Anyone that uses Amazon or Netflix will be familiar with their suggested algorithms. These learn from the products and films that customers regularly view and purchase to understand their buying or watching habits and provide them with products that correspond to their interests.
Google Adwords also combs search histories to provide the most relevant websites and products to tailor consumers’ browsing habits. While these algorithms are not yet perfected, they are early examples of machine learning and the growing ‘algorithm economy’.
Now, due to the increased efficiency and understanding of data management and easy access storage, particularly in the cloud, machine learning is becoming increasingly sophisticated and prevalent in a range of industries, including car insurance and telematics.
At its core, telematics is simply a way to transmit data on how a car is being driven. However, there is an increasing understanding that the data gathered can be used far more widely than just reporting on driver behaviour. Collecting information on collisions can not only help with fraud detection, but also forms the basis of trend analysis, says Andrew Lee, head of Market Intelligence and Analysis at Octo Telematics.
For example, if there’s a road which has a high volume of accidents, examining how these crashes occur can help understand why this area is so dangerous, and therefore help prevent accidents.
With 65% of businesses using telematics devices and around 750,000 telematics insurance policies in the UK, there is a huge amount of data being gathered. The more data available, the more we can understand. With smartphones, wearables, sensors on cars and the telematics boxes themselves, all connected through the Internet of Things, there is a huge amount of information coming at once, meaning that it could be very hard to sift through and analyse.
Machine learning could be the answer. If systems can ‘learn’ what the important data is and use automated and evolving processes to analyse it, there can theoretically be no limit to the amount of information that can be received and put to good use.
Better storage facilities also mean that there would be no need to discard information that is not of use now, but may be in the future. Even more importantly, when that information becomes useful, an AI that is teaching itself all the time will realise when to retrieve it and incorporate it into calculations.
The benefits are obvious to insurers. They are already looking at telematics data to detect fraud and build up databases of collision data for use in pricing risk. With the ability to analyse even more data points for each accident, they can also look at the type of car and how passengers are affected by crashes.
It also becomes possible to more fully understand the full range of forces at work and be able to build a better picture of each incident. No longer relying solely on evidence from one or two data points, insurers looking to more accurately assess and price risk can take advantage of a far greater number of data points to create precise trend data. Of course, this is of huge benefit to the drivers that are providing the data.
As they provide more information to their insurers, they can create a better, two-way relationship and benefit from both the trend data that is collected, as well as receive policy calculations based on their own, unique driver DNA. This collaboration between drivers and insurers can only lead to even better User-based Insurance (“UBI”) policies. These policies will be fine-tuned to the needs of an individual, rather than a demographic or age group.
However, this information can also be of value to car manufacturers. If drivers are regularly receiving the same type of injury, designers can use the analysis to make their vehicles even safer. If one type of accident is constantly resulting in a broken arm, for example, designers will be able to strengthen restraints or increase cushioning to avoid this.
Incorporating learnings will become doubly important in the design and ongoing development of autonomous cars. As well as safety functions to protect occupants, driverless cars will need to be constantly updating their awareness of their environment and understand not only the conditions, but the driver and incident history of the area they are travelling through. The vehicles will then be able to match their speed and approach to the road appropriately.
In extreme cases, we may even see legal issues arising from the data available and a company’s failure to analyse it properly. If a driver is in an accident and the data shows that they have been driving poorly for some time without flags being raised, stern questions may be asked regarding the purpose of collecting the data and why steps were not taken.
This is already in place for fleet owners subject to the Corporate Manslaughter and Corporate Homicide Act of 2007. Under the Act ‘aggravating factors’ such as deliberate concealment of an illegal nature of activity could mean that employers face huge fines. Insurance companies covering private vehicles could, potentially, find themselves in a similar position. Making sure there are no grounds for accusation will be key.
In any data-driven industry, there has always been a balance between having enough information and having so much that it becomes unwieldy and impossible to see the wood for the trees.
Through applying smart, learning algorithms and systems that can autonomously decide which information to store and which to analyse immediately, it is possible to sort through the volumes of information gathered and continue to build up a wide database of information and trends that will not only generate real value now, but also in the future.
The author of this blog is Andrew Lee, head of Market Intelligence and Analysis at Octo Telematics
Comment on this article below or via Twitter: @IoTNow_ OR @jcIoTnow