Most visitor counting systems report an accuracy figure. What they rarely tell you is how it was calculated, whether it holds consistently across hours and locations, or what happens when accuracy drifts undetected. Indivd verifies its own data because we do not trust it by default. This article explains how that process works and what the numbers in your report mean for your business.
Included in: ✓ Indivd Basic ✓ Indivd Pro ✓ Indivd Complete
Related articles
This article contains the following topics:
- Why accuracy verification matters
- Full traceability: what that means for you
- How the QA process works
- Counting QA: what the report shows
- ReID QA: what the report shows
- How to read the accuracy metrics
- Stores
- Malls
- Airports
Why accuracy verification matters
Computer vision is not inherently accurate. Without verification, counting systems can drift significantly without anyone detecting it. The error is often inconsistent: varying from hour to hour and location to location, without a pattern that makes it visible.
Many providers advertise accuracy of 99% or higher based on controlled test conditions. In real retail environments, with varying ceiling heights, lighting, and visitor densities, the same systems can deviate far more, and that deviation can change from one hour to the next. A system that is 95% accurate at 10:00 may be 70% accurate at 14:00 on the same day. If nobody is checking, nobody knows.
The business consequences are real. Research shows digitalization can reduce costs for information-intensive retail processes by up to 90% (Parviainen et al., 2017). That potential depends entirely on data quality.
- Staffing: Inflated or deflated visitor counts lead directly to overstaffing or understaffing.
- Performance assessment: Conversion rates calculated on inaccurate footfall produce false conclusions about store performance.
- Campaign measurement: Effectiveness measured against a drifting baseline is effectively unmeasurable.
The problem is widespread. McKinsey (2021) found that only 15% of retail leaders are satisfied with how they measure customer experience. A survey by Redant (2022) found that 39% of retailers lack confidence in their data quality. With physical retail margins generally below 5% (Indivd benchmark study, 2023), there is no room for decisions built on unverified data.
Indivd's QA process has involved over 500,000 manual assessments to date. The methodology has been independently assessed by KPMG in audits that placed Indivd among global leaders in people counting accuracy. The full process is documented and available to anyone who wants to assess it.
Full traceability: what that means for you
Every Indivd QA assessment can be audited in full: the sample used, the methodology applied, the manual validation steps, and the final calculations. Nothing is a black box.
In practice this means:
- You can challenge any result and ask for the underlying data. We can show you exactly where a discrepancy came from.
- Your internal stakeholders, auditors, or commercial partners can review the methodology independently.
- When you use Indivd data to support a business case, you have a documented, verifiable accuracy baseline behind it, not just a number.
A stated accuracy figure means very little without knowing how it was measured, over how many samples, and under what conditions. The QA report you receive answers all of those questions, line by line and zone by zone. For the data protection perspective, see the DPIA Guide: Quality assurance.
How the QA process works
QA is performed after the initial installation at every location, repeated periodically as part of your contract, and triggered by significant changes to camera placement or configuration.
Counting QA: A masked video sample of real visitor activity is reviewed. Indivd's automated counts are compared frame by frame with manual validation. Discrepancies in entering and exiting counts are analyzed line by line.
ReID QA: Masked image pairs showing real visitor movements between two zones are reviewed. For each pair, a reviewer determines whether the two images show the same person. The automated system's decisions are then compared against the manual decisions to calculate accuracy.
Both processes use masking throughout. No personally identifiable information is accessible at any stage of the review.
You receive a QA report by email when each assessment is complete. If you have not received a report for a location you expected to be assessed, contact your Indivd account contact.
Counting QA: what the report shows
The counting QA report covers every measurement line at your location.
| Field | What it means |
|---|---|
| Absolute accuracy | The percentage of counts that were correct. A result of 98.49% means the system was correct on 98.49 out of every 100 visitors counted. |
| Data collection dates | The specific dates during which the QA sample was collected. |
| Participants | The total number of visitors included in the assessment. Larger samples produce more reliable results. |
| Confidence interval | The statistical reliability of the result. A 99.99% confidence interval means the result represents normal operating conditions, not a lucky sample. |
| Deviations | Lines where accuracy fell below 97%, listed with a reason. Common causes: low count volume, camera placement distance, or insufficient lighting. |
Note: Low-traffic lines such as emergency exits or secondary stairwells naturally show lower accuracy figures. A single missed count represents a larger percentage of a small total. These lines are always flagged with a reason and do not indicate a problem with the overall system.
ReID QA: what the report shows
The ReID QA report covers anonymous visitor tracking across zones at your location.
| Field | What it means |
|---|---|
| ReID accuracy | The percentage of match decisions where the automated system agreed with manual review. A result of 95.52% means the system was correct in 95.52 out of 100 cases. |
| Capture rate | The proportion of visitors the system successfully re-identifies across zone boundaries. A 60% capture rate means the accuracy figure applies to 60% of total visitors moving between zones. |
| Data collection date | The date the ReID sample was collected. |
| Participants | The number of visitor movements included in the assessment. |
| Deviations | Zones where accuracy fell below threshold, listed with a reason. Most commonly: low resolution due to camera placement or insufficient lighting. These are flagged as improvement opportunities. |
How to read the accuracy metrics
Both reports use absolute accuracy as the primary metric. One missed entry and one missed exit count as two separate errors. There is no smoothing that obscures where the system fell short.
The overall location accuracy, shown at the top of each report, averages across all lines weighted by traffic volume. Indivd's minimum threshold is 95% for individual lines with sufficient traffic.
The confidence interval tells you how statistically reliable the overall figure is. A 99.99% confidence interval means the result reflects consistent real-world performance, not a favourable sample.
Note: If a line in the deviations section shows a low accuracy figure, check the participant count first. Lines with fewer than 50 to 100 counts are inherently more volatile. The deviations section always includes a reason for context.
Stores
QA gives you a verified accuracy baseline for every entrance and zone boundary you measure. When you use Indivd data to evaluate a window display, test a layout change, or measure campaign ROI, a 5% change in traffic could be real or measurement drift. The QA report tells you which. It is also the document you can share with brand partners or internal stakeholders who question the data before acting on it.
Malls
Credible tenant performance reporting depends on verified counting accuracy across every entrance and circulation line. When you present cross-visitation data or footfall figures in a lease discussion, the QA report is what stands behind the numbers. Full traceability means any tenant who questions a figure can be shown exactly how it was produced.
Airports
Qualification rate figures reported to concession partners or used in commercial benchmarking need to be defensible. A documented methodology, a 99.99% confidence interval, and complete traceability provide the audit trail that commercial and operations teams require when those figures are challenged or used in contractual reporting.
Comments
0 comments
Please sign in to leave a comment.