Every camera manufacturer now sells analytics, and the feature lists are long enough that comparing two of them is close to meaningless. There is a simpler test.
The test
Does this feature reduce the number of recordings a person has to look at, or does it add another thing to configure? Analytics that pass that test are worth money. The rest are a line item.
The ones that usually pass
Object classification, meaning the system can tell a person from a vehicle from an animal. This is the single most valuable one, because it removes the majority of nuisance alerts at source rather than filtering them afterwards.
Line crossing and intrusion zones, where the alert is tied to a specific boundary rather than to motion anywhere in frame. A fence line, a dock door after hours, a gate.
Search by attribute after the fact. Being able to ask for people in a time window rather than scrubbing through six hours is where most of the day-to-day value actually shows up, and it is the feature people notice last.
The ones that usually do not
Face recognition, for most commercial properties. It brings real privacy and policy obligations, it needs a curated database to compare against, and very few of the buildings asking for it have a use that justifies either.
Behavior detection billed as spotting suspicious activity. Loitering detection on a fixed zone is straightforward and can be useful. Anything claiming to infer intent is a claim worth being skeptical about.
Counting, unless somebody is actually going to act on the number. Plenty of sites buy it and nobody ever opens the report.
The condition everyone skips
Analytics are only as good as the frame they are given. A camera set too wide, aimed at the wrong thing, or blinded at night will produce bad analytics no matter what is running on it. The software cannot recover information the lens never captured.
Which means the honest order is: get the camera positions right, then decide which analytics are worth enabling. Doing it the other way around is how sites end up paying a license fee to be notified about headlights.
Where the processing happens
Analytics run either on the camera, on the recorder, or in the cloud, and the difference affects what you can change later.
On the camera is usually the most reliable, because the analysis happens on the full quality image before anything is compressed for transmission. The limitation is that the capability is fixed at purchase. Improving it means new cameras.
On the recorder is more flexible, since one upgrade covers every camera, but the recorder is working from a compressed stream and has to do it many times over, which limits how many channels it can handle well.
In the cloud is the most flexible of all and depends entirely on upload bandwidth, which brings you back to the same constraint that governs cloud recording.
None of these is the right answer universally. The useful question to ask a supplier is where the analysis runs and what it would take to improve it in two years, because that is the answer that tells you whether you are buying a capability or renting one.
Elite Security will say plainly where analytics will earn their cost on your site and where they will not, which is usually a shorter list than the brochure. More in the blog, or start with AI analytics and LPR.