The short answer
Measure what changes your decisions: visits and unique visitors, capture rate, conversion, dwell time, zone heat and occupancy. Count on-device and store only aggregate numbers so you never touch biometric data. That keeps analytics defensible while giving malls, landlords and retailers the same operational levers — leasing, staffing, marketing ROI and layout — that facial-recognition systems claim.
Key takeaways
- You can get rich shopper insight without faces: visits, conversion, dwell time, heat maps, occupancy and repeat-visit patterns.
- Detection (a person is present) is not recognition (who the person is); privacy-first setups do only detection and leave no biometric templates.
- Edge processing that discards frames and stores only aggregates removes most data-protection friction.
- Metrics form a funnel: pass-by, capture rate, conversion and average transaction value turn raw traffic into usable sales insight.
- Traffic data pays off when wired into leasing, staffing, marketing ROI and portfolio benchmarking — not as numbers for their own sake.
- Jurisdiction matters: GDPR, national regulator guidance and the EU AI Act differ, so validate configuration with counsel before launch.
Why count people without recognizing them
Footfall is the number of people who enter, pass through or spend time in a retail space — but volume alone says little. The value sits in the patterns: when visitors arrive, how long they stay, whether they return, and how they move between stores and zones. For a shopping centre, mixed-use scheme or a retail network, this is an operational and commercial signal, not a surveillance goal.
Dropping facial recognition does not mean dropping analytics. The distinction is fundamental: detection establishes that a person is present (enough for a counter), while recognition attempts to determine who they are. A privacy-first pipeline performs detection and permanently forgets each individual once they leave the frame, creating no stored image and no biometric template. This removes a layer of legal friction and builds trust with tenants and visitors while preserving the management levers that matter.
- For owners: evidence for rent, tenant assessment, event planning and operational staffing.
- For tenants: conversion, floor occupancy, shift scheduling and promotion impact.
- Why skip biometrics: far simpler retention, disclosure and regulatory alignment.
The metrics ladder: what to measure
Start with simple entry counters to build a baseline you can compare across time. Then layer a funnel: how many people pass the frontage (pass-by), what share actually enter (capture rate), what share of entrants buy (conversion), and at what average transaction value. Joining visit counts to sales through the till turns traffic into true conversion — the core KPI for both leasing decisions and marketing ROI.
For larger properties add spatial metrics: occupancy by zone, dwell time per area, heat maps and movement direction. Re-identification (Re-ID) technologies that track individuals by non-facial traits such as clothing colours let operators distinguish passers-through from genuine shoppers using a dwell threshold and count unique visitors without double-counting those who exit and re-enter.
- Traffic / visits — baseline volume and hourly dynamics.
- Capture rate — the share of passers-by who enter a store or zone.
- Conversion — purchases as a share of visits.
- Dwell time — average length of visit in-store and per zone.
- Occupancy — real-time load that drives staffing and queue planning.
- Unique visitors and repeat visits — without double counting re-entrants.
Sensors and privacy by design
Measurement options differ in depth and privacy footprint. Infrared and simple door counters are cheap but give no behaviour or demographics. Wi-Fi and Bluetooth tracking detect device signals to estimate dwell, returns and zones, but only sample phones with discovery enabled and themselves draw regulatory attention. Video analytics processed on the device gives the richest picture — provided raw frames never leave the camera and are not stored.
Privacy by design means processing aggregate counts only, doing detection on-device, and never writing facial images to disk or the cloud. Some vendors earn independent validation such as the EU Privacy Seal precisely because their pipeline generates no biometric templates and cannot re-identify a returning person. Before buying, ask where each frame is processed, what is retained, and how the system separates staff from visitors.
- Infrared counters — low cost, but little behavioural insight.
- Wi-Fi/BLE tracking — device-sample based; needs a careful privacy impact review.
- Video with edge processing — full data when frames are not retained.
- Re-ID — tracking by clothing and traits without faces: unique visitors, passers-through and staff exclusion.
From counts to landlord and tenant decisions
Footfall data earns its keep when wired into decisions. In leasing, traffic is evidence: a zone with reliably high footfall justifies higher rent, while quiet areas support reconfiguration or flexible terms. A tenant's capture rate shows how well a store turns interest into entry, feeding directly into renewal and rent conversations.
For operations, traffic patterns level staffing of security, cleaning and service teams to actual peaks. For marketing, comparing before, during and after a campaign or event demonstrates objective return. Across a portfolio, identical metrics let owners benchmark stores and spot underperformers quickly, while movement between zones reveals which areas act as anchors and which need wayfinding or new tenants.
- Leasing: ground rent and renewals in objective numbers.
- Staffing: schedules matched to measured occupancy, not guesswork.
- Marketing: campaign and event ROI from footfall deltas.
- Layout: uncovering cold zones and bottlenecks.
- Portfolio: benchmarking sites on consistent metrics.
Legal guardrails and where your responsibility starts
This is general information, not legal advice. In the EU, video processing is framed, among other things, by the EDPB's guidelines on video devices: the pivotal question is whether personal data is processed at all. A system that anonymizes immediately and keeps only aggregates carries a lighter burden; if frames are stored or biometric templates are formed, stricter rules apply, including special-category data. A truly anonymous count may fall outside personal-data processing, but that depends on the concrete configuration.
Before launch, a data protection impact assessment (DPIA), a lawful basis, limited retention and visitor signage are recommended practice. Under the EU AI Act, certain real-time remote biometric identification in publicly accessible spaces is heavily restricted. Russia has its own personal-data law (152-FZ) with different tests for anonymity and biometrics. Always confirm the configuration with qualified counsel for your jurisdiction.
