What Video Analytics Costs and When It Pays Off

What makes up the cost?

Every video analytics project goes through the same stages that IEC 62676-4 describes for video surveillance systems: planning, design, installation, testing, commissioning and maintenance. Each stage has its own cost line.

Component	What it includes	Cost type
Site survey	Review of tasks and losses, check of the cameras already installed, launch plan	One-time
Hardware	New cameras only where existing ones do not cover the zone. Hardware to run the system if it is deployed on your site	One-time
Monitoring setup	What counts as a violation, in which zones and hours, who receives event cards	One-time per rule
Training an AI model for your task	When the system has to see an object or action specific to your production	One-time per new task
Check on your own events	Before going into daily use, you see how many cards your people confirmed and over what period	One-time
Integrations	With your accounting system, staff entrance, messengers	One-time per integration
Support	Updates, camera health monitoring, adjustments after changes on the floor	Monthly or yearly
Your people’s work	Time to review cards, training of responsible staff, personal data paperwork	Ongoing, your internal cost

Where will the system run, and how does that change the cost?

We deploy the system in one of three ways. What you get is the same: event cards, a log, reports. What differs is the cost structure and where the frames physically sit.

Question	On your site	In your cloud	In our secure cloud
Launch	Hardware and setup	Setup	Smaller upfront payment
Monthly	Support	Support plus your cloud provider’s bill	Subscription
Connection to the site	Video never leaves the premises	A stable link from the site is needed	A stable link from the site is needed
Working without internet	Works	Depends on the link	Depends on the link
Personal data	Frames stay on the premises	Frames stay within your perimeter, under your contracts	We process frames as a data processor, under contract
Scaling	Additional hardware when the current setup is full	Gradual, as the bill grows	Gradual, as the subscription grows

Transfer of personal data to foreign parties is governed by Article 29 of the Law of Ukraine “On Personal Data Protection”. Under Article 24, a processor must protect data just as you do. Your accountant can tell you how the hardware affects your balance sheet.

For a plant with dozens of cameras, on-site deployment is usually cheaper, while the cloud pays off for a few cameras at remote locations without technical staff. We make the final choice together after the survey, taking your security team’s requirements into account.

What drives the price most?

• Number of cameras. The more cameras there are and the more complex the scene each one has to make sense of, the more computing is needed.
• Number of rules. Each rule has to be set up, checked on your own events and supported.
• Ready-made or trained for your task. People, vehicles, license plates and zones are ready out of the box. A new object type, such as a specific container or tool, means training an AI model for your task. This is the most expensive and least predictable part, so we do not promise the result in advance: you see it on your own events before deciding to expand.
• Camera condition. If the cameras already installed are suitable, the hardware budget almost disappears.
• Integrations. Reconciliation with your accounting system or the staff entrance adds work for each system.
• Evidence requirements. Keeping frames longer means a larger archive.
• Site operating hours. A 24-hour plant with a night shift needs rules that have been checked both by day and by night, so the check on your events takes longer than at a site with a single day shift.
• Cards per shift. The more events there are, the more of your people’s time goes into reviewing them and the more every percent of false alerts costs.

How do you calculate payback?

The formula is simple.

Payback (months) = One-time costs / (Monthly benefit − Monthly costs)

The monthly benefit is what you can measure before and after: hours of inspectors who no longer walk the floor, lower write-offs or idle time, fewer penalties from customers. Monthly costs include support and your people’s time to review cards.

Example with notional numbers (c.u. means conventional units; this illustrates the method and is not our pricing):

Item	Base scenario	Pessimistic scenario
One-time costs	100 c.u.	100 c.u.
Inspector time freed up	4 c.u./mo	2 c.u./mo
Loss reduction	6 c.u./mo	3 c.u./mo
Support	2 c.u./mo	2 c.u./mo
Card review time	1 c.u./mo	1.5 c.u./mo
Net benefit	7 c.u./mo	1.5 c.u./mo
Payback	≈ 14 months	≈ 67 months

The example shows how sensitive the result is to the benefit estimate. Measure the benefit during the pilot, on your own data. Calculate card review time as cards per shift × seconds per card × hourly cost. If half of the cards are false, this line doubles, so system precision directly affects payback.

Some benefits are hard to express in money, such as fewer nonconformities at audits or faster retrieval of a recording after a complaint. Describe them separately in words instead of forcing them into a number.

What costs appear after launch?

A video analytics system lives alongside the plant floor, and every change there can require work.

• New containers or a different workwear design change how the monitored items look, and the system has to be adjusted to them.
• A remounted camera or moved equipment changes the picture. Rules need to be reconfigured and briefly rechecked.
• The frame archive grows according to the retention period you set in your video surveillance policy.
• Cameras age and get dirty. The system must alert you when a camera stops delivering a picture. Otherwise a rule silently stops working.

You also need people on your side: someone responsible for reviewing cards every day, an IT specialist for the network and access, and at the start a lawyer for the personal data paperwork. Include their time in the calculation from day one.

How do you avoid overpaying at the start?

1. Start with a site survey. It shows how many of the cameras already installed are suitable and which tasks are realistic.
2. Run a pilot on a few cameras and one or two rules with the clearest benefit.
3. Do not expand the pilot until you see confirmed results together with the number of events and the period.
4. Tie payment for training new models to the result on your own events.
5. Expand to new cameras and rules once the pilot has paid back its share or shown a measurable benefit.

What questions should you ask a vendor?

• Which tasks do you handle out of the box, and which need training for our site?
• What happens if training for our task does not reach the required result?
• How do you show results: on how many events, over what period, on which days and cameras?
• Where will the system run: on our site, in our cloud or in yours, and how does that change the price?
• Does the system need internet to work, and what happens without it?
• Where are frames stored, who has access and when are they deleted?
• What does support include, and how much does an adjustment after rearranging the floor cost?
• Does the system work with the cameras we already have?
• Who makes the decision on an event, and what does the log for an auditor look like?
• Can we take our data with us if we end the contract?

Sources

1. IEC 62676-4:2025, Video surveillance systems for use in security applications, Part 4: Application guidelines
2. Law of Ukraine “On Personal Data Protection” No. 2297-VI, Articles 24 and 29