Our client is an Internet of Things (IoT) company specializing in designing devices that provide signal-based services, including a location-based service for vehicles. They currently manage approximately 600,000 devices across various services.
Proactive Battery Maintenance

Service Provided
Client challenge
This client faced challenges in providing customers with accurate, real-time information about the current state of their device batteries. As a result, customers struggled to determine which devices were ready for deployment and usage, as well as to predict when batteries might fail. The client’s primary objective was to address these issues, ensuring both internal and external customers could save time and optimize the usability of their devices. Complicating matters further, the client lacked historical records of device battery failures, including timelines and patterns. With no existing data to reference, it became clear that the solution would need to include approximating normal battery life and failure rates by analyzing patterns within the available data.
Our solution
We discovered that our client had previously attempted to address this issue but faced inconsistent and inaccurate results. To tackle the problem effectively, we began by analyzing the client’s historical challenges. We then designed an architecture to extract necessary fields from the raw data, which contained a vast set of fields, including those required to solve the problem. A significant complexity of this task was managing data in Terabytes.
Our next step was building a multi-step aggregation in Google BigQuery to structure the data, perform efficient calculations, and optimize disk reads and CPU usage.
We developed a classification system, labeling statistical points from device messages as either “warning” or “normal.” This fed into a state designation methodology, classifying each device into a specific state at a given time. Using Google Cloud Platform (GCP), we scheduled daily pipelines to update these structures every morning, producing a final presentation table for predictive analysis and dashboard integration.
From this work, we delivered:
- A Power BI dashboard: Visualizing the current state of device populations across various metrics (e.g., percent dead, replaced, normal), replacing ad-hoc analyses and improving daily monitoring.
- A single-stress survival model: Developed using Python, this model leveraged reliability engineering to predict future failures, allowing the client to project performance months or years ahead and plan proactively.
This solution offered the client more accurate analysis, enhanced decision-making, and significant progress toward a new initiative.
Results
Internal Achievements:
- Forecasted approximately 60,000 device failures over the next two years, projecting a conservative financial impact of $600,000. Additional value is anticipated as the process scales across more devices and with organizational growth.
- Enhanced the battery replacement process by integrating battery health metrics with warehouse testing.
- Leveraged a predictive model to estimate future battery demand and optimize supply chain planning.
- Consolidated battery health statistics with system health data for future maintenance planning.
- Provided critical insights to operations and engineering teams to determine whether future failure rates justify further automation investments.
External Contributions:
- Collaborated with customers to accurately forecast replacement needs, minimizing redundant orders.
- Integrated predictive models within the platform, advancing toward a fully self-service solution.
Additional Case Studies
Fostering Workforce Diversity with Crew
At Crew, we’re dedicated to helping our clients achieve their goals of building a more diverse
Making the Shift to Virtual in Just 24 Hours
Similar to many businesses, our client had to swiftly transition their workforce to a virtual environment
Comprehensive Salesforce Implementation
Our client, a leading national medical device manufacturer, sought a skilled Salesforce partner to assist in