Bosch is not going to surprise you. That is what you buy it for.

That sets the order of priorities: that the equipment behaves the same in year eight as in the first one, that the image is predictable and that the data coming out of the camera can be integrated with the rest of the installation. There are no pleasant surprises in its catalogue, and none of the other kind either.

What it puts first, and this is not common, is analytics inside the unit itself as standard and not as an extra invoiced separately. The camera tells what is moving from what is not, so it does not spend recording quality on a still wall either: the drive lasts longer with the same quality on what matters.

What it sacrifices is price and apparent modernity. The interfaces are an engineer's, the documentation is dense and nothing is meant to be set up in twenty minutes. In exchange, once the system is built it stays put, and in industry that is worth more than a pretty screen.

In industry and transport: an industrial unit, a process plant, a tunnel, a station, a platform, a large car park. Places with dust, vibration, temperature and a night shift, where what is valued is that the equipment stays standing and that the recorder does not fill up because three cameras have spent a month looking at an empty corridor.

And where there is already intruder detection or fire detection from the same house, because then the video and the alarm talk without an intermediary: the camera steps up its quality when a detector goes off and whoever is on duty sees what triggered the alert. Verifying stops being a phone call, which is where the minutes that matter get lost.

In the small shop, the office or the home, where what people want is four cameras, an application on the phone and a morning's installation. It would work, and it would be paying for engineering on a scene that does not ask for it.

In projects where what rules is how many cameras fit into the money available: at that price there are much wider catalogues. And when the customer wants everything in the cloud, on subscription and with no server of their own, which is ground other houses have covered more.

Here the on-board analytics are the argument, not the decoration: they classify people and vehicles, count, alert on intrusion into a zone, on an object left behind or one that has disappeared, and they do it inside the unit. Against the motion detection of old —which goes off with a shadow and with a bag— the difference is that the alert can be attended to.

The ceiling is worth saying beforehand and not in the demo: these are geometric rules. A line, a zone, a time, a size. What does not fit into a shape drawn over the image does not exist for the camera, and neither does what happens on the camera next door.

What comes now is a recommendation with an interest of our own, so it is declared: in the same house there is IRIS Neural, the NVMS from Infinity Neural, the other company in the group. It is our own product. We would rather you knew it by reading this line.

What it brings is what cannot be drawn: that two people are arguing, that this worker is where a crane is working today and was not yesterday, that the odd thing is the hour. And they sit together with no trouble, because what the camera already solves on its own does not need asking of anything else.

In an industrial unit the temptation is to hang the camera from the roof, which is where the structure is and where you do not have to drill a wall. At nine metres you see a lot of floor and no faces, and the sheet roof moves with the wind just enough for an analytics zone to start giving alerts on gusty days. That gets discovered in November, when everything is already closed.

The angle, the height and the light decide for good what will be possible with that image: whether one day it will be able to count, read a number plate or recognise somebody. It is settled in the project, with the plan on the table, and not on the day of fitting.