Store cameras once answered one question: who took the merchandise. That still matters. It is no longer the only reason retailers buy video.
The shift is that the same hardware now produces operational data, which means the budget conversation has moved from loss prevention to something broader.
Retail technology coverage tends to run ahead of retail technology budgets. Plenty of column inches go to the far horizon, from agentic shopping assistants to what quantum AI might eventually do for demand forecasting and supply chain optimization.
Camera analytics are the unglamorous end of that spectrum, and they are the end already running in production across thousands of stores.
What a modern system actually includes
A current setup is a set of connected components rather than a wall of monitors in a back room.
Cameras feed a video management system running on a local recorder, in the cloud, or both. Access control, door sensors and point-of-sale data connect to the same platform, so a manager can match a transaction to the moment it happened rather than scrubbing through footage by timestamp.
The analytics layer is where most of the recent change sits. People counting, heat maps showing dwell time by aisle, alerts when a queue passes a set length, and search tools that filter recorded video by attribute.
One commercial detail worth checking early: whether analytics are licensed per camera or included. Some manufacturers, including Hanwha Vision, build license-free analytics directly into the camera, which changes the total cost meaningfully across a hundred-store estate.
Security outcomes you can measure
The NRF’s 2025 Impact of Theft and Violence study reports an 18% increase in average shoplifting incidents in 2024 against 2023, alongside a 17% increase in threats or violence during those incidents.
That second figure is the one that changes procurement. Staff safety and evidence quality become priorities alongside inventory loss.
Measure things you control. Time to close an incident report, percentage of cases with usable footage, and after-hours alarms verified by video before anyone drives to the store. That last metric alone often justifies the system in a multi-site operation.
Do not promise a shrink reduction. Dollar General reported gross margin rising 78 basis points to 31% in Q1 2025, with lower shrink cited as one contributing factor after removing most self-checkout stations. Video may support results like that. It is one part of an operational plan, not the cause.
Operations, from the same cameras
The analytics that matter in practice are unglamorous.
People counting gives entries and exits against pre-defined areas, which turns into conversion rate when set against transactions. Heat mapping scores traffic by visitor density and length of stay. Queue management alerts staff when customers have been waiting longer than a threshold you set, which is the difference between opening a second register at the right moment and hearing about it in a review.
Camera placement determines what you get. Fisheye cameras cover customer flow across an open floor and expose bottlenecks. Multi-sensor units track a full journey across zones, which quick-service operators use to time ordering, payment and pickup separately.
Anyone evaluating retail store video surveillance should look at one problem most vendors ignore: cart detection. Hanwha Vision’s retail AI pack identifies shopping trolleys as well as people, which matters because a family sharing one cart registers as four shoppers in a people count and one basket at the till. Detecting the cart corrects the conversion maths. It also flags abandoned carts in the car park, which is a recovery cost most retailers absorb without measuring.
The genuinely useful applications tend to be small. Comparing dwell time between two endcap layouts. Checking whether a seasonal display blocks a walkway. Pulling a thirty-second clip to show new staff what a good handoff looks like.
Face detection is not facial recognition
This distinction decides your compliance exposure and almost every article gets it wrong.
Face detection identifies that a face is present in frame and can trigger an alert or capture a clearer image. It does not identify anyone. Facial recognition matches a face against a database to determine identity.
Most retail camera analytics, Hanwha Vision’s included, perform detection. Confirm which one a system does before you assume a local restriction applies to it, and get that confirmation in writing from the vendor.
Compliance and customer trust
Rules vary by jurisdiction and some are specific.
New York City requires a biometric information notice at every customer entrance when a business collects biometric identifier information, and the notice must be at least 8.5 by 11 inches. Portland, Oregon prohibits private entities from using facial recognition in places of public accommodation.
California finalized CCPA regulations in September 2025 covering cybersecurity audits, risk assessments and automated decision-making technology, with compliance requirements scheduled from 2027. If analytics inform decisions about people, that framework is worth reading now rather than in 2026.
Audio is separate and catches people out. Consent requirements differ by state, which is why many retailers disable camera microphones by default.
Device security deserves attention too. Ask whether cameras carry unique certificates applied during manufacturing and whether the product line meets a recognized cybersecurity standard such as UL CAP. A camera on your network is an endpoint, and it will be treated as one during any security review.
This is not legal advice. Check the rules for each store location and take qualified counsel before deploying facial recognition, biometric analysis or audio recording.
Choosing and rolling out
Define outcomes first. Faster investigations, verified alarms, shorter checkout lines. “Better coverage” cannot be evaluated.
Pilot in one store. Test the point-of-sale and access control connections specifically, since integration is where these projects usually stall. Set role-based permissions, encryption and retention before go-live rather than after.
Check what the VMS does with POS data. Some retail platforms save receipt data alongside the video, so a flagged transaction pulls up its own footage. That single capability removes most of the manual work from exception investigations.
Update signage and procedures at the same time. Staff training should cover who may view or export footage, when review is appropriate, and how incidents get documented.
Compare on the full picture. Camera coverage, storage, search function, integrations, access control, support and total cost, including whether analytics carry a recurring license. A single standout feature is a poor basis for a decision you will live with for seven years.
FAQ
Do security cameras reduce shrink?
They can contribute, but the result depends on how footage, alerts and incident procedures are used. Shrink improvements usually come from several operational changes at once, and attributing them to video alone is not supportable.
Can I use facial recognition in my stores?
It depends where you operate. Portland, Oregon prohibits private use in places of public accommodation and New York City requires notices when covered biometric data is collected. Confirm applicable rules with counsel, and check whether your system actually performs recognition or only detection.
What is the difference between face detection and facial recognition?
Detection identifies that a face is present. Recognition matches that face against a database to determine identity. Most retail analytics perform detection, and the two are treated very differently under local law.
Which analytics deliver value fastest?
Queue management and people counting, because both produce an action the same day. Heat mapping is useful for layout decisions but works on a longer cycle.
How much footage do we need to keep?
Long enough to cover your claim and investigation windows, and no longer. Retention drives storage cost and increases exposure, so set the period deliberately rather than defaulting to the maximum the system allows.
Do analytics require a separate license?
It varies by manufacturer. Some charge per camera per year, others build analytics into the device at no extra cost. Ask directly, because across a large estate the difference is substantial.
The takeaway
Retail video earns its place in two ways: supporting safety and investigations, and showing managers how queues, layouts and staffing actually behave.
Define outcomes you can measure, write the privacy and access rules before installation rather than after, and compare vendors on total cost across the whole estate. That is a duller process than a demo, and it is the one that produces a system people still use in year three.
