A structured review of Veesion's gesture-detection layer for retail theft for loss prevention, store operations and multi-site retail teams: what it does well, where the constraints sit and who should look elsewhere. Every score traces to something Veesion publishes.
Veesion is a specialist and it documents the specialism closely. Its own FAQ names 11 gesture types in 3 categories, states that concealment, personal bag and backpack are always active and cannot be switched off, and then publishes a section on what the system does not detect, including an item carried openly out of a shop with no concealment and a theft committed entirely with the person's back to the camera. The installation side is documented to the same level: the box contents, the recorder prerequisites, the stream it reads and its settings, and the distance and height range a camera has to sit in. Retention is stated as a policy rather than as a capacity, 30 days of history after which Veesion states videos are permanently deleted. It scores lowest where a detection has to become an intervention, because the published output is a push notification with a clip, typically under 30 seconds with a median around 18, and nothing on site makes a sound. Choose Veesion to catch concealment on a shop floor. Look elsewhere when the same cameras have to answer a safety or operations question.
Five criteria, each scored from Veesion's own current documentation as of August 24, 2026. Spot AI competes with Veesion, so no score here rests on an anonymous source, an aggregated user rating or a private benchmark: each one sits next to the documented fact behind it. Anything Veesion does not document is recorded as not publicly specified rather than assumed. Veesion publishes in several languages and its installation and app FAQ carries most of the technical detail, so facts here come from veesion.io in French and English plus that FAQ. Where a fact was not on those pages, this review records the absence rather than filling it from a reseller listing. This is not a paid placement, and Veesion had no input into it.
The detection side is published down to the gesture, the stream settings and the camera height. The end of the path is published just as plainly, and it is a phone.
Veesion is a gesture-detection layer that rides on the video system a shop already has. A compact server, which Veesion describes as about as simple as an internet box, is installed on the existing recorder, reads the streams continuously and pushes a short video alert to a phone when a gesture linked to concealment is recognised. Veesion states it is compatible with any video system that supports an RTSP protocol, which it says is the case with all recent installations, and names HIK, DAHUA, TVT and Uniview as the brands its customers generally use. The design target is a store manager or a loss prevention lead who has more cameras than attention.
The detection set is closed and it is published in full. Veesion's FAQ states the AI is trained to detect 11 specific gesture types organised into 3 main categories, and that behaviors outside these categories will not be detected. Body gestures cover concealment, described as the most important gesture and always active, plus an optional gesture close to the body. Bag gestures cover personal bag and backpack, both always active, plus optional suspicious bag, baby stroller and shopping bag. Other gestures cover shelf sweeping, defined as products collected four or more times in a row and placed rapidly into a bag, along with consumption, deblistering and a slightly suspicious category for gestures too unclear to classify. Presets run Essential, Moderate and Maximum, with an expert mode for per-camera control.
The commitments page goes further than the product pages do. Veesion states it performs no biometric processing, that its system cannot identify a person or recognise that the same person came in three days ago, and that only gestures are analysed. It publishes the three components behind that claim, human detection, then pose estimation, then object recognition, and explains why pose estimation at store camera quality cannot produce a signature that would identify anybody. Recruitment at Veesion is conditioned on signing an ethics charter. The company publishes 6,000+ stores equipped, 55+ countries, about 100,000 suspicious gestures detected each month and a 38 million euro Series B raised in June 2025.
Veesion is strongest at recognising one narrow thing on a shop floor and describing exactly where that recognition stops, and thinnest anywhere the cameras have to answer a different question.
Both columns are documented. Nothing here comes from an anonymous review.
Both columns describe documented behavior. Veesion's column was checked against its own documentation on August 24, 2026.
Swipe the table sideways to see every column.
Veesion data comes from Veesion's own public documentation as checked on August 24, 2026. Gaps are marked as not publicly specified.
The honest split, stated the way a shortlist call would state it.
