What are the best practices for collecting and analyzing driver behavior data?

Although there are several options - Mobile App, OBD based etc, a final decision of which device to go with depends on various factors 1) Cost: Upfront costs vs Scaling costs Mobile app would have higher upfront cost (few millions dollars for building app, data pipelines/analytics) OBDs have higher scaling costs 2) User base: For a smaller book of business, OBD might make sense, but may not for millions of users 3) UW vs Claims: Data from OBD and Mobile app have been historically used for UW purposes and Crash detection alone for Claims. but Dash cams supplement this claims information for at-fault determination and more. 4) Data Accuracy: OBD and Dash cams can provide more accurate data (audio/video, speed, crash, braking etc)

Extracting data directly from a vehicle's Controller Area Network (CAN Bus) has been highly effective, capturing over 100 parameters, varying with the vehicle's protocol. It is essential to choose a device that supports the specific protocol of the vehicle, considering the cost-effectiveness relative to the number of data points extracted. CanBus devices typically yield better and more accurate results due to the extensive range of parameters they can access. However, the hardware component is just one part of the solution; the software component is equally critical its responsible for processing the data and converting it into insights using advanced algorithms. Thus selecting the best software is as crucial as choosing the right hardware

The adoption curve is a common feature in the roll out of nearly every new technology. However, in the case of telematics, incentives have proven to be one of the most effective strategies for increasing technology adoption. There are various methods of providing incentives, each tailored to fit specific business models. At Equiptal, for example, we have developed an algorithm that calculates the utilization percentages of heavy machinery operators. Based on these utilization rates and performance metrics, management awards bonuses to the top-performing operators. This approach not only boosts adoption but also enhanced productivity.

Historically, the primary driver for Drivers to sign up for Telematics programs were incentives. These have been provided in different forms - Discounts on Auto Policy is most common - 5-10% off Rewards - earn 100 points for taking 10 trips on consecutive days etc Gamification - Compare your driving behavior with your friend/family Driving Behavior Feedback/Goals - This is usually associated with Rewards or Discounts, but is more focused towards improving driving habits - 5 consecutive trips without braking events, improve your driving score by 10 points to get 5% more discount etc

Analyzing and optimizing telematics data is crucial to extract meaningful and actionable insights to enhance driver behavior. But, we need to accept reality before delving deep. 1. IoT Data is Not Clean: IoT data often contains errors, outliers, and inconsistencies. 2. Context is Key: Understanding the context of driver behavior is essential. Imagine detecting numerous acceleration events happening almost simultaneously. Do you penalize the driver immediately? Or do you consider the context? Contextualizing Data: Is the vehicle stuck on marshy land? Is it on a snowy patch? Without understanding the context, drawing conclusions can lead to inaccurate insights. For me, Optimizing data is contextualizing it before deriving any insights.

Data validation rules also called Cleansing, Normalization is an important step in determination of a driver score. Why is it even needed?? Thanks to fragmentation. Ex: There are 1000+ Android phone manufacturers and 20000+ Android phone models. No two devices would provide the same information even when they are in the same vehicle side by side. Personally, I have noticed one device detecting 10+ braking events while another device in the same vehicle detecting only 2 or 3. Most important step in creating a baseline is to determine the ground truth. Once the ground truth data is available, how a device data looks in comparison to that data would help optimizing the model

Analyzing driver behavior involves multiple parameters, which can be tricky. However, with the advent of machine learning algorithms and data visualization tools, this process becomes increasingly accurate as you feed and train your algorithm. Data cleansing and setting thresholds are crucial in this analysis, as they help eliminate errors and abnormalities. Ultimately, presenting the story that the data tells through data visualization is an effective way to convey the insights gleaned from the analysis. It has been demonstrated that visual representations facilitate a smoother understanding of information compared to text-based or number descriptions.