For traffic, count entries; counting exits too is a check, not a second visitor. A shop with one door that counts every movement either way gets two ticks for every visit, so its conversion rate comes out at half the true figure unless the ticks are halved. Halving assumes every visitor passes a counted door twice, which fails where a shop has a second, uncounted door or opens into a mall.

Counting in and out separately gives a free accuracy check: the shop is empty at closing time, so the day's ins and outs should be equal. If the day ends with 702 in and 689 out, 13 taps were missed, about 1.9%, and you know the error is at least that big.

How to count by hand

Stand where you can see the whole entrance without blocking it, count one direction per key, and write the total down at the end of every hour. The method below is built for one person with a phone; a second person on a wide or double door splits it on a fixed line, as at an event.

1. Pick the hours. The same hours every week: at least one quiet and one busy hour, on the same days.
2. Write the rules from the table above on the count sheet, so the next person counts the same way.
3. Stand inside, to one side, with a clear view of the threshold line and not in the way of shoppers.
4. Count entries with one key, exits with another. On the people counter, I counts in and O counts out, and a missed or double tap can be undone.
5. Tally groups on a second counter if you want conversion per group. The multi counter keeps named counters such as "groups" and "children" side by side.
6. Note the totals on the hour, and anything unusual: a coach party, a fire alarm, a queue out of the door.
7. Rotate counters on long sessions. A University of California, Berkeley study of manual pedestrian counts found errors were greater at the beginning and end of the observation period, possibly from unfamiliarity or fatigue.

Expect hand counts to run low. That study, by Diogenes and colleagues, compared counts made with sheets and with clickers against video at San Francisco intersections and found both "systematically underestimated pedestrian volumes", by 8 to 25% (SafeTREC, 2007). A shop door is quieter than a city crossing, but the direction of the error is the point: missed taps make conversion look better than it is.

Matching the count to the till

Take transactions from the till report for exactly the hours you counted, hour by hour, never the day's total. Most point-of-sale systems can report transactions by hour; if yours cannot, note the receipt number at the start and end of each counted hour. Refunds and voids are not sales: leave them out of the transaction count, or at least treat them the same way every week.

Worked example: a Saturday in a gift shop

The shop counts 11:00 to 12:00 and 15:00 to 16:00 on Saturdays. Staff are left out; people and shopping groups are tallied separately. The till report gives transactions for the same two hours.

1. Per hour: 21 ÷ 84 = 25.0%; 26 ÷ 132 = 19.7%.
2. For the sample: add first, then divide. 47 ÷ 216 = 21.8%. Averaging the two hourly rates, (25.0 + 19.7) ÷ 2 = 22.4%, gives the quiet hour too much weight.
3. Per group: 47 ÷ 132 = 35.6%. Neither figure is wrong; report the one you chose and keep to it.

The busier hour converted worse, which is the pattern Perdikaki and colleagues found: conversion falls as traffic rises, and sales rise with traffic but with diminishing returns. In their data, staffing changed how much extra traffic turned into extra sales. An hour where conversion drops as the door gets busy is often an hour where the till queue or the shop floor is short of people.

The mistake that makes conversion look like 74%

Dividing the whole day's transactions by the sample hours' traffic is the easiest error to make, and it produces impossible numbers. If the shop took 160 transactions all day, 160 ÷ 216 = 74%, because 216 is two hours of visitors and 160 is eight hours of sales.

Scaling the sample up is better but still a guess. Two hours of an eight-hour day, times four, suggests 864 visitors and 160 ÷ 864 = 18.5%. That assumes the counted hours are typical of the day. Here the afternoon hour is a busy one, and busy hours tend to convert worse, so the true daily figure may well be higher. Matching hour for hour avoids the assumption altogether.

How big a change is real?

A conversion rate from a few hundred visitors is only good to several percentage points, so small week-to-week moves are noise. Treating each visitor as buying or not, the margin of error at 95% is about 1.96 × √(p × (1 − p) ÷ n), where p is the conversion rate and n the visitors counted.

So a fall from 21.8% one Saturday to 19% the next, on two hours' counting, does not show that anything changed. The same pattern over several weeks, or a move bigger than the margin, does. The approximation is generous to you, because groups make visitors buy together rather than one by one; real margins are a little wider. The arithmetic of sample sizes is set out in the guide to how many people to survey.

When a hand count stops being enough

A hand count is for sampling hours; once you need every hour of every day, or several doors, it is a sensor's job. One person can count a single shop door well for an hour or two. They cannot count twelve hours, every day, at several entrances, and the undercount grows with the flow.

Hand counts still have a use alongside a sensor: they are how you check one. Count the door yourself for an hour and compare. If the sensor and your count differ by more than a few percent, find out why before trusting a conversion rate built on either.

What this guide does not cover

This guide covers counting visitors by hand in a shop and working out a conversion rate from till transactions. It does not compare people-counting sensors, interpret footfall data bought from third-party providers, or cover the privacy rules on camera-based counting, which depend on where you are. Published conversion benchmarks are not used here because the ones we found came from counter vendors with their own definitions of a visitor.

Simple Online Counter is a free, browser-based set of counters that keeps its counts in your browser's own storage. Its people counter counts entries and exits separately and is built for a door; the till report, the counting rules and the judgement about what changed are yours.