Implementing Flow Vision Calibration of WiFi KPI
1. Background
Location WiFi is an excellent tool for understanding general trends in shops and malls. However, WiFi has limitations in delivering consistent specific numbers. These limitations have recently been exacerbated by changes in how mobile devices interact with WiFi networks.
As an essential evolution, Flow has released a calibration methodology that combines WiFi monitoring with advanced computer vision and big data regularization-based machine learning techniques to ensure accurate trend data and accurate specific values in WiFi-enabled locations.
At the end of November 2020, Flow Vision Passersby functionality was released for several existing clients to gather ‘ground truth’ data to constantly calibrate WiFi data gathered from existing Cisco Meraki access points. Using proprietary machine learning algorithms, we calibrate the passersby trends and values for all locations, even if the location is not equipped with Flow Vision. (See methodology below)
2. Affected KPI
Passerby - The total number of people passing in front of a location
Capture Rate - The percentage of passersby entering a location.
3. Methodology
Using AI clustering techniques that take into account all available KPI, locations are clustered into groups (clusters).
Candidate locations are identified (Masters) and Flow Vision running on IP cameras is then deployed. For chains with a very wide variety of shop styles cameras may be required in up to 20% of locations. Typically 10% of locations require cameras.
Ground truth data from the Master location for each cluster is used for constant WiFi calibration for all locations (Nodes) in its cluster.
* Calibrated data will be reported according to the client's current data refresh timeframes.
To ensure the highest accuracy, the Master locations also serve as the essential ground truth source:
1 location out of the Master Locations is taken apart.
The calculations for ‘passersby’ using FV adjustment logic on the WiFi data of the selected location are performed using solely data from the other Master locations.
The calculated numbers are compared against the numbers obtained from the IP camera to compute the accuracies.
The process is repeated for all Master locations.
Accuracies are averaged.
4. Accuracy
After the calibration process, the following passersby accuracy rates can be expected. Accuracy Rates are an average of all locations for an active 150 location chain.
| Uncalibrated WiFi Trends | Calibrated WiFi Trends | Uncalibrated | Calibrated Specific WiFi Values |
Master Locations | 90.3% | 99.9% | 1605.2% | 84.4% |
Node Locations | 90.3% | 92.1% | 1605.2% | 134.3% |
5. Adjustment of Past Data
For existing clients newly adopting the Flow Vision calibration methodology, there will be a significant change in actual values reported. To allow for meaningful YoY comparisons, Flow recommends that all data be brought into line. This process is powered by the newly acquired ground truth data from Flow Vision, past WiFi data and AI estimation algorithms.
We can enable this adjustment at the client’s convenience. We will consult with you regarding this process and the implementation timeframe. This service will be provided free of charge.