At kasko2go, where I have been a part of the team since 2020, we are not just another name in the insurance industry. Our mission has always been clear and ambitious: to mitigate risks in the insurance sector. If you're already acquainted with us, you know that our commitment to this goal is unwavering.
When I first joined kasko2go, I was captivated by a bold vision: "Usage-Based Insurance (UBI) is the future of car insurance!" But having been around the block a few times, I knew the importance of digging deeper. Through numerous discussions and analyses, I came to understand the essence of our mission - the creation of a risk score. This score reflects the risk an insurance company assumes when onboarding a new customer. Simply put, a lower score indicates a higher risk associated with the policyholder.
Confronting such a challenge, my initial step was to embark on a steep learning curve, arming myself with the right questions. As an outsider to the UBI domain, I lacked the luxury of direct access to the data for analysis. Thus, my decision to join the company was largely based on asking critical questions that could be answered without delving into the UBI data.
In this article, I aim to share those pivotal questions that guided my decision-making process. My hope is that by sharing these insights, I can assist you in making more informed decisions about UBI - whether it's to fully embrace this technology, utilize it in a specific manner, or even reconsider its application, particularly from a risk assessment standpoint.
Note: My perspective is primarily technical, so this discussion will focus on the solution's viability. However, I will touch upon business viability towards the end of the article.
.jpg?width=600&height=400&name=car-dashboard-wheel-close-up%20(1).jpg)
1. How are you defferent from your competitors?
When exploring UBI/Telematic solutions, one crucial question stands out: "How are you different from your competitors?" This question is vital in assessing which provider aligns best with your needs. Generally, UBI/Telematic providers offer one of three main solutions:
a. Device-Based Solution:
This approach involves embedding a device in the car or connecting it to the driver's phone. An OBD2 device, plugged into the car's diagnostic port, gathers extensive telemetry data (e.g., sensor readings, pedal usage, steering patterns). Alternatively, phone-connected devices typically provide location and, in advanced cases, acceleration data.
Pros:
- No need for drivers to install apps.
- Constant presence in the car ensures data capture.
Cons:
- Costs associated with the device.
- Maintenance and service requirements.
Insurance companies aren’t equipped for physical device sales and support, often leading to significant expenses and operational challenges.
Conclusion:
The device-based solution is beneficial for its reliability and direct data collection. However, the significant costs and maintenance, combined with the operational challenges for insurers in handling physical devices, make it a less feasible option for many. Insurers must weigh the benefits of accurate data collection against the practicalities and costs of device management.
b. App-Based Solution:
This model either involves a standalone app provided by the UBI provider or integrating their software into an existing app. The data is collected using the smartphone's sensors. Initially, the variability in sensor quality, especially in budget phones, impacted data accuracy. However, advancements in technology have mitigated this issue.
Pros:
- Eliminates the need for physical devices.
- Easy installation and no physical integration with the vehicle.
Cons:
- Data may be less accurate than from dedicated devices.
- Drivers can deactivate the app at will.
- Insurance companies typically struggle to engage users with their apps, as regular interaction with insurance services is uncommon.
Conclusion:
The app-based solution excels in its ease of use and avoidance of physical devices, making it more user-friendly. However, the challenges lie in the variability of data accuracy and user engagement. Insurers must consider whether the convenience of an app compensates for the potential lack of consistent use and the inherent variability in smartphone sensor quality.
c. Connected Car Data Collection:
A newer approach involves sourcing data directly from car manufacturers. This method is gaining traction and is possibly the most efficient in terms of data collection and user convenience.
Pros:
- Eliminates the need for additional devices or apps.
- Provides highly detailed data.
Cons:
- Complex process of obtaining data from manufacturers.
- Limited to newer car models.
- Diverse protocols and formats across different manufacturers pose logistical challenges.
Conclusion:
This solution represents the cutting edge in UBI/Telematic data collection, offering detailed and reliable data without the need for additional devices or user intervention. However, its complexity, the requirement for agreements with numerous manufacturers, and limitations to newer car models make it a challenging option to implement universally. Insurers need to evaluate if the high-quality data and user convenience are worth the complexities and restricted vehicle applicability.
