Emerging Usage-based Insurance

Digitalization, big data, the sharing economy, and connected and self-driving cars are all testaments of an ongoing revolution that will change traditional ways of doing things, and impact in particular the insurance market – from risk estimation, tariff assessment, and insurance distribution to customer engagement.

The wide adoption of the smartphone in our daily lives has converted devices into behavioural proxies, which, combined with big data technologies, provide significant improvement in ways to estimate risks for car insurance, and enriching socio-demographic-based tariffs with driving data. This is not new but the telematics black box has become a roadblock to mass market adoption.

Drivies: A New Approach to Usage-based Car Insurance

Following a Telefónica intrapreneurial initiative, the authors created Drivies. Drivies estimates driver risk by assessing each user’s actual driving patterns, and then offers fairer car insurance prices through a prepurchase scoring model. The app leverages gamification techniques including competition theory, comparison among users, and social networking interactions. By providing detailed information of the user’s driving activities and its evolution overtime, Drivies creates a compelling value proposition for each driving user, and a higher-quality acquisition channel for insurers.

Unlike many other app-based telematics solutions, the Drivies business model relies on identifying the best drivers first, before they buy a policy, and not an app that estimates and adjusts the tariff post-sale. This enables the delivery of risk selections and telematics tariffs that are based on true driving behaviours.

Furthermore, Drivies plays well in a marketplace environment. Thanks to collaborative arrangements with local brokers, the app provides insurance options from the main car insurance companies, from low-cost offers to high-value ones. However, the risk profiling data that is collected is exploited in an exclusive way by one single insurance company at any point in time. This way, users are offered a wide variety of preferred insurance brands, coverages, and prices to choose from (including a differential telematics tariff, with better prices for good drivers), while the differential value of the knowledge capture through telematics data is preserved for that specific insurer.

A compelling value proposition is also critical for any app to succeed in our app-saturated times. Most app-based telematics solutions are purely focused on one single insurance proposition – meaning that they focus on providing an optimized price for good driving. However, such price-driven messaging is only relevant for those drivers closer to their renewal date. It also attracts price-driven users, who are typically more prone to churn.

Drivies’ value proposition is actually focused on helping drivers make the most of their driving. The app offers relevant driving information to enable the driver to improve their driving and compete with others for good driving. It also helps the driver earn discounts based on their app usage. The insurance discount is a secondary advantage for Drivies’ users. Such strategy helps us capture high-value customers, who are not entirely price-led.

Technical Differentiations

Building an app that automatically records a driver’s driving patterns creates a series of challenges.

The app must:

• automatically detect the start and the end of a user’s journey;
• handle any positions a mobile phone may be placed into;
• distinguish the different driving manoeuvres (e.g. accelerations, braking, mobile phone manipulation, cornering, etc.);
• minimize battery consumption;
• assess whether the driver is driving or not.

To overcome these technical challenges, we have decided to use accelerometers and gyroscopes (if present), while minimizing the use of a global positioning system (GPS), since a GPS is a main source of energy consumption on the device.

Accelerometers measure the acceleration forces acting over the mobile device, while gyroscopes measure the rotation speed, which is used to find out one’s driving orientation.

When the solution is being deployed, the app may face additional problems:

• Different vendors produce different personalization for the Android operating systems in particular: the same version of an operating system can behave differently on various types of phones depending on the manufacturer.
• Drivies must work behind the scenes. Current versions of the operating systems can create limitations to those background modes.
• Drivies is currently able to overcome many mobile phone-related challenges. Still, battery-related issues exist and our goal is to continuously strive to optimize usage.

Our Spanish Beta: Lessons Learnt 18 Months Post-Launch

Drivies was launched in the Spanish market in July 2015 as a beta concept. Since then, more than 125,000 drivers have downloaded the app, and 24,000 of them have requested a car insurance quote through the app. Almost 40 million kilometres and 3 million journeys have been tracked and analysed through the app.

Product: The Value of Using Fast Iterations and Clear Key Performance Indicators (KPIs) to Optimize Performance

Drivies was developed following lean startup methodology. We constantly iterated and tested our concept to validate our key hypothesis for success, and we still do. A new version of the app is released every two weeks, enabling us to introduce new features or to remove those features that have proved a hindrance to app performance. All customer acquisition activities are tracked on a per-channel, per-week basis and adjusted according to one of our main KPIs: cost per acquired driving user. We use A/B testing to compare new user experiences, functionalities, and marketing messages. These metrics would be useless without the tracking of additional parameters: user acquisition (cost per user), user retention (ratio of users that use the app for at least two weeks), and lead conversion (ratio of users that buy a policy through the app). The application of lean startup and lean analytics brought relevant improvements in Drivies. Figure 1 shows the cohorts of Drivies’ drivers (percentage of users who drove from day 1), and Figure 2 highlights the evolution of our monthly active drivers since January 2016.

Despite targeting genders equally, more than 70% of Drivies’ users are male.

Drivies started with a three-insurer marketplace. We quickly learned that to optimize our sales conversion we needed to open the application to more insurers to offer more options to end-consumers. Today, Drivies works with over 20 insurers. The Drivies app offers a variety of interaction options (chat), prices, offers, and adverts (top recommendations) to the users of the app. These have been pivotal to increase our sales.

Information Brings Unique Insights

App-based insurance telematics, and emerging machine learning techniques, provide great opportunities to enhance risk assessments. They could also potentially disrupt the global insurance market

The data we collected unveiled unique pieces of information from users’ demographic, driving and claims patterns, behaviours, and habits. Attitudes and preferences are reliable predictors of risk. They are actually just as useful indicators as one’s driving skills or one’s demographics. Drivies has changed the way we think about risk, how we compute it and evaluate our customer base at Telefónica. We know that risk assessment is imperfect. However, we also know that one can assess the risk for specific policies from reviewing data patterns across a limited number of days through the app. Our insight is constantly evaluated and updated.

Figure 3 shows how our models have evolved when introducing these new sources of information. We can clearly rank user profiles and policies based on the level of risk. And as we gain more insights we discover new patterns more rapidly.

As the telematics data brings continuous updating to our models, we can systematically evaluate the stability and confidence of our results. This is the power of dynamic data analysis.

There is more to the data we collect than the information it provides. Quality and size are two additional parameters we monitor.

For any big data problem, data structuring and monitoring is a big issue, as well as a great opportunity. As the amount of data constantly increases, the number of analyses that need to be covered rapidly expands, but also our learning capabilities. This enables reliable business insight extraction and facilitates data quality and reliability checks that were previously impossible.

Figure 4 shows several maps of Spain describing the car insurance industry in 2014. In chart (a) we display the rate of claims incurred associated with different regions, whereas in chart (b) we highlight the distribution between insured versus registered vehicles. There are large oscillations from one region to the next. This means either that some regions gathered a high number of insured vehicles with low claims ratio or that the demographic data is biased. Chart (c) shows policy owners based on declared address areas based on policy information. Because pure demographic-based approaches are biased and do not capture the right set of information, good risks are unlikely to get the tariffs they deserve. Our pilot overcame this limitation and identified where the actual driving took place. This is depicted in chart (d).

Conclusions

Drivies demonstrates that an app can not only be a good assessor to measure the driving behaviours of driving app users, it can also significantly improve the assessment of the potential inherent risk embedded within specific risks, on a per-user basis. By leveraging concepts such as gamification, Drivies has shown itself to be a unique asset to acquire high-value consumers and serve them well.