Insurance Analytics: Applying Analytics in the Underwriting process has been saved
Insurance Analytics: Applying Analytics in the Underwriting process
Data Analytics within the Dutch Insurance industry
Insurers have invested in Data Analytics but see a limited return in business value”. This is one of the outcomes of a research amongst Insurers in EMEA. This fourth blog will give a concrete example of how Deloitte’s Predictive Modeling approach resulted into an improved underwriting process for an International Insurance firm in the Life Insurance industry.
Go directly to
- Life Insurance Value Chain
- Our Approach
- Results and Insights
- Conclusion and follow up
- More information
Over the last years most insurers have invested in Data Analytics solutions and understand that investing in these solutions is key to survive in a fast changing environment. However, a recent study among 68 EMEA Insurance companies showed that 90% of interviewed firms struggles to see a positive business case on Data Analytics solutions. Insurance companies are facing multiple challenges that prevent them from reaching the potential of these solutions:
- Data Analytics experts are scattered across the organization; each unit or function has their own expertise and activities are not optimally coordinated.
- There is a gap between Data Analytics expertise and business sense.
- Data Analytics solutions are not implemented into business processes, therefore using the solution is too cumbersome and the business stops using it.
- The value of Data Analytics solutions is not defined or not measured structurally, therefore it is unclear if the investment and maintenance is justified.
- There is no company-wide vision and strategy for Data Analytics, therefore direction and drive for initiatives is missing.
- New technology developments like Big Data and AI give even more potential of using Data Analytics. Insurers feel that they have to jump on to not get behind of competition or behind of InsurTech startups, but forget that in order to profit from these technologies they will need a solid Data Analytics capability first.
In our first blog different operating models for Data Analytics capabilities were described, including their respective pros and cons. It was explained that there is no one size that fits all.
In the second blog the process for setting up a business case for a Data Analytics organization was explained and examples of impact and required investments were given.
In our third blog a concrete example was described of how Analytics has delivered value to our customers: our Operations Analytics approach which was implemented in multiple Dutch and International Insurance organizations.
In this fourth blog we will discuss another concrete example of how value was delivered with analytics to our clients. We will describe the importance of closing the loop of the Life Insurance Value Chain and how we have fed back learnings from actual Claim data back into the product design and underwriting process for an International Life Insurance firm.
Closing the loop of the Life Insurance Value Chain
To explain how we have delivered value in the Underwriting process we first take a step back to describe how we view the Insurance Value chain within Life Insurance firms. The typical life cycle of an insurance product consists of the following:
- Product design and marketing
- Underwriting & Pricing
- Distribution & Administration
- Claim handling & Cover Charges
- Service, communication and churn
This typical life cycle does not account for measuring the activity, the progress or the performance of an industrialized insurance product. Also, the quantified learnings from the life-cycle of an insurance product are seldom fed back into the design and marketing of new or revised products. With that, the process remains incomplete; i.e. the process stops after the development and release of Insurance products.
One of the ways to close the loop is to feed the information gained from processing claims back into the underwriting process. For a long time, it was hard to measure the relation between loadings and claims due to a lack of data – both in quality and quantity. This data was either not being collected, or hidden away in mainframes that do not provide easy access. However, with data being more and easier available and with the availability of better analysis tools, an opportunity arises. It is now relatively easy to compare loadings with actual historic claim data and analyse various characteristics and mismatches, without needing sophisticated technologies and within relatively short timeframes,
Evaluating the relation between loadings and claims and understanding characteristics that drive mismatches will enables these organizations to better price risk.
Use case on comparing loading versus claims
As an example, we will elaborate on how we connected claims and loadings for an international Insurance Firm in the Life Insurance industry. This Insurance firm had difficulties in pricing its products competitively and defining the right loading for individual consumers. Previous attempts of connecting claim data back to loadings were unsuccessful due to difficulties in data transformation and data quality, and not having the right data transformation and analysis skills.
To set the direction for our analysis, we had posed the following business questions together with the client:
- What, if any, are the relations between loadings and the corresponding claims experience?
- Is there a correlation between loadings and claims, and how significant is it?
- Are there any benefits of closing the feedback loop?
To answer on these business questions, we used our CAP (Certified Analytics Professional) based approach. This approach is the Deloitte NL standard for Predictive modelling projects and is based on best practices from the independent premier global professional certification institute for analytics practitioners. The image below shows the steps of our CAP-based approach.
The reason to use this approach is to assure quality of the end result by applying a well-structured methodology. When building analytics solutions it is very easy to get lost in the data and focusing on building the best model only. By forcing to set up for example a data ETL document and a model approach, Data Scientists will need to focus on the quality of other important Analytics steps and it becomes easy to check the Model quality for a second person. Also, it helps with clear definitions of the business problem and keeping in mind this business problem throughout the whole modelling process.
To explain how we applied this approach for our case, we already elaborated on the business questions that were defined in the problem framing step. In the next step, the model approach, we defined what type of models could help us in answering these business questions. We decided on a two-folded approach:
- Data discovery on loadings versus claims, to analyse differences between them and getting a better feeling on characteristics that drive the difference
- Predictive model for loading predictions, with the actual claim data as training labels, to check if we could make better loading predictions when accepting a new customer
After having defined the model approach, the required data for our analysis was collected. This step consists of defining what data is needed, talking to data owners to describe and validate the data understanding and transforming the data in an analysable format. Claim data was connected to customer and product data. Since we wanted to create a predictive model that we could apply in the underwriting process, we only used data that is (possibly) available in new insurance applications, with the only exception being the actual claim data that we used for training labels and our value to predict.
When the data was ready for our analysis we first built the data discovery interface and the predictive model as designed in the model approach. Finally, we looked at the results and tried to answer our business questions based:
Results and Insights
The analysis gave lots of interesting insights in the loading versus claims and the predictability of the best fitting loading. In this section, we will describe one of these insights.
The data discovery dashboard made it possible to visualize how different loadings and exclusions (based on characteristics such as employment, health, workplace hazards, or lifestyle) provide different levels of protection against the increased risks of claims. To do this, a dashboard was set up in which more than 4M rows of data were loaded, containing both historic claim data records and all characteristics of customers. Data discovery tooling, like Qlik, Tableau and PowerBI allow for in memory loading of data and with that very fast filtering, aggregation and drill-down.
For example, visualizations were made on which the “Premium Increase Ratio” was plotted against the “Relative Claim Ratio”. The premium increase ratio means the relative increase in premium compared to the premium asked to ‘standard’ customers. A premium increase ratio of 2 means that the premium of these customers is twice the standard premium. The Relative claim ratio is the claim amount compared to claim amounts of ‘standard’ customers (with standard Premiums, so a premium increase ratio of 1). A relative claim ratio of 2 means that the claim amount of these customers is twice the claim amount of standard customers.
Each new client of this Insurance product is categorized into different risk categories. For example, some clients that face hazardous environments or have a pre-existing medical condition might have exclusions in his or her policy to make claims resulting from this hazardous environment or medical condition. Other clients that for example have a risky job profile might get a Loading profile, which means a higher premium. The risk categories are determined by actuaries, the categorization of new customers to this predefined risk categories is done in the underwriting process.
The visualization shows for example that:
- Category Exclusion 1 (“A – EX1”) provides an excellent protection against claims: the premium increase ratio is approximately 1,1 while the relative claim ratio is approximately 0,7. This means that while the premium of customers with this category is increased by 10% compared to standard customers, the claim ratio is actually 30% less. This also shows that there might be an opportunity for better pricing to these customers (so lowering the premium increase ratio)
- Category Loading 4 (“A – LOAD4”) is providing insufficient loading: premium increase ratio is approximately 1,3 while the relative claim ratio is approximately 1,5. This means that while the premium is increased with 30% compared to standard clients, the claim amount is 50% higher
- Ideally, all categories should be exactly on the blue diagonal. This line stands for equal premium increase ratio versus relative claim ratio. Therefore, this line also immediately gives insight into which loadings do or do not provided a sufficient premium increase versus the increased risk of claims
These insights could for example be used as followed in future product design; thus entailing the benefits of closing the loop in product design:
- Extra-premium loadings which do not provide sufficient premium increase versus claims can be re-evaluated by the actuarial team. Do you increase the premium, or convert it to an exclusion?
- Extra-premium loadings which generate more premium compared to their relative claims (i.e. loadings which are above the blue diagonal) can be kept as-is (revenue generators), or reduced prices to improve your competitive advantage
- Exclusions which do not provide sufficient protection against claims (i.e. exclusions which are not near the 1.0 claims ratio threshold) can be scrapped in the underwriting process to improve operational speed and reduce costs.
Conclusion and follow up
This article is the fourth in a series of blogs on Data Analytics in the Dutch Insurance market. This article shows how value was delivered by applying data analytics within the underwriting domain of a life insurance organization. By analyzing and visualizing loading versus actual claims, insurers can better price their risk.
For more information, please contact Merijn Slegers or Abhishek Choudhury via the contact details below.