Key Performance Indicators
Introduction and key questions
Our client sought to assess the effectiveness of their current fitness club model. The objective was to analyze foot traffic, member retention, and the efficiency of different club areas to identify optimization opportunities.
To achieve this, we addressed the following key questions:
- How can analyzing member activity and movement patterns over time help evolve the club model?
- What actionable scenarios can be envisioned to improve operations and member experience?
Action #1
Automating Video Retrieval from Surveillance Cameras
To ensure continuous data collection while minimizing human intervention, we automated the video retrieval process via an API. An automated script was developed to retrieve and store video footage at regular intervals, ensuring a reliable and structured data pipeline.
Action #2
Prioritizing Cameras for Optimal Coverage
We selected and prioritized surveillance cameras based on two key factors:
- Field of view: Cameras with a wide-angle view were preferred to maximize coverage while limiting the number of required devices
- Strategic location: High-traffic and high-density areas were prioritized to ensure optimal data collection in critical zones.
This selection process optimized resources while ensuring comprehensive visibility of member movements.
Action #3
Defining Traffic Zones and Key Equipment for Analysis
To maximize insight generation, we defined specific traffic zones and equipment to monitor, focusing on:
- High-density areas along key customer journeys and operational touchpoints
- Critical equipment such as treadmills and weight machines, ensuring targeted tracking and performance analysis
Action #4
Video Processing and Computer Vision Application
Once videos were retrieved, we processed them through a structured workflow:
Frame Extraction: Videos were broken down into still images at regular intervals, enabling detailed, frame-by-frame analysis
Computer Vision Model Application: AI-driven models were applied to detect and count individuals in predefined areas, generating precise data on foot traffic and crowd density
Validation of Detection Accuracy: The AI results were cross validated against manual annotations to ensure high accuracy and reliability
Action #5
Time-Series Management for Data Smoothing
To refine the reliability of detection results, data smoothing techniques were applied, reducing fluctuations and filtering noise. Moving averages and other statistical methods were used to stabilize traffic curves, providing a clearer picture of long-term trends.
Action #6
Equipment Usage Analysis and Traffic Visualization via Looker
Using foot traffic and interaction data, we calculated equipment usage rates and integrated the insights into dynamic Looker dashboards. These visualizations enabled:
- Real-time monitoring of high-traffic areas and equipment demand
- Data-driven decision-making to adjust club layouts and operational strategies
Action #7
Scaling the Model with Google Cloud Platform (GCP)
Once validated at a smaller scale, the model was deployed on Google Cloud Platform (GCP) to handle large-scale data processing. Key tools included:
- Google Cloud Storage for scalable video storage
- BigQuery for efficient data analysis
- Vertex AI to process and refine AI models at scale
This transition ensured high-performance processing and seamless scalability as data volumes grew.
Action #8
Optimization Proposals for Club Layout and Equipment Mix
Based on insights from foot traffic and equipment usage, we proposed strategic adjustments, including:
- Rearranging equipment to optimize space usage and prevent congestion
- Redesigning traffic flows to enhance member experience based on observed behaviors
- Adjusting the equipment mix (e.g., modifying the number of treadmills, bikes, or weight machines) to better match demand patterns