In essence, each solution offers unique advantages and faces distinct challenges. The choice depends on balancing the priorities of data accuracy, user convenience, operational feasibility, and cost-effectiveness.
2. What is the accuracy of the data?
The accuracy of data is a fundamental aspect to consider when selecting a UBI/Telematic provider. Understanding the limitations and reliability of the data they provide is crucial for effective decision-making. Here’s a breakdown of what to expect:
GPS Data: The cornerstone of most Telematic systems, GPS data is provided in various formats depending on the solution. With device-based solutions like OBD2 or data directly from car manufacturers, you get comprehensive Telematic data, including information from the car’s onboard computer. In contrast, app-based solutions primarily offer GPS data and occasionally accelerometer data.
System Limitations: It’s essential to recognize that these systems are not flawless. A typical structure involves a data collection device that transmits data to the internet, either through a cellphone or a modem. However, there are situations, like driving through a tunnel, where you lose connection. This loss affects not just your ability to stream a podcast but also the transmission of crucial Telematic data. While GSM (cellular) connections might hold, GPS connections are often lost in such scenarios.
Data Gaps: These connectivity issues inevitably lead to gaps in the data. A proficient Telematic provider is aware of these challenges and should have strategies to mitigate them. If these gaps aren't properly addressed, the quality and usability of the data can significantly deteriorate. However, with effective handling, you can expect to receive around 80%-85% of the data from your fleet.
Expectation of Completeness: It’s important to set realistic expectations regarding data completeness. Achieving 100% data capture is currently beyond the scope of these technologies. There will always be some level of data loss due to various factors, including connectivity issues and environmental constraints.
3. Are you scoring for me or just providing me with data?
In the realm of UBI/Telematic services, it's important to discern whether a provider is offering just data or including a scoring system as well. As we've seen, there are three types of Telematic providers. Typically, the first two types - those offering device-based and app-based solutions - tend to provide scores. The third type, which focuses on bridging car manufacturers with Telematic consumers, usually does not engage in scoring, primarily because their business model is centered around data connectivity rather than risk calculation.
When a provider does offer scoring, it's crucial to understand the mechanics and implications of their system. Scores are often derived from parameters indicative of "driver behavior," such as data from GPS and accelerometers, analyzing aspects like speeding, braking, acceleration, and sometimes cornering.
This approach, however, raises questions about the direct correlation between such "driver behavior" data and the actual on-road situations. While some companies might argue that patterns of sharp acceleration and harsh braking indicate risky driving, thus lowering the score, this isn't always a foolproof assessment.
The discrepancy arises in the definition of a "bad driver." For a Telematic provider, a "bad driver" might be one whose driving patterns trigger extreme sensor values. However, for an insurer, a "bad driver" is defined more in terms of insurance claims – someone who is more likely to claim more than what they contribute in premiums. This distinction highlights a significant divergence in perspective. Both the insurer and the Telematic provider might use the term "bad driver," but they apply it in entirely different contexts.
This understanding is crucial in determining the role of your UBI/Telematic provider and how their services align with your insurance objectives. Knowing whether you're receiving just raw data or a data-driven score - and how that score is calculated - will guide your decision-making process and strategy in leveraging UBI/Telematic technologies.
4. How much data do you have to prove that your score works? And how do you define a score that works?
In the realm of UBI/Telematic services, understanding the volume and quality of data a provider has to validate their scoring system is crucial. This becomes particularly significant when aligning the provider's concept of a "bad driver" with the insurer's perspective for effective risk assessment.
Actuaries at medium to large insurance companies often require extensive data to validate risk models, typically data from 100,000 policies over five years. While this is manageable for traditional risk models with decades of data, it poses a significant challenge for Telematic companies. Gathering such a volume of data involves not just deploying 100,000 devices or apps annually but ensuring consistent engagement from users. Moreover, this data should ideally be homogenous in terms of demographic and geographic factors to maintain consistency and accuracy. This process must be repeated over five consecutive years, making the task even more daunting.
