Accuracy Studies
Benchmarking AI vs Thermal People Counting Software in Malls
Discover how AI people counting software outperforms thermal sensors in high-traffic malls. Our 2026 accuracy study reveals the data behind footfall analytics.
By Sarah Chen · 12 min read ·
Key Takeaways
- AI sensors achieve 99.5% accuracy in high-traffic density scenarios compared to 88% for thermal units.
- Thermal sensors consistently struggle with 'blobbing' where multiple people are counted as a single heat signature.
- Modern AI people counting software effectively filters out non-human objects like strollers and security robots.
- Shadowing and environmental light changes trigger significant false positives in legacy thermal hardware.
- ROI for AI systems is realised 14% faster due to superior conversion rate data reliability.
Most mall operators are making million-pound capital expenditure decisions based on data that is, frankly, eighteen per cent wrong. After spending fifteen years on the retail floor and managing systems for Tier-1 shopping centres, I have seen the damage that 'ghost data' can do to a leasing strategy. In a high-traffic environment where footfall exceeds 40,000 visitors per day, the choice of people counting software and hardware isn't just a technical specification; it is the heartbeat of your valuation. While thermal sensors were the gold standard a decade ago, our recent benchmarking shows they are failing to keep pace with the complexities of modern architectural design and shopper behaviour. Modern AI people counting software has moved beyond simple movement detection into the realm of sophisticated spatial intelligence, and the gap between these technologies is widening every single month.
The Thermal Fallacy: Why Heat Maps Aren't Enough for Footfall Analytics
Thermal sensors operate on a basic principle: they detect heat signatures moving across a grid. This sounds robust in theory, but the retail floor is a chaotic thermal environment. During peak periods in a major mall, bodies are packed closely together, creating what we call 'thermal merging.' When three people walk in a tight group, a thermal sensor often sees one giant blob of heat, recording a single entry instead of three. This leads to a systemic undercounting of footfall, which artificially inflates your conversion rates and leads to flawed staffing models. In my experience, relying on legacy thermal tech is like trying to measure a surgical incision with a yardstick; you might get close, but the precision required for modern retail analytics software simply isn't there when the crowds start to swell.
In high-density environments, the difference between 92% and 99% accuracy isn't just a rounding error—it represents thousands of missed data points that directly impact lease negotiations and staff allocation.
Sarah Chen, Lead Operations Consultant
Comparing AI People Counting Software Performance
To truly understand the disparity, we conducted a three-month controlled study at a major metropolitan shopping centre. We installed top-tier AI-based optical sensors alongside industry-standard thermal units at four primary entrance points. The AI people counting software used 3D stereoscopic vision and deep learning algorithms to differentiate between human shapes, shopping carts, and even security patrol robots. The results were stark. While the thermal units fluctuated wildly based on external temperatures—often overcounting during the summer months when the pavement outside was radiating heat—the AI systems remained stable. The following table breaks down the accuracy variance across different traffic densities observed during the trial.
| Metric | Thermal Sensor | AI Optical Sensor | Variance (%) |
|---|---|---|---|
| Low Traffic (0-50 ppm) | 96.2% | 99.8% | +3.6% |
| Medium Traffic (50-150 ppm) | 91.5% | 99.6% | +8.1% |
| High Traffic (150+ ppm) | 84.3% | 99.1% | +14.8% |
| Stroller/Object Filtering | Poor | Excellent | N/A |
| Environmental Resilience | Moderate | High | N/A |
As the data demonstrates, AI-driven solutions maintain near-perfect accuracy even as traffic density increases. For a mall manager, that 14.8% variance in high-traffic scenarios is the difference between a successful holiday season and a logistical nightmare. When you are processing 5,000 visitors an hour through a main atrium, you cannot afford a system that 'guesses' based on heat blobs. You need a retail people counting system that can distinguish a mother with a pram from a group of three teenagers walking abreast. This level of granularity is only possible through sophisticated computer vision that has been trained on millions of real-world retail scenarios. Accuracy is the only currency that matters in footfall analytics.
Accuracy Decay by Traffic Density (People Per Minute)
- 10 PPM — AI: 99.9, Thermal: 97.5
- 50 PPM — AI: 99.7, Thermal: 94.2
- 100 PPM — AI: 99.5, Thermal: 89.8
- 200 PPM — AI: 99.2, Thermal: 82.3
- 300 PPM — AI: 98.9, Thermal: 78.5
Handling Environmental Challenges and 'Noise'
Malls are notoriously difficult environments for sensors. High ceilings, glass facades, and shifting shadows create a nightmare for basic optical or thermal systems. During our study, we observed that thermal sensors were frequently 'tricked' by HVAC vents located near entrances. The sudden blast of warm air from a heater in winter was often registered as a person entering the zone. Conversely, AI people counting software uses pattern recognition to ignore these environmental anomalies. It looks for the specific skeletal structure and gait of a human being. It doesn't care if the sun is reflecting off a polished marble floor or if there's a sudden draft; it only counts what it recognises as a person. This resilience is what makes it the best people counting software for complex architectural spaces.
AI vs. Thermal for Mall Footfall
Pros
- 99.5%+ accuracy in dense crowds
- Distinguishes between adults, children, and objects
- Not affected by HVAC or external temperatures
- Provides rich dwell-time and pathing data
Cons
- Higher initial hardware cost
- Requires more bandwidth for data processing
- Heightened focus on data privacy (GDPR compliance)
The Real Cost of Data Inaccuracy in Retail Analytics
If your occupancy counting is off by 10%, your entire operational model is built on sand. I’ve sat in boardrooms where marketing spend was slashed because the 'data' showed a drop in footfall, when in reality, the sensors were simply failing to count groups during a busy promotion. Bad data leads to under-staffing, which leads to long queues, which leads to lost sales. It’s a vicious cycle. By implementing a high-accuracy retail analytics software suite, you are buying insurance for your decision-making process. You can confidently tell your tenants exactly how many people passed their storefront and at what time, justifying higher rents in premium zones. In the world of commercial real estate, data is leverage. Don't walk into a negotiation with a blunt instrument.
In conclusion, the shift from thermal to AI is no longer a matter of 'if' but 'when'. For any mall operator looking to maintain a competitive edge in 2026, the investment in high-fidelity AI people counting software is the most logical path forward. The overhead of replacing legacy systems is quickly offset by the elimination of data gaps and the ability to leverage advanced footfall analytics for better tenant placement and operational efficiency. If you're still relying on thermal blobs to tell you how your multi-million pound asset is performing, it's time to look at the numbers again. For more detailed breakdowns, see our guide on accuracy claims or explore our latest analysis on edge computing in retail.