Video analytics and privacy: what responsible camera AI looks like
Responsible video analytics focuses on counts, flows and behaviour patterns, not on identifying people. It has a clear purpose, a limited scope and no new cameras. If you plan to use CCTV for store insight, privacy should be part of the design from the start.
Customers and staff are right to ask what happens to the video. A good answer is specific.
Why does video analytics raise privacy questions?
Cameras record people. When software starts reading that video, the natural worry is that it will track or profile individuals. That worry is reasonable, and any retailer adopting the technology should be able to answer it plainly.
The issue is not whether cameras exist. Most stores already have them. The issue is what the analysis is allowed to produce. A system that reports how many people entered is very different from one that tries to say who they are.
What does privacy-first camera AI look like?
There is no single checklist, and local law and company policy apply. But a few principles are widely sensible.
- Counts and patterns, not identities. Report how many people came in, where traffic flows and how long visits last. Sakshi works with these measures. It does not claim to identify individuals.
- A defined purpose. Know why you are collecting each measure. Footfall for staffing is a clear purpose. Data collected just in case is not.
- Use what you have. Running on existing CCTV means you are not widening the camera footprint inside the store.
- Limited access. The people who see store data should be the ones who need it to do their job.
- Honest signage and policy. Customers and staff should be able to learn what the cameras are used for.
Staff and compliance monitoring
Privacy matters for employees as much as for shoppers. When cameras are used to check routines, such as the points covered in catching SOP and uniform lapses in real time, the way it is introduced matters.
Tell the team what is being checked and why. Frame the purpose as consistency and support, not punishment. Use the data to spot patterns across a store or time of day, and to fix training and process. Teams that understand the purpose usually accept it far more readily than teams that find out later.
Questions to ask any vendor
Before adopting video analytics, it helps to ask direct questions.
- What exactly does the system report? Counts, flows, dwell, or something more?
- Does it need new cameras, or can it use the ones already installed?
- Who inside my company can see the output?
- How is the data stored, and for how long?
- Can I explain the purpose clearly to a customer or an employee who asks?
If the answers are vague, treat that as a warning. Good vendors are specific. You can also read how the basic idea works in making cameras do more than watch for theft, and how the numbers are used in footfall, dwell time and bounce ratio.
Where Sakshi fits
Sakshi turns existing CCTV into data about footfall, movement, dwell time, headcount and SOP compliance. The focus is on how a space is used, not on who is in it. More than 10,000 cameras have been turned into data this way across our rollouts. The Sakshi section has more on what it reports.
How to get started
- Write down the purpose. List the measures you want and the decision each one supports.
- Involve legal and HR early. Check local rules and company policy before you begin.
- Brief your teams. Explain to store staff what is measured, what is not, and why.
- Limit access. Decide who sees which reports, by role.
- Review after the pilot. Ask staff and managers what concerns remain, and adjust the setup.