It goes in where something is already there. Up to a certain point.

The setting: a hospital, a municipal complex, a network of schools or a factory with fifteen years of layers on top.

A classic platform, on your servers, with two pieces that get left out of the quotation.

• The server runs on Windows and also on Linux, and that second option is not a catalogue detail: for an IT department that already maintains Linux it means one licence less, less surface to patch and a machine that does not reboot on a Tuesday. For one that only knows Windows, it means the opposite, and then it goes on Windows.
• It is sold as software on your own hardware or as the house's closed appliances, with tiers according to the number of cameras and the storage. The appliance route is shorter to commission and stiffer to extend; the software route is the opposite. You choose, and you choose at the start.
• Each server records its own cameras, and above them sits a management layer for seeing many systems from one place: status, users, updates. That is the piece that gets left out of the quotation and the one that makes the difference when there are fourteen buildings — without it, fourteen systems are fourteen visits.
• Recording to local drives, NAS or a storage array. Failover —another server picking up the cameras of the one that went down— is a function of the tier and not a tick box: if the project needs it, it may not be a matter of adding a licence but of changing tier, which is one of the nastiest price jumps and the one that gets discovered late most often.
• Desktop, web and mobile clients. With the usual warning, which weighs here because many of these installations run across several sites on tight lines: the load is not set by the recording, it is set by how many people watch from outside at the same time, and that gets counted for two years from now and not for delivery week.
• A programming interface and an SDK, plus the integrations with the house's own access control and intruder detection, which are the ones that arrive already done. With third parties you check the system, the version and who maintains the connector before quoting: «it is compatible» answers none of the three.

First, what is actually true. And then the four places where the sentence breaks, which are the four that turn up in a real migration.

It has the clearest customer profile of these six, and that makes the conversation easy in both directions.

On this platform half the value comes before anything is installed: in the inventory, and in telling the truth about what gets kept.

• The real inventory, unit by unit and camera by camera, with model, firmware and condition. Without that you cannot say what gets integrated and what gets replaced, and in an installation with years on it that inventory is half the project and half the budget.
• The integration check before quoting: what can be seen from the new client, with what limitations and what has to be replaced, with its price, even when it spoils the quotation. And the phased plan building by building, with what survives each phase written down.
• The sizing: cameras per server, drive space for the agreed retention at the real resolution and the real frame rate, and —the part almost nobody counts— how many people are going to watch from outside two years from now. With the calculation handed over, not the total.
• The choice of tier and of route —software on your own hardware or a closed appliance, Windows or Linux— justified in writing against a three-year horizon, because it is the decision that sets the ceiling.
• The migration, which is where everything goes wrong: living alongside the old system in read-only while its retention runs out, exporting whatever has to be kept in an open format, and the cost of those weeks in the quotation and not in an awkward conversation in May.
• The integrations with access control and intruder detection, tested by triggering the event on the real system and not in the manufacturer's demonstration. And the acceptance tests written before anything is built: switch a server off, fill a drive on purpose, open the real number of remote connections planned and measure.
• And the maintenance with a calendar, with the date of the manufacturer's support contract inside it: that is the one that gets left to lapse without anyone noticing, and here letting it lapse is expensive. Plus documentation: a live inventory, which server records what, a map of licences and tiers, the retention calculation and the administrator credentials in your name.

Here the split depends on a decision taken at the start, and it is worth knowing that before you take it: on Linux or on Windows, and with a closed appliance or without one.

• IT keeps servers, drives, backups, the video network, accounts and updates. If it was built on Linux, that demands an IT team that really maintains it —and if you have one, it is on home ground and it wins—. If it was built with the house's closed appliances, it is a good deal less work and a good deal less room to touch anything. Both are defensible decisions; the one that fails is the one taken without asking IT.
• And IT keeps the task that, in an installation with many buildings, decides whether this ages well: using the central management layer to know what version each system is on. Without that, in three years there are fourteen different versions and nobody able to say which.
• Security keeps users and permissions, which camera is seen by whom, recording rules and alarms, and above all the export. In a hospital or a school, who takes images out, with what authorisation and what gets logged is a data protection decision, and the person answerable for it has to be able to take it without asking IT for permission.
• What is not left to day-to-day use: the retention, the recording quality and how the storage is shared out. Two tick boxes, consequences in terabytes.
• The integrator is left with the camera drivers when a new model comes in, the version jumps, the growth of the storage and —the one most particular to this platform— the inherited equipment that has stayed inside. That ages, and when each piece gets retired is reviewed every year with a date. Plus the coordination with the access control side: same house, different products and different release calendars, and that is real work.
• Training: half a day for the operator on their views and their exports; one day for whoever administers permissions and retention; and a technical session with IT on the server, central management and backups. All three with a written script for whoever joins next year, because someone will.

This platform is not bought for its analytics, and in an inherited fleet that has a consequence worth seeing before the demonstration.

If your problem is a fleet of many brands and many ages, there is more than one answer in the index and it is worth looking at them together: the difference between them is not in the list of functions, it is in which part of what you have installed survives.