We get shown a lot of competing quotes. The spread is remarkable — for what is broadly the same capability, we've seen ₹500 per camera per month and we've seen ₹8,000.

The difference is rarely the intelligence. It's what got bundled around it, and how much of it you actually needed.

The four things you're paying for

Almost every AI CCTV quote decomposes into the same four items. Getting them itemised is the fastest way to compare offers honestly.

Software licence. The detection itself, charged per camera per month. This is the part you're genuinely buying.

Edge hardware. A device at your site that runs the models. Charged either as a one-off or amortised into the monthly fee. Capacity scales with camera count, so a 120-camera warehouse needs more than a 4-camera shop.

New cameras. Sometimes genuinely needed — a gate camera at face height for recognition, or an ANPR camera at plate height. Often not needed at all.

Installation and tuning. Usually underquoted. Tuning is what determines whether the system survives its first month, and a quote that omits it is quietly transferring that work to you.

Where quotes get inflated

Rip-and-replace. The most expensive line on any quote is a new camera estate. It's also the least often necessary. If your DVR exposes an RTSP stream, software can run on what you have. Ask directly: which of my existing cameras will you reuse, and why not the rest? A vendor who wants to replace all of them should be able to justify each one.

Per-feature pricing. Face recognition as an add-on, ANPR as an add-on, loitering as an add-on. This turns a predictable line item into an unpredictable one, and it means the vendor's incentive is to keep capabilities behind a paywall rather than tune what you have.

Cloud processing. If video is analysed off-site, you're paying twice — once for the processing and again for the bandwidth to ship every frame out of your building. It also means your detection stops when your internet does.

Where quotes get suspiciously cheap

Below about ₹500 per camera per month, something is usually missing.

Sometimes it's tuning — you get a generic model with default thresholds, which produces exactly the false-alarm flood you were trying to escape. Sometimes it's support, so the first time detection drifts you're on your own. Occasionally it's honesty about what the model does: "AI-enabled" on a spec sheet sometimes just means motion detection with a marketing budget.

The test is simple. Ask what happens in the second week — who reviews the events with you, and who adjusts the thresholds. A cheap quote that has no answer isn't cheap; it's incomplete.

What we charge, and why it's one number

PGAK is ₹1,000 per camera per month. That covers the software, the detection capabilities, the updates that arrive along the way, and the tuning. It does not change by city, by industry, or by which features you switch on.

We price it that way for a straightforwardly selfish reason: per-feature pricing makes us argue with customers about whether they can have loitering detection, and that's a stupid argument to have when the model is already running.

Doing the arithmetic

A 60-camera warehouse is ₹60,000 a month. That is a real number and worth being sceptical about.

The comparison that matters is not against your current CCTV bill — it's against what you're currently absorbing. Most warehouses we work with were writing off more than that in monthly shrinkage before anyone was watching. For a shop with twelve cameras, ₹12,000 a month sits against a single significant theft.

If that arithmetic doesn't work at your site, it doesn't work, and we'd rather say so during the audit than after the invoice.

The question to ask every vendor

Which of my existing cameras will you use, what will you charge me monthly, and what happens in week two?

Three answers. If any of them are vague, keep looking.

Related reading: PGAK pricing · AI CCTV for warehouses · Case studies