When focusing on app-based data collection, the financial implications are considerable. Costs include not only incentives for app installation but also continuous efforts to ensure user engagement. This investment becomes a prerequisite before even assessing the effectiveness of the technology for risk evaluation.
The uncertainty extends to whether any Telematic providers currently have enough data to conclusively prove the effectiveness of their systems. This lack of definitive proof challenges the viability and potential benefits of these expensive and complex projects.
This requirement for extensive data also complicates pilot projects in Telematics. To effectively test whether the Telematic product meets an insurer's needs, a large-scale deployment is necessary, conflicting with the typical cautious approach of pilot projects aimed at minimizing exposure and reputational risks.
In conclusion, insurers face the challenge of balancing the potential advantages of a Telematic system against the significant investment in data collection and the risks associated with pilot projects. This balance requires a careful consideration of the provider’s ability to not only gather but also interpret data in a manner that aligns with the insurer’s criteria for risk assessment and operational practicalities.
5. How do I use it in my risk assesment models?
The question of integrating telematics data into existing risk assessment models is a technical challenge that pushes the boundaries of traditional actuarial science. Traditionally, actuaries rely on structured data tables and employ various mathematical techniques like Generalized Linear Models (GLMs), Random Forests, and XGBoost to identify factors indicative of higher claim risks. These models, however, are based on finite parameters like age or car make and model. These parameters have a limited range of possible values – a person can't be 1000 years old, and there's a finite number of car makes and models.
In contrast, data from telematics providers is seemingly unbounded. You can record an infinite number of events for each driver, which introduces a significant complexity: the issue of dimensionality.
To put this into perspective, consider a simple risk assessment based on age, with age groups ranging from 18 to 85. If you need a sample of 100 policyholders for each age group to draw meaningful conclusions, that's 6700 policyholders in total – a manageable number for an insurance company. Add another variable like gender, and the requirement doubles, as you now need 100 males and 100 females in each age group, totaling 13,400 policyholders.
Now, introduce a variable with infinite possibilities, like the telematics data. The question arises: how much data is required to maintain the same assessment quality? The answer isn't straightforward. The challenge is not just the volume of data but also its complexity and variability. Each additional variable from telematics data (like driving speed, braking patterns, time of day, etc.) exponentially increases the complexity of the model. This complexity can quickly escalate beyond the manageable capacity of traditional modeling techniques.
The practical implication is that incorporating telematics data into risk assessment models requires advanced analytical techniques capable of handling high-dimensional data. It also demands a significant increase in data volume to ensure statistical significance and accuracy. Insurers need to consider these requirements and evaluate whether their current systems and methodologies can adapt to this new level of complexity. Furthermore, the task involves not just collecting and processing data but also interpreting it in a way that aligns with established insurance risk models and business goals.
In Conclusion
Embarking on a UBI project requires thorough consideration and understanding of several key factors:
1. Financial Commitment: Be prepared to invest significantly, possibly hundreds of thousands of dollars, in deploying and maintaining a vast number of apps or devices.
2. Data Usability Costs: Beyond the initial investment, additional resources will be necessary to ensure that at least 80%-85% of the collected data is usable.
3. Scoring System Relevance: Be aware that the scoring provided by the UBI system may not align perfectly with your specific needs or expectations regarding risk assessment.
4. Data Volume and Duration: The quantity of data you need to collect, and the duration over which it must be collected, may be extensive. This requirement can make limited-scope pilot projects challenging, as they may not yield sufficient data for meaningful analysis.
5. Dimensionality Challenge: Even with adequate data collection, the complexity of telematics data ('Dimensionality Hell') could pose significant challenges in effectively utilizing this data within your existing risk assessment models.
Bottom line, a UBI project involves a substantial financial outlay and carries a high risk of yielding data that may not be fully usable or applicable to your objectives. It’s crucial to keep these factors in mind and conduct a detailed cost-benefit analysis before initiating such a project.