If the loss is people hiding product in clothing, bags, backpacks or a stroller, and there is somebody on the floor who can walk over inside a minute, Veesion is built for precisely that and documents it better than anything else here. The eleven gestures can be switched per camera so a shop whose customers carry their own bags does not drown in alerts, the alert arrives in well under a minute in Veesion's own figures, and 6,000+ stores across 55+ countries have already been through the install.
If nobody is free to intervene, the published path ends at a phone and no speaker, strobe or talk down appears anywhere on the pages this review read. The other case is scope: everything published is retail concealment, the gesture list is explicitly closed, and there is no route for a customer to define a detection, so back-of-house safety, loading bay activity or an operations question on the same cameras needs a second platform.
Spot AI ships 15+ pre-trained Video AI Agents spanning vehicle break-in, fire, intrusion, personal protective equipment, forklift near-miss, falls and hazard-zone crowding, so a mixed estate gets shipped coverage on both sides. Iris builds anything else in natural conversation in about eight minutes, full-resolution video stays on the Intelligent Video Recorder in the building and the platform is SOC 2 Type II, NDAA-compliant and HIPAA-aligned.
The two platforms answer different halves of the same shop. Veesion reads one narrow behaviour extremely well, publishes exactly where that reading stops and hands the result to a person with a phone. A platform built around a wider agent set answers security, safety and operations questions on the same feeds and can respond through speakers already mounted on site. The question is how much of your loss is concealment on a shop floor during trading hours.
The cheapest way to settle it is to run both in the same store for a fortnight and count two things separately: how many concealment events each one surfaces, and how many of those anybody actually reached in time. The second number is where the architectures diverge.
Quote the 30-day deletion policy alongside whatever your insurer or claims process needs. A shop that has to produce footage six weeks after an incident has to plan for that inside the window, not after it.
See what the AI catches on your live feeds before any platform decision.
Customer-reported outcomes from named Spot AI customers.
Blackmon Oil runs a vehicle loitering filter to keep parking lots clean and safe at its around-the-clock stores, with a smaller overnight crew.
Silver Bay Seafoods replaced fragmented legacy camera systems across 22 locations, including remote Alaska facilities, and lifted operational efficiency 15%.
Cambridge City cut footage search from two hours to 30 seconds after consolidating seven municipal locations on one platform.
"The biggest benefit of Spot AI is how easy it is to look up incidents, see the footage, and then share it and collaborate."
Yes, for a staffed shop floor. It scores 3.3 out of 5 here, carried by 11 gesture types in 3 categories with concealment, personal bag and backpack always active, by a published list of what the system does not detect, and by installation and retention terms documented down to the stream settings. It fits stores losing product to concealment. Teams with thin staffing or a scope beyond theft should read the response section first.
Yes. Veesion states it is compatible with any video system that supports an RTSP protocol, which it says covers all recent installations, and it names HIK, DAHUA, TVT and Uniview as the brands its customers generally use. A compact server is installed on the existing recorder and reads the sub stream rather than the main stream, with recommended settings published as D1 resolution or higher, a bitrate above 1000 kbps and 10 frames per second.
Eleven gesture types in three categories. Body: concealment, always active, plus an optional gesture close to the body. Bag: personal bag and backpack, both always active, plus optional suspicious bag, baby stroller and shopping bag. Other: shelf sweeping, consumption, deblistering and slightly suspicious. Veesion also publishes what it misses, stating that an item kept in plain view will not trigger an alert, that a theft committed entirely facing away from the camera cannot be analysed and that eye movements and expressions are outside what it reads.
Thirty days. Veesion states that qualified alert history is kept for 30 days and that after that period videos are permanently deleted from the system, and it advises downloading or screenshotting an important clip inside that window. Where full-resolution video lives is less specific: Veesion states data is secured on its own servers without publishing a region or a hosting arrangement.
It depends on what the cameras have to do besides catch concealment. If the sites that cost money are the ones nobody can reach in time, or the same feeds have to cover safety and operations as well, a camera-agnostic platform such as Spot AI fits, because 15+ pre-trained Video AI Agents run across the fleet already installed and an incident can be answered with talk down, strobes and horns through standard speakers. The alternatives roundup compares them side by side.