
Indivd
•
•
Intro
A Central European shopping centre recorded 98.10% absolute counting accuracy and 98.62% anonymous matching accuracy, each checked against manual review.
A large, multi-level shopping centre in Central Europe had two separate quality assurance checks carried out between August and October 2026.
The counting check covered 78 counting lines and 31,732 manually counted crossings, with samples collected on three trading days: 1 August, 25 September and 21 October 2026. It reported 98.10% overall absolute accuracy.
The matching check covered 20 zones and 10,500 visitor movements, with samples collected on 14 and 27 October 2026. It reported 98.62% overall anonymous matching accuracy and a 60% capture rate.
Both reports state a 99.99% confidence interval.
The centre
Two questions behind every movement figure
A shopping centre reads its visitor data on two levels. The first is how many people pass through each entrance, walkway and tenant doorway. The second is how visits connect: whether visitors to one area also reach another, and how that changes after a campaign or a layout change.
The two levels depend on each other. Movement between zones is built on top of the counts, so an error in either one passes into every figure that combines them. A miscount at a line changes the size of a flow. A wrong match between zones can suggest a connection that did not happen, and a missed match can leave a real one unrecorded.
The centre in this report had both levels checked, in the same building and in the same season. Its tested areas include grocery and home anchors, fashion and footwear shops, a drugstore, internal walkways, a food court and a meeting place.
The check
Two layers, one manual standard
Both checks use the process set out in the help center article How Indivd verifies data accuracy. The customer decides where, when and how much is tested, and reviewers only ever see masked material.
Counting. Short video samples were converted into full-body masked video, where every visitor appears as a coloured box. Reviewers counted each time a box crossed a counting line, and Indivd's automated count for the same footage was compared with that, frame by frame.
Anonymous matching. Indivd QA reviewers examined pairs of masked images showing visitor movements between two zones and decided, for each pair, whether the images showed the same anonymous visitor. Indivd's automated decision was then compared with the manual one.

What a reviewer sees during counting QA (example from another location, not the centre in this report). Each visitor is a full-body masked box, and each green line is a counting line, labelled with its ID and its running count. The reviewer counts the boxes that cross each line.
Masked fragments are deleted automatically after review. What is collected, how long it is kept and who can access it is documented in the DPIA Guide: Quality assurance and the Anonymization Policy.
This assessment was carried out by Indivd's QA team. As a general process, customers can choose where, when and how much to test, and can run the check themselves or appoint a third party.
The count
Every line checked against a manual count

The counting check reported 98.10% overall absolute accuracy across 78 counting lines, at perimeter entrances, internal walkways, zone boundaries and tenant doorways.
DEFINITION
Absolute accuracy is 100 percent minus the absolute error rate, where a missed entry and a missed exit count as two separate errors that never cancel.
The verification covered:
78 counting lines
31,732 manually counted crossings
104 review hours, accumulated across counting lines
Three trading days: a Saturday in August, a Friday in September and a Wednesday in October 2026
Spreading the sample across three different days puts weekend and weekday trading under the same standard.
Individual lines were also assessed. 73 of the 78 lines reached 97% or above. The report flags five below that, each with its reason: too few crossings for a stable percentage, or a counting point set far from its camera. Low-traffic lines are more volatile by nature, because one missed crossing is a larger share of a small total.
The matching
98.62% across 20 zones, every zone at 95% or above

The matching check reported 98.62% overall anonymous matching accuracy across 20 zones, from anchors and shops to walkways, the food court and a meeting place.
The verification covered:
20 zones tested
10,500 visitor movements assessed
Masked image pairs reviewed manually
Two sampling dates, 14 and 27 October 2026
Zone results ranged from 95.33% to 100%. All 20 zones reached at least 95%.
What the numbers mean
A checked basis for footfall and for movement
The overall figures in both reports average across lines and zones, weighted by the traffic each one carried, so a busy entrance counts for more than a quiet side door.
For the centre, the two results give a documented reference for both layers of its visitor data. Entrance and walkway counts can be read knowing how closely they matched a manual count. Connections between zones, such as how far visits extend from an anchor into the rest of the centre, can be read knowing how often the matching behind them agreed with manual review.
The routine QA shown here is Indivd's own operating cadence, with over 500,000 manual assessments to date. It sits alongside an independent audit: KPMG once audited Indivd's counting against manual ground truth at a major retailer's store, with an average accuracy of 99.11%.
More QA reports
98.5% overall absolute counting accuracy across 62 shopping-centre lines
97.72% average anonymous matching accuracy across 18 shopping-centre zones
97%+ absolute accuracy in people counting across 35 fashion stores
Common questions
How accurate is people counting in shopping centres?
It depends on the system and on each measurement point. Many providers quote 99% or higher from controlled tests, but a shopping centre has dozens of entrances, walkways and tenant doorways, each with its own light, ceiling height and crowd density. The figure that matters is one verified in the centre itself, line by line, against a manual count.
How is people counting accuracy measured?
A sample of real visitor activity is recorded, and the system's count is compared with a manual count of the same footage. The comparison is made for each counting line, and the result is reported as an overall figure alongside the line-level results, the sample size and a confidence interval.
What is the difference between absolute and net accuracy?
Net accuracy compares totals, so a missed entry and an extra entry cancel out and the result looks better than the counting behind it. Absolute accuracy counts every miss and every extra as a separate error, which makes it the stricter measure and the more useful one for comparing hours, days and locations.
How do shopping centres measure movement between zones?
Movement between zones is measured by anonymous visitor matching: associating observations from different areas with the same visitor without attaching a personal identity. It supports measures such as cross-visitation, which shows how often visitors to one area also reach another.
How often should footfall accuracy be checked?
After installation, at regular intervals, and after any significant change to camera placement or configuration. Accuracy can drift without anyone noticing, so a single check at setup is not enough to rely on the data over time.
How is visitor privacy protected during accuracy checks?
In Indivd's QA process, reviewers only see full-body masked video or masked image pairs, never raw footage, and the masked material is deleted automatically after review. The full process is documented in the DPIA Guide: Quality assurance.