Turn the cameras you already have into an AI counting system — for drinks, dishes, products, packages, ice and more. Send us a short clip and we'll show you what's possible.
Compatible CCTV required — the items must be clearly visible in your footage.
Most owners have no reliable way to know. Dishes, drinks and stock go out on comps, mistakes and theft — and never reach the POS or the paperwork.
of revenue is commonly cited as lost to theft & shrinkage*
counting systems already built on real client footage — dishes, drinks, ice blocks
spent on cameras before we've looked at your footage
*Industry estimate; ranges vary by source and region. TrueCount AI measures your actual gap during the pilot.
Six steps, and the first three cost you nothing.
One line is enough: “I want to count drinks leaving my bar.”
A short recording off your existing CCTV that shows the workflow — a file or a Google Drive link. Nothing to install.
Camera angle, item visibility, lighting, overlap, movement, the counting area and the hardware you already have. You get a straight answer on what's possible.
Processing on a computer at your location, or in the cloud — whichever suits your footage and your equipment.
The counter is trained and tuned for your items, your camera view and your counting point — that's the paid build.
You get the counts, timestamped, plus the agreed reporting — and the comparison against your POS or paperwork if you want it.
That's the whole journey. You're not buying an AI project — you're buying a managed counting service built around the cameras you already have.
A tray of four counts all four. Multiple waiters, multiple dishes — each one tracked.
Bamboo steamers, sauce & yogurt cups, empty bowls — even items on shoes or aprons — are skipped.
We train it on clips from your own cameras, so it learns your dishes, your angles and your lighting — accuracy where it counts.
Moving a dish to a side table, returning empties or staff walking around does not count.
You don't necessarily need new hardware — we build around compatible cameras you already own. We check yours before anything is bought.
It counts dishes and movement, not identities — built to respect staff privacy.
If your camera clearly shows the item, and there's a consistent point where that item moves through the view, we can evaluate building a counting system for it.
Every plate that crosses the pass is counted and timestamped, then lined up against the POS. A tray of four counts four; steamers, sauce cups and empties are ignored.
Each drink picked up off the bar is counted as it goes out — so drinks that were poured but never rung up show up as a gap at the end of the night.
Blocks carried out over the loading platform are counted per platform and per shift, so what leaves for dispatch can be checked against what was produced.
The three above were built on real client footage. The same approach applies wherever a clearly visible item passes a fixed point — these are candidates we can evaluate for you:
Every use case is different. Camera angle, lighting, item size, movement and overlap all affect what's possible — so we evaluate your actual footage before confirming anything.
Three different businesses, three different cameras. Items outlined on the real footage, the counting area marked, and the counter ticking up only when something genuinely goes out.
On real CCTV footage — kitchen, bar and ice plant — the system's counts matched what was hand-counted on review, including a 4-dish tray and drinks picked up off the bar, while correctly ignoring bamboo steamers, sauces and decor bottles. Notice the marked counting area in each clip: that's what the counter reacts to.
TrueCount can use compatible existing cameras to build a counting system around your workflow. Not every system is compatible — here's what we look for.
You don't necessarily need to buy anything. If your CCTV covers the right spot, we build around it. We tell you honestly if it doesn't.
This is the big one. If a person watching the footage can clearly see that a dish, a drink or a bag is going out, the use case is suitable for evaluation.
The item needs to stand out from its surroundings. Very dark footage, heavy glare, strong backlight or deep shadow across the counting spot make it harder.
The camera needs a clear view of the area where counting happens — not too far back, not blocked, not aimed straight into a window, not constantly walked in front of.
A drink on the bar in front of the camera is easy. The same drink as a few pixels across the room is not. There's no magic resolution — what matters is how big the item looks in the frame.
Items should be distinguishable one from another. Ten bags stacked so the camera only sees the top one, or items constantly blocked by people, will reduce accuracy.
TrueCount works best where items pass through one consistent place — the pass, the end of the bar, the loading platform. We mark that area in your camera view.
Your camera system — cameras direct, or through an NVR — needs to provide a video stream TrueCount can access. Recorded files work too; that's how the free evaluation starts.
For counting at your own location, a suitable computer connected to the camera network. We look at what you already have first — and if it isn't enough, there's the cloud option.
TrueCount counts an item when it crosses the area we define — not every time it appears on screen. That's what stops staff walking around, items moved aside and returns from inflating the count.
The chain is short — and we'll check your existing computer before you spend anything on hardware.
Don't have a machine that can handle it? That's fine — we can run the counting in the cloud instead. Nothing needs buying before we've confirmed what's feasible.
You don't need every box ticked — some of these depend on how your system ends up being deployed. Send us a sample and we'll tell you where you stand.
Send us a short sample video. We'll evaluate your setup and tell you what's possible — before you spend anything.
If your camera can clearly see it, we can evaluate it. Clear visibility makes a use case suitable for evaluation — we test your actual footage before confirming the final setup.
Building the AI is a one-time fee per product type — the more types, the less each one costs. Then it has to run somewhere — that part isn’t optional, but the cost can be: nothing per month on your own hardware, or a monthly fee if we run it for you. Billed in your local currency.
Start with the free demo — pricing only matters once you've seen your own numbers.
1 camera · a short clip
One-time · the more types, the cheaper each
It has to run somewhere — you pick where
Because the work is teaching the AI to recognise a new thing. Ten types is ten things to learn — so each one gets cheaper.
For example: 5 types = $750 · 10 types = $1,000 · 20 types = $1,400
You never pay more for wanting more. If a bigger bundle works out cheaper, we charge you the lower price — 10 types costs less than 9, and we won't let you overpay for it.
Because what costs us to run is video, not variety. Whether a camera is counting 2 products or 40, it's the same stream to process.
Or pay nothing: run the counter on your own machine and there is no monthly fee at all.
We quote the monthly once we've seen your cameras during the free evaluation — we'd rather price your real setup than sell you a plan that doesn't fit it.
Three real-shaped setups, priced end to end. Every one of them starts with the free evaluation.
1 camera · counting Coke, Pepsi and water separately
1 camera · 20 different product types
3 cameras · the same 5 product types on each
Note: 5 types, not 15. Extra cameras don't multiply the build.
Cameras don't multiply the build. Three cameras watching the same five products is still five product types — usually the same trained model, so you pay for it once. What extra cameras change is the running side: each one is another video stream to process, configure and keep healthy, which is why the monthly follows cameras and the build follows product types. If cameras differ a lot in angle or lighting, some extra tuning may be needed — we'd tell you that before starting.
These aren't plans to choose between — they're the steps in order: get your footage checked free, pay once to build the counter, then decide who runs it. Own it and run it yourself with nothing per month, or hand the running of it to us.
Free evaluation on one camera. A product type means one distinct product the AI has to be trained to recognise — 500 identical bottles is one type; Coke, Pepsi and water counted separately is three. Build prices above assume normal conditions on one camera at one location; very large catalogues, several cameras, several locations or difficult footage mean more work, and we quote that before anything starts. The counter then has to run somewhere: nothing per month if your own hardware can handle it, or a monthly fee based on your camera setup if we run it for you. Continuous counting depends on your cameras, network and processing hardware — we confirm what your setup supports during the free evaluation. Chains & groups get volume terms. Prices adapt to your market & currency.
Not necessarily. TrueCount is designed to work with compatible existing CCTV. We evaluate your current camera setup and footage first to see whether it can be used — and we'll say so plainly if it can't.
Not every camera system is compatible. Your camera or NVR needs to provide a video stream TrueCount can access, and the camera needs a clear view of the items being counted. Send us your camera details or a short sample video and we'll evaluate it.
There is no single resolution that works for every counting task. What matters most is whether the items are clearly visible in the camera view — a lower-resolution camera close to the action can beat a sharp one across the room. Send us a sample and we'll evaluate it.
That's okay. We'll evaluate your setup and recommend the appropriate option. Depending on the use case, the counting can run on a suitable computer at your location or through TrueCount's cloud. You don't need to buy hardware before we've established what's feasible.
Usually not. If your cameras are connected to an NVR, we evaluate the NVR's available video streams for compatibility. We can't promise every NVR works, so we check yours before committing to anything.
It depends on how your system is deployed. With on-site processing, the counting itself happens on your own computer; internet is used for things like dashboard access, reports, updates and remote support. Continuously uploading video to the cloud is not always required.
One distinct product the AI has to be trained to recognise — not one physical object. 500 identical bottles going past the camera is one type. Coke, Pepsi and water counted separately is three types. The build price is per type and drops as you add more: $150 each for 1–9, $100 each for 10–19, $70 each for 20 or more. You'll never pay more for wanting more — if a bigger bundle works out cheaper, we charge you the lower price. Very large catalogues, several cameras or difficult footage are quoted individually, and we tell you that before any work starts.
Because what costs us to run is video, not variety. A camera counting 40 product types is the same stream to process as one counting 2 — but a second camera is genuinely twice the work. So the build is priced on how much the AI has to learn, and the monthly on how much video we process. It starts at $100/month for one camera; beyond that we quote your real setup after the free evaluation rather than sell you a plan that doesn't fit it. And if you run the counter on your own machine, there's no monthly at all.
The counter we built for you: the trained model, the counting software and your configuration, set up to run on a suitable machine at your site — with no monthly fee to us. What that option doesn't include is the managed side: nobody is watching it for you, reconciling against your POS or sending reports. That's the monthly service.
If your camera clearly shows the item and there's a consistent point where it moves through the view, we can evaluate building a counting system for it — drinks, dishes, trays, ice bags, ice blocks, boxes, packages and other clearly visible products. Clear visibility makes a use case suitable for evaluation; camera angle, lighting, item size, movement and overlap decide the rest, which is why we always test your real footage first.
On real client footage across 5 test clips, counts matched the hand-counted dishes exactly after review — including a 4-dish tray — while ignoring steamers and sauces.
It's built to count dishes and movement, not identities. It's privacy-friendly by design.
You send a short clip of a busy service — a file or a Google Drive link. Within 24–48 hours we send back a short sample clip of it counting dishes on your own footage, plus the rough gap vs your POS — so you see it work before spending anything. The trained system that runs on your footage ongoing is the paid build. No install, no hardware, no commitment.
No. A two-model agreement check ignores steamers, sauce/yogurt cups, empty bowls and items on shoes or aprons.
Worldwide. We currently work with restaurants across the USA, Mexico and France/Europe, in English, Spanish and French.
Most IP cameras and NVR systems provide video streams such as RTSP. TrueCount connects to a compatible stream, runs the detection and tracking either on a machine at your site or in our cloud, and returns counts with timestamps that can be reconciled against your POS or paperwork. Compatibility depends on the specific equipment — send us the make and model and we'll confirm.
Tell us what you want to count and send a short sample of your footage. We'll evaluate your setup and reply within one business day — no obligation.