Machine Learning Underwriting: Benefits and Use Cases

18 May 2022
8 min read
Discover how to get the most value from machine learning-based intelligent underwriting, its core use cases, and benefits.

Although the insurance industry is known to be resistant to any changes, now it’s experiencing a successful digital transformation. Machine learning (ML) and AI technologies have altered underwriters’ daily activities beyond recognition, improving accuracy, reducing the underwriter process time, and increasing overall machine learning underwriting efficiency.

From better underwriting risk calculation to automated claims processing, ML helps insurers make more data-driven decisions and streamline operations. In this article, we explore key ML use cases in underwriting and their benefits to businesses and the insurance industry. Without further ado, let’s see how ML brings value to underwriting.

Insurance underwriting

Underwriting 2.0: Shift to intelligent underwriting

In insurance, traditional manual underwriting is associated with piles of physical and digital paperwork, back-and-forth communication between clients and agents, entering and rechecking lots of information submitted into the system. This is without mentioning a cumbersome and time-consuming process of assessing clients’ insurance risk to decide whether to issue the contract and under which conditions.

McKinsey has surveyed insurers on how much their underwriters spend time on mundane admin tasks. 35.3% of the surveyed pointed to spending a third of their time on such tasks, while 29.4% specified that almost half of their working time goes to this activity.

Interestingly, such efforts don’t lead to anything more, but human errors, inefficient operations, and increased underwriting expenses, primarily due to the waste of expensive human hours. Instead of automating the underwriting system for insurers and making more efficient use of expert talent, insurers suffer the consequences of manual underwriting.

Still, the rise of ML and AI technology proves that, in a few years, underwriting as we know it today is doomed to failure. According to the GlobalData Engineering Technology Trends Survey, 62% of insurers are already investing in disruptive technologies and at least half of them state that AI and ML will be crucial for businesses in the next years.

It’s not really surprising if we consider how intelligent underwriting algorithms help underwriters better analyze risks, provide more accurate pricing, and process applications faster and more efficiently. With ML and deep learning technologies, the underwriting process time is reduced to a few minutes (instead of hours or even days) because most activities are automated and data-driven. As Matthew Josefowicz, a technology strategist and opinion leader in insurance, summarizes it,

“New technology is allowing underwriters to spend more of their time building and strengthening relationships in the field and, in many cases, acting as the frontline interface of the company with the marketplace.”

