Why we publish our numbers

We spent a week reading every video analytics website we could find. Ninety-seven of them, from cloud security platforms to dashcam fleets to the open-source projects people run at home.

Seven publish a price. Fewer than half say what hardware their software needs. Not one publishes how they measure false alarms.

That is not an accident. When you sell through “book a demo”, the number arrives late, after the buyer has invested time, and it arrives attached to a quote nobody can compare. The economics of the product stay private, and so does the fact that many of these products run on a cloud GPU that costs more per month than the cameras it watches.

We have run cameras for a long time. The questions an operations manager asks are always the same three. What does it cost per camera. What do I have to buy. What happens when it is wrong.

So we have written down how we will answer them, before we have an answer. The benchmarks page carries the definitions: what a camera-per-box count has to sustain to be counted, how power is measured, what a false-alarm rate is measured against. No result appears there until it has been run on named hardware, on a date, by a method anyone can repeat.

Three things follow from committing to that in public.

First, we have to keep them true. A published stream count is a promise you can test on your own site in an afternoon.

Second, we cannot hide compute cost inside a subscription. The box is yours. The model runs on it. The cloud is used when you ask a question that needs it, and we say when that is.

Third, comparison becomes possible. Our competitor research is not secret. If a vendor publishes better numbers than ours, we would rather know, and so would you.

The market has been selling fear and demos for a decade. We would rather sell arithmetic.