A camera system is graded on one question: can you identify the person

That is the wrong first question, and answering it first is how systems end up with twelve cameras and no usable footage. The right sequence is to list the specific things you need to see (the face of a stranger at the front door, a license plate at the garage ramp, whether the fire exit was propped, what happened at the register) and then work out what each of those needs in pixels per meter at its actual distance.

That process routinely reduces camera count in some areas and increases it at the two or three points that matter. A single camera positioned to give 500 px/m across a doorway is worth more than four cameras giving you a wide, dark, unusable view of a lobby.

Thirty days is the common commercial default and it is a reasonable starting point, because most incidents surface within a week or two and thirty days covers the ones that surface later. Insurers, franchise agreements, licensing conditions and specific industry rules sometimes require more, so check what you are already contractually committed to before choosing a number.

The design point is that retention does not have to be uniform. Ninety days on entrances, cash handling and loading areas with thirty days everywhere else usually costs 30 to 40 percent less than ninety days across the board, and it protects exactly the footage anyone ever actually asks for.

Often, and in a pre-war building it can be the difference between a viable project and an unaffordable one. HD-over-coax formats put genuine high definition (up to 4K on current equipment) down the same RG-59 that used to carry an old analog camera, and the distance capability is far beyond Ethernet’s 100 m limit. There are also converters that run Ethernet over existing coax where you want IP cameras on legacy cable.

The trade-offs are real and we will state them: limited on-camera analytics, no PoE switch infrastructure you can reuse later, and a ceiling on future expansion. In a building where fishing new Cat6 through plaster-on-lath and terra-cotta would cost more than the cameras, that trade is usually worth making. In a building with accessible pathways, it usually is not.

Check your upload bandwidth first, because that is what decides it. Continuous cloud recording needs roughly 1 to 2 Mbps of sustained upload per 1080p camera, and some platforms specify 4 Mbps guaranteed per camera. Twenty cameras therefore wants 20 to 40 Mbps of dedicated upload, around the clock, and a 40-camera site wants around 80 Mbps. Most commercial circuits in this region cannot deliver that without a fiber upgrade, and if it is not shaped it will flatten the tenant’s phones and internet.

Cost is the other half. Cloud video commonly runs $10 to $30 per camera per month for 30-day retention, with 4K and long retention plus analytics reaching considerably higher. Forty cameras at $20 a month is around $48,000 over five years before any circuit upgrade. There is a genuinely good middle path: record locally at full rate, and use cloud for remote access, event clips and off-site redundancy. That gets you the resilience without the bandwidth bill.

Yes, and they should be. Port forwarding puts a recorder’s web interface directly on the public internet, where automated scanners index it within hours. Combined with default credentials, that is the mechanism behind the large surveillance botnets, and it is not a defensible architecture for a business.

The right arrangement is no inbound ports at all: remote access over a VPN, or a vendor-managed outbound tunnel from a hardened bridge. Cameras themselves should be on their own network segment with no route to the internet and no path to the office network, management on HTTPS only, UPnP disabled on both the router and the recorder, and every default account changed. That is a one-time configuration, not an ongoing cost.

Common areas (lobbies, hallways, stairwells, laundry rooms, garages, exterior) are ordinarily fair territory for video, and in a multi-tenant building they are usually where the value is. What changes the analysis is audio, which we recommend leaving off by default, and any placement that lets a camera see into a private space.

The specific things to avoid are cameras in or covering restrooms, changing areas and similar spaces, which are prohibited outright, and camera positions where the field of view includes the inside of an apartment when a door opens. That last one is a mounting angle decision made on site, and it is the sort of thing we would rather adjust with a ladder on installation day than argue about later.