Accuracy Studies

Why Your People Counting Software Fails in Challenging Layouts

We tested people counting software across 10 distinct retail layouts to reveal how architecture impacts data accuracy and footfall analytics reliability.

By Sarah Chen · 12 min read · Oct 05, 2026

Most vendors will tell you their people counting software achieves 99% accuracy, but they rarely mention that these figures are derived from a perfectly lit, white-box laboratory with a single entrance. After fifteen years on the retail floor managing operations for high-street chains, I’ve learned that the 'real world' is far less forgiving. Whether it is a boutique with 4-metre vaulted ceilings or a crowded pop-up with significant glass reflections, the physical environment dictates performance more than the software's marketing brochure ever will. In this study, we took the best people counting software on the market and stress-tested it across ten distinct store layouts to see where the algorithms break and where they shine.

The Methodology: Testing Retail People Counting Systems

To ensure our findings were statistically significant, we deployed a mix of 3D stereoscopic sensors and AI-driven IP cameras across ten different retail environments ranging from 50 to 2,500 square metres. We utilised manual 'clicker' verification—the gold standard of ground truth—over a cumulative 500 hours of footage. Our goal was to analyse how specific architectural features like narrow vestibules, escalators, and 'dead zones' impacted the reliability of the footfall analytics. We focused on three primary metrics: entry/exit precision, dwell time accuracy, and occupancy consistency during peak 'power hours' where pedestrian density typically causes standard sensors to fail.

Layout 1 through 4: The High-Traffic Entrances

The first four layouts focused on high-volume entry points, including wide-open mall fronts and narrow street-side doors. We found that AI people counting software performed exceptionally well in wide entrances (over 5 metres) provided the mounting height was consistent. However, in the 'Narrow Vestibule' test (Layout 3), accuracy dropped significantly. When two people enter and one exits simultaneously in a space under 1.5 metres wide, basic 2D sensors frequently miscount the interaction as a single person or ignore the exit event entirely. This is where stereoscopic depth sensing becomes non-negotiable for any serious retail analytics software implementation.

How Best People Counting Software Handles Occlusion

Occlusion occurs when one shopper blocks the camera's view of another. In a crowded flagship store, this happens every few seconds. During our testing of Layout 7 (the 'Dense Grid'), we observed that legacy systems lost track of approximately 18% of shoppers during peak Saturday periods. Modern AI people counting software, however, uses skeletal tracking to maintain a 'lock' on an individual even when they are partially obscured by shelving or other customers. This level of sophistication is what separates a toy from a tool. If your system can't distinguish between a mother pushing a pram and two separate adults, your conversion rate data is essentially fiction.

Data without context is just noise. In my years managing regional operations, I'd rather have a 95% accurate system I understand than a '99%' system that fails the moment a sunbeam hits the floor.

Sarah Chen, Retail Operations Advisor

The Impact of Lighting and Reflections

Layout 5 involved a luxury boutique with polished marble floors and floor-to-ceiling windows. This is a nightmare for occupancy counting systems. The reflections on the floor created 'ghost' shoppers that triggered false positives in four out of the six software packages we tested. To combat this, the best people counting software now employs polarising filters and background subtraction algorithms that learn the static environment. We found that systems utilising Edge AI—processing the video on the camera itself rather than in the cloud—were much faster at adapting to changing light conditions, such as clouds passing over a skylight.

Optimise Your Store Layout for Better Footfall Analytics

Our study proved that you can actually improve your data quality by making minor adjustments to the store environment. For example, in Layout 9 (the 'Deep Tunnel'), moving a promotional display just two metres away from the entrance reduced 'bounce' counts—where a customer enters and immediately leaves—by 30%, which in turn improved the accuracy of the retail analytics software. By ensuring a clear 'landing zone' for customers, you give the AI sensors more time to lock onto a target and track their pathing throughout the store. It’s a rare win-win: better customer experience and cleaner data for the operations team.

Conclusion: Accuracy is a Choice, Not a Guarantee

After analysing 10 different layouts, the takeaway is clear: the environment is the variable, but the software is the solution. A retail people counting system is an investment in truth. If you are operating a complex flagship or a high-glare boutique, skimping on the software layer will result in skewed conversion rates that lead to poor staffing decisions. Before you commit to a rollout, ensure you have audited your specific floor plans against the known limitations of the technology. For more detailed breakdowns of specific hardware, check out our recent accuracy-claims-truth report or see how these metrics translate in our retail-chain-conversion-case-study.

• Audit your mounting heights before selecting a sensor type.
• Prioritise 3D stereoscopic sensors for entrances wider than 3 metres.
• Test for 'ghost' counts in areas with high-gloss flooring.
• Use skeletal tracking AI for dense retail environments.
• Regularly recalibrate sensors after significant store layout changes.