Running a successful video analytics pilot: objectives, duration, validation, decision

Why start with a pilot

Video analytics depends heavily on conditions on the ground: camera placement, lighting, footfall density, the layout of the space. What works well at one site may need adjusting at another. A pilot exists to replace assumptions with observations, on a small scale and within a controlled budget.

A pilot is not an extended sales demo. It is a structured trial with written objectives, a validation method and decision criteria agreed in advance. Run properly, it answers one simple question: does this solution deliver information that is reliable and useful enough, at our sites, to justify a rollout?

Two roles are worth naming from the outset: an internal pilot owner, who knows the site and will actually use the indicators, and a technical contact who handles network access and camera settings. Without an owner on the business side, a pilot can easily produce figures that nobody looks at, and end without a clear conclusion.

Framing measurable business questions

The starting point is not the technology but the decision you want to make better. "Understand our customers better" is too vague to evaluate. A good pilot question is specific, measurable and tied to a possible action:

• at what times does the checkout queue exceed the level we consider acceptable, and for how long?
• what is the occupancy of the dining room and terrace by time slot, on weekdays and at weekends?
• how many vehicles go through each wash bay, and what is the median cycle time?
• how long does a loading dock stay occupied per truck?

Pinning down the decision you expect

Two or three questions are enough. For each one, write down the decision it should inform (scheduling, layout, staffing, investment) and the level of accuracy needed to make it. Adjusting a rota tolerates more error than an indicator used to compare performance across sites.

Also list the non-video data you will need. Counts, durations and occupancy come from video; receipts, amounts and transactions come from the till or another system. If the pilot needs to combine them, the integration is part of the scope and should be assessed from the start.

Choosing the site, cameras and time period

The pilot site should be representative of your network, neither the easiest nor the most unusual. Too easy a site gives a flattering picture; an exceptional one makes it impossible to extrapolate.

On the camera side, the preliminary audit establishes which cameras can be used as they are, which need reframing and whether a dedicated one should be added. A pilot limited to a few well-chosen cameras beats one that connects many of uneven quality.

The duration must cover your usual variations in activity: weekdays and weekends, quiet hours and peaks, and if possible a special event (a promotion, market day, month end). There is no one-size-fits-all duration; it depends on the rhythm of your business. Allow for a short tuning phase at the start, separate from the measurement period itself.

Finally, sort out compliance before going live: documented purposes, information for customers and staff, and the applicable CNDP formalities. The pilot is a data processing operation in its own right, and your business is responsible for it.

Validating results: manual samples and cross-checking sources

A number on a dashboard is not automatically a correct number. Validation is the heart of the pilot and should be planned from day one:

• pick representative sampling windows: peak hour, quiet hour, evening, weekend, difficult lighting;
• build a manual baseline for those windows, live or from recordings, ideally with two people counting;
• compare window by window, not just in total, because errors in opposite directions can cancel each other out;
• cross-check against other available sources, such as till receipts or payment terminal transactions, bearing in mind that they do not measure exactly the same thing;
• analyse the gaps to pinpoint problem situations (groups, occlusion, backlighting), adjust the settings, then run another round of checks.

Decision criteria: scale, adjust or stop

At the end of the pilot, the decision should follow from the criteria set at the start, not from a general impression. There are three possible outcomes, and all of them are legitimate:

• scale up: measured gaps are within the acceptance threshold, the indicators have already informed at least one concrete decision, and teams are using them;
• adjust: accuracy falls short in certain situations or on certain cameras, but the causes are known (framing, lighting, unsuitable camera) and can be fixed;
• stop: the business question cannot be measured reliably in your conditions, or the indicators, though accurate, do not add enough value. A well-documented stop is a useful result in itself, because it prevents a misdirected investment.

From pilot to multi-site rollout

If the decision is to scale up, the pilot provides the blueprint: a reference configuration for cameras and zones, a validation procedure and agreed indicator definitions. Each new site still needs its own check, since different framing, lighting or entrances can affect accuracy. Spot checks when each site goes live, followed by periodic checks, keep indicators comparable across sites.

Rollout is also the time to standardise hardware (Edge AI units, additional cameras), fleet monitoring and dashboard access rights. VIDEO.MA supports this process project by project, from feasibility study to pilot to rollout; our SaaS platform, which will make multi-site management easier, is being prepared for a later phase.

To learn more, see our Solutions, Custom solutions and Hardware pages, or request a demo to scope your own pilot.