- Run a DPIA before deploying cameras or counters.
- Anonymize at capture so few obligations remain.
- Cap retention of any raw data and inform visitors.
- Jurisdiction: GDPR and the EU AI Act in Europe, 152-FZ in Russia — verify separately.
Honest limitations of the method
Counting without identity yields aggregates, not individual shopper profiles: you will not know who buys, and you cannot tie a repeat visit to a named person without special technology such as Re-ID. Connecting footfall to transactions takes discipline and separate tooling. Camera-based demographic estimates are approximations and can themselves be risky in some jurisdictions.
Every counter has finite accuracy, so analyse trends within each location rather than comparing absolute figures across different zones or vendors. Reassess your configuration periodically: both regulatory practice and vendor capabilities move quickly, and what looked anonymous yesterday may need re-evaluation tomorrow.
- No individual profiles without special tools such as Re-ID.
- Linking traffic to sales needs separate integration and data discipline.
- Demographics by camera are estimates, and risky in some jurisdictions.
- Compare trends inside a site, not raw numbers across different systems.
A working implementation path
To move from idea to a running scheme in a reasonable timeframe, run a short audit: decide which questions the data must answer, choose sensors scaled to the property, lock a privacy-preserving configuration and validate it with counsel, then build reporting around each metric and periodically verify that decisions really rest on the numbers.
The checklist below can be handed to a project or operations team as a ready template before launching or upgrading a footfall system.
- Write down 3–5 management questions the data must answer.
- Pick the sensor type for your scale: entry counter or zone analytics with edge processing.
- Confirm no facial images are written to disk or sent to the cloud.
- Check how the system separates staff from visitors.
- Validate lawful basis and configuration with counsel for your jurisdiction.
- Join footfall to till data to compute true conversion.
- Configure reports: traffic, capture rate, conversion, dwell time, occupancy.
- Complete a DPIA and post visitor signage.
Put it into practice
Audit checklist: footfall analytics without facial recognition
A reusable template for shopping-centre, retail-network and mixed-use operators. Work through these items before launching or upgrading a people-counting system to get legally defensible and operationally useful analytics.
- Defined 3–5 management questions (leasing, staffing, marketing, layout) the data must answer.
- Selected the sensor type for your scale: simple entry counter, infrared, Wi-Fi/BLE, or edge video analytics.
- Video is processed on-device; raw frames are not stored or sent to third parties.
- System performs presence detection only and generates no facial-recognition or biometric templates.
- Staff are excluded from counts without mandatory badges.
- A mechanism exists for unique-visitor counting and separating passers-through from shoppers.
- Footfall is joined to till data to compute capture rate and conversion.
- Lawful basis and configuration validated with counsel for your jurisdiction.
- A DPIA is complete and visitor signage is posted.
- Retention periods for any interim data are fixed with a deletion procedure.
- Regular reporting is configured per metric with an accountable analyst.
- Periodic reassessment is scheduled (annually or when regulation changes).
Questions people ask
Is anonymous people counting by camera processing of personal data?
If the system anonymizes immediately and stores only aggregate counters — creating no retained images or biometric templates — such processing can fall outside personal-data rules altogether. But the answer depends on the concrete configuration and jurisdiction: in the EU the EDPB's video-device guidance frames the analysis, while Russia's 152-FZ sets different tests. If frames are kept or biometric data is formed, stricter obligations apply. Confirm with qualified counsel.
What is the difference between face detection and face recognition?
Detection establishes that a person is present in a frame, which is all a people counter needs. Recognition attempts to identify who that person is, which requires biometric templates and triggers much stricter rules. Privacy-first configurations perform detection only, generate no templates and permanently forget each individual after the frame is processed — this distinction is central to whether the system is legally low-risk.
Which metrics matter most in mall lease negotiations?
The most persuasive figures for leasing are zone traffic, a tenant's capture rate (the share of passers-by who enter) and conversion to purchase. Consistently high footfall justifies rent levels, while a low capture rate shows the problem sits with the store rather than the traffic. Joining traffic to sales provides objective evidence for renewals, tenant mix and landlord-tenant alignment.
Do I always need a data protection impact assessment (DPIA) for visitor counting?
A DPIA is standard practice for processing that may create high risks to people's rights, and regulators — including the EDPB in its video-device guidance — recommend it for surveillance-style processing. Even if your system is fully anonymous, running a DPIA disciplines the process by fixing the lawful basis, retention, signage and security measures. It documents a responsible approach and eases dialogue with the supervisory authority.
Can I identify repeat visitors without facial recognition?
An ordinary anonymous counter cannot: it forgets each individual once they leave the frame. To count repeat visits and unique visitors without faces, re-identification (Re-ID) technology matches a person across multiple sensors using non-facial traits such as clothing colour, without forming biometric data. Re-ID also helps filter passers-through by setting a minimum dwell threshold before someone counts as a genuine visitor.
Sources and further reading
Sources were checked when this page was generated. Confirm changing dates, rules and prices with the original publisher.
- Guidelines 3/2019 on processing of personal data through video devicesEuropean Data Protection Board (EDPB)
- Part 1: The next evolution in people counting (Re-ID)Sensormatic Solutions
- How does in-store measurement work without identifying anyone?Pygmalios
- What is footfall? How to track and use it to drive retail successMRI Software
- Traffic Analytics | Product BriefRetailNext
- How does Quividi handle consumer privacy? (VidiReports)Quividi