Problem Description:
The Financial Industry especially Insurance has been undergoing significant transformation in recent times due to IT & digital technological innovations.
Though this change has significantly increased the growth of insurance across the globe, the adoption of insurance is still very less in terms of percentage.
The next wave of transformation – Machine Learning & Technology Innovation will massively transform & imbibe insurance in each human and objects on earth in a few years from now.
The proposed research is to help the Insurers/Financial industry/Technology companies across the globe to understand & prepare themselves for the new transformations ahead.
A point of view on the subject is presented as below for your perusal to come out with in-depth research analysis.
Machine Learning in Insurance
Point of View
Introduction:
700+ years of Insurance industry has seen 3 major eras:
Manual Era (15th CE to 1960) Ô Systems Era (1960s to 2000) Ô Digital Era (2001-20X0?)
One of the most common things in all 3 eras is that the fundamental Insurer’s business has been firmly relying on data analytics all through-out for ages (though the volume of data and purity of data has been significantly improving decade to decade). However, the main difference between the eras is the speed of changes in adoption to leverage technologies to better the way risks are assessed and secure their capital intact.
While spectacular increase in data volumes (thanks to telematics taking the lead), insurance actuaries and experts are spellbound, the main challenge is the existing analytical models and algorithms are not sufficient to do advanced analysis to support insurers, which possibly could be done best only by machines!!
Digital era has seen technology supporting customer engagement, straight through processing, single view of customer, add-on selling, analytics, etc., the next wave of the era will obviously be towards “Machine learning & AI” (also the Top of the Gartner’s Hype cycle). This is indicated to bring in much larger transformation in the insurance business landscape and the way in which insurers will do the business.
The bigger picture: machine learning changing the way insurers do business
Machine learning has not been a new concept, in fact the infamous advanced model ‘Neuron Network’ research was conducted during 1965-70s, which did not kick-off due to the reason that processing power was not enough to effectively handle the long run time.
Most of the insurers have been following “Supervised learnings” for decades – illustratively means assess risk with known parameters applied on available information (structured data & unstructured data) in different combinations to get the desired results.
New age insurers are striving to get to “Unsupervised learnings”, where pre-set goals are defined, if there are changes in the variables, the system recognizes the changes and tries to reset according to the goal set (e.g. GPS map suggests alternative routes dynamically based on traffic conditions). In the insurance industry, this learning is adopted for usage based insurance – Telematics/IoTs are early examples.
“Reinforcement learning” – which primarily depends on Artificial Neuron Network, in which the target/goal will change dynamically depending on the objective. According to the environment, the variable will self-adjust to the dynamic targets/goals, e.g. driverless cars. The insurance world has not yet adopted this style.
Ultimately, the advent of technological innovations such as clothes carrying out pattern recognition and transmitting signals, underwater transportations, Neuromorphic Chips, 3D/4D Printing and Holograms, Insurers are only going to have more opportunities to explore and transform their business models and ways in which they do business in the future.
Machine learning (and advances in data analytics) will change the way insurers (a) Price risk (b) Estimate losses and
(c) Monitor fraud
a) Pricing:
Machine learning will impact three major areas of pricing:
Firstly, determining the price based on possible occurrences of claims. Machine learning could consider the behaviour pattern of insured/ objects across different sources in the similar segment and simulate the illustrative behaviour pattern to assess the price. Machine learning will have the ability to prescribe the pricing based on the future claims.
It will also allow insurers to collaborate with OEMs vendors & service vendors to improve the quality and jointly determine the product price. Machine learning could help Insurers to get the future behaviour pattern of the particular segment of the clients using the products and provide ways in which the claims occurrence could be prevented. This will help the insurer to collaborate for determining the price.
Thirdly, real-time assessment of price using feedback loop of claims record & dynamically changing behaviour patterns in real-time. For example, Machine learning can provide spot discounts on premium when the road traffic is very low.
b) Estimate losses
Machine learning can help the insurer to assess the exact intensity of the accident to assess the loss, which is currently not possible. This will be based on the inputs from IOTs, monitors, level of damage, etc. Machine learning can help the insurer to save huge cost in loss estimation.
c) Monitor Fraud
Machine learning primarily needs identification of object, human and recorded activity to apply algorithms so that it can monitor and detect fraud in real-time (moment of truth).
In the current world, most of the objects (whether it is electronic goods, or furniture or equipment under insurance) already have unique identification marks, but this is a label without intelligence. In future, most of the goods will become IOT enabled, having capabilities to communicate. Likewise, most of the humans are uniquely identifiable (through SS numbers, IDs and biometrics). Though humans and objects identity are less integrated, it is not too far for the convergence of biometric enablement in IOT using block-chain technology to support Insurers in a massive way to monitor & detect the fraud practices.
For example, one of the most vulnerable fraud practice challenge Insurers is dismantling the vehicle and selling the parts & filing a “car lost” complaint and getting insurance coverage. If objects, humans & block-chain technology are integrated together with Machine learning, the fraud can be detected even while the car is getting dismantled & help to prevent loss and associated cost to detect.
Culture change:
Most of the large scale insurers are aware of machine learning & benefits especially in pricing & fraud detections. However, insurers are not foreseeing machine learning as a way to do loss estimation in regular scenarios, but are inclined to make use of machine learning in complex loss estimation scenarios especially in commercial insurance, maybe IoT enablement may take the lead.
The resistance to open way to machine learning has been from different layers & functions of insurance organizations – starting from Underwriters, IT teams, operational management – for various reasons those includes
1) Loss of jobs: One of the surveys mentions there will be significant reduction in people count due to automation like 1 machine doing 3 humans work.
2) Focus on current issues : With many diverse systems, less integrated application environment, cost pressure, pricing pressure, non-availability of required skill sets, etc., insurer’s current priorities are to build the scalable foundation for their current journey. However some of the insurers have set separate teams to collaborate with technology vendors to explore the possibilities to focus on machine learning.
3) Non-availability of resources: Dearth of required skill set – which include a person having multi-dimensional understanding of business & need to spend considerable time.
4) Financial Constraints: As truly speaking, machine learning is still in the development / POC/ research stages which will require considerable time & resources to be available at an affordable cost, many waiting for a better commercial model that could help to achieve early ROI.
5) To be Tried & Tested: Though there are instances where Machine learning has saved millions of dollars, many insurers want clear evidence to adopt.
6) Access to data: Current regulatory restrictions and privacy norms might restrict the data access, without which the accuracy of estimations may be challenged.
7) Training: Machine learning requires a large diversity of training for real-world operation, with real-time examples for machines to simulate various scenarios.
8) Experimentation: Insurers are not very clear when and how to start on this.
9) Infrastructure: A neural network designer has to fill many millions of database rows for its connections – which can consume vast amounts of computer memory and hard disk space. Furthermore, the designer of neural network systems will often need to simulate the transmission of signals through many of these connections – which must often be matched with incredible amounts of CPU processing power and time.
Conjecture:
While Machine learning has the ability to prevent fraud during the moment of happening, current machine learning can certainly determine the likely fraud cases based on algorithms, patterns, behavior patterns. For comprehensive machine learning leverage, there are larger dependencies like support from regulatory authorities to access privacy data, integration of IOT, human identity and block chain technology, IOT capabilities in the insured objects & infrastructure.
