The cloud does not remove the box at the premises. It makes it dumber.

The problem it solves is almost never about video: it is about maintenance.

Everything turns around the bridge and around one decision: what goes up, when, and at what quality.

• The bridge is the one doing the work: it collects the streams, keeps them on its drive and sends them as configured. The sizing consists of choosing how many days it holds locally, and that is not a whim — it is how long the line at that premises can be down without losing anything.
• What goes up is not everything at full blast: a light version of the video goes up continuously, and the full quality according to the contracted retention or when it is asked for. That setting decides the line bill and what you will actually have three months from now, so it has to be written into the handover, camera by camera.
• Scaling: you add premises, not servers. Each premises is one more bridge in the same console and there is no central machine on the client's side, so a chain goes from twenty to sixty branches without redesigning anything.
• Redundancy: the recording lives in two places —bridge and cloud—, which solves the classic problem of somebody walking off with the recorder. And the other way round: a camera hanging off a single bridge depends on that bridge. If the premises is critical, either there are two, or the risk is accepted knowingly.
• Access and intruder detection are integrated, not unified, and video alerts can travel to an alarm receiving centre. The API is one of the most open in this group: it lets you cross a sale at the till with the image from that second, or build your own panel with the status of all sixty shops.

The same five, and here with different answers from a platform with its own camera. It is best not to copy conclusions from one to the other.

The answer depends less on the security than on the line at each site.

On a cloud platform all the work goes into the preparation.

• The survey of each premises before promising anything: what cameras there are, which of them are any use, what protocols they speak, what line it has and how much real upload it gives at eleven on a Tuesday. Measured, not the one in the operator's brochure.
• The sizing: which bridge goes at each site, how many days it holds locally, what quality each camera uploads and how many cloud days each one deserves. With the sustained upload figure alongside, which is the one that decides whether this works.
• The rollout: bridges, a video network separated from the company one, cameras added one by one and checked by day and by night, and user roles by position.
• The commissioning of what raises alerts: which events generate which alert, to whom, at what hours and what the person receiving it does. With the integration to the alarm receiving centre tested by triggering the real event.
• The migration and the maintenance: coexistence with the old recorders, exporting whatever has to be kept and then their removal. Then fields of view, checking that each camera uploads what it should, alerting on silent premises and an annual review of the contract.

The split is easier than on a classic system, and there is one place where a technician is still needed: the network.

• IT: the network at each premises, the line, keeping the video traffic separate, the accounts and two-factor. There are no video servers, but there is a new and permanent dependency on the connection. When shop seven changes operator and nobody says anything, the one that finds out is the video system.
• Security or whoever is in charge of each premises: watching, searching, marking, exporting for the police or the insurer and dealing with an alert. It is a web page and it is learnt in a morning, which makes it possible to give the manager of each branch access to their own cameras and to nothing else.
• The detail that matters in a chain: with twenty premises there are twenty managers, and they change. If nobody reviews who still has access, the system ends up with accounts belonging to people who no longer work there. That review belongs to somebody with a name, and it is decided at handover.
• It demands of the integrator more knowledge of networks and bandwidth than of video software. The problems here are almost never the program's: they are the line, the switch, the configured quality or an old camera that closes the connection every two hours.
• Training: a morning for daily use, with one page per premises. And a session with IT on what happens when the line goes down and how a bridge's status is checked. Somebody on the client's side being able to tell «it is not recording» from «it is not uploading» saves half the calls.
• Documentation: a list of premises with their bridge, their line and their calculated upload; a sheet per camera with what it sees, what it uploads and the days it keeps locally and in the cloud; a map of roles; the storage region and the processing contract; and the administrator accounts in the client's name.

Almost nobody chooses video software by looking at a single platform. These are neighbours in the same conversation, and the index leads to all twenty.