Measuring the customer journey and conversion rate at a gas station using existing cameras relies on three key components: dividing the site into zones (fueling area, store, checkout, car wash, charging stations), counting the number of visits and the duration of stays in each of these zones, and then cross-referencing these figures with checkout data. The store conversion rate is calculated by dividing the number of receipts by the number of visitors who entered the store during the same period.
The advantage of video surveillance over a door sensor is that it measures what happens between the two endpoints of the customer journey. It shows how many drivers move from the drive-through lane to the store, how many stop in front of a shelf without making a purchase, and when checkout lines begin to form. XXII, a French computer vision software provider, generates these types of metrics using its CORE platform by repurposing existing cameras, without facial recognition.
Four metrics are sufficient to build a reliable picture of the site.
Counting vehicles and people by zone provides the entry volume: vehicles entering the service area, people entering the store, and vehicles going through the car wash. The duration of stay indicates the time spent at the pump, in the store, and in line. The occupancy rate of the service lanes and pumps shows actual congestion during peak hours. Finally, the flow of movement between zones describes the path taken: how many drivers who fill up then enter the store.
These four metrics are calculated without identifying any individuals. They yield counts and durations, not individual profiles.
The calculation depends on the conversion you’re tracking. Three metrics are useful and are calculated separately.
The “pump-to-store” conversion rate is the ratio of the number of people who entered the store to the number of vehicles that refueled. This is the most revealing metric of a gas station’s commercial potential: two locations with identical traffic can show significant differences depending on the store’s visibility, signage, and the distance between the pump and the door.
The store conversion rate compares the number of receipts to the number of visitors who entered. It is analyzed by time slot and highlights times when there is a staff shortage, when lines are discouraging, or, conversely, when supply meets demand.
Conversion by store zone calculates sales of a category relative to the number of visitors who passed in front of the corresponding shelf. It allows for evaluating an end-cap location or a promotional campaign beyond just overall volume.
It is not necessary to cover the entire site to obtain actionable data. Four points are sufficient to get started.
The entrance and exit of the shopping aisle, to determine actual foot traffic and its distribution throughout the day. The store entrance, to distinguish between visitors and people simply passing by the window display. The checkout area, to measure line length and processing time. Finally, the electric charging area, if available, where occupancy times do not follow the same pattern as the fueling lanes at all.
The cameras at these four locations were often installed for security purposes, with angles designed to resolve disputes. A security camera angle is not always suitable for counting. A preliminary check of the fields of view, framing, and nighttime lighting prevents problems from being discovered after the system goes live.
The GDPR applies whenever processing involves personal data, and the image of an identifiable person is considered personal data. Two principles significantly limit the scope of such processing.
The first is the absence of biometrics. Counting silhouettes and measuring durations does not require any facial templates. Facial recognition falls under biometric data, which is governed by Article 9 of the GDPR as a special category of data, subject to much stricter regulations. A traffic analysis solution that does not use biometrics avoids these requirements.
The second is data minimization. The analysis can produce only aggregated counts, without retaining the images used to calculate them. The data output from the system is therefore a number of passages or an average duration, not a video sequence.
The usual obligations still apply: informing individuals through posted notices, defining a specific purpose, limiting retention periods, and conducting an impact assessment when the processing warrants it. The CNIL has published a position paper on so-called “augmented” cameras that helps clarify the intended use. Statistical counting and individual behavioral monitoring are not treated the same way in this guidance.
Three conditions determine feasibility. The resolution and framing must allow for the identification of individuals within the relevant area, which requires a camera that is neither too high nor too low. Access to the video stream must be possible—either via the local network or the recorder—with the appropriate permissions. The site’s network connection must support the selected processing method, bearing in mind that on-site processing significantly reduces the required bandwidth since only the processed data is transmitted back.
In a network of stations, the best approach is to test a site representative of the most common constraints, not the best-equipped site. The results obtained at this site provide a realistic estimate of the adaptation work required for the rest of the network.
Metrics are only valuable when linked to a decision. Three uses come up regularly.
Staffing levels, by aligning work shifts with actual peak times rather than with an average. Store layout, by comparing high-traffic areas with low-traffic areas before changing the layout. Monitoring sales promotions by measuring the impact of a promotional event on foot traffic in front of the specific shelf section, rather than just on overall sales.
For a chain, the main benefit lies in comparing locations. A resort where the conversion rate from the slopes to the shop is significantly lower than that of comparable resorts indicates a specific problem: signage, hours of operation, layout, or checkout wait times. Without measurement, this discrepancy remains hidden in sales reports.
Start by defining the decision you want to make, then the two or three metrics that support it. Next, set up monitoring in the corresponding areas at a pilot location for a period long enough to account for weekly fluctuations. The resulting data set serves as a benchmark for rollout across the rest of the network, and XXII’s CORE platform was designed for this type of gradual scaling up across an existing camera network.