Written by Oren Atia
Insurance companies have relied heavily upon the Generalized Linear Model (GLM) for many years, and some still utilize it today. Before AI revolutionized our lives, implementing this model was achieved by categorizing users into groups based on factors such as age and cost of asset; these parameters were then applied to a formula that used limited computing power in order to draw conclusions.
GLM has experienced a remarkable evolution, now allowing for an exponential increase in the sheer volume of data that can be processed to create personalized differential risk per customer. Employing this technology without grouping data (meaning, that the information that was once grouped is now stand alone data) allows for much better accuracy: more singularized information enables precise and efficient analysis.
There is, though, an important caveat: there are currently models that present better performance in relation to the information that insurance companies use.
At Seenity, we recognize that “Risk” is composed of these essential components:
The asset being covered (e.g., a car)
Those responsible for it
Each of these elements can then be further divided into numerous risk factors. And it’s important to note that – while the data associated with an insured asset is usually straightforward and objective – The circumstantial factors can be vast and varied.
Let’s look at car insurance as an example. Thanks to modern vehicles being equipped with innovative security systems that prevent damage and accidents, there is a clear influence of the user when it comes to the underwriting process and other inquiries are made about their past insurances.
When looking at circumstantial risks though, what should insurance companies consider? Every risk has its own components but for example, in terms of natural disasters such as floods, weather patterns and topographic altitude will likely have strong correlations. Additionally, if we look at property damages caused by crimes rates in the area will also be highly relevant.
But there is still one key question: what is the rule of thumb when gathering circumstantial data for a model?
We have found that any data point related to location and time should be considered, as it has relevance in assessing circumstantial risk factors. Philosophically, any element in the universe that has both a location and a time can be seen as having value in constructing a risk assessment model with circumstantial considerations.
The GLM model featured at the start of this article offers a great degree of flexibility by allowing any data that satisfies its threshold conditions – time and location – to be incorporated.
The power of models today lies in their ability to incorporate a multitude of data with no need for pre-grouping; enabling these models to train and provide better results.
Seenity specializes in the development of risk assessment models, leveraging circumstantial data with time and location indications. This data has relevance in two scenarios – during the model’s creation and when performing predictions in real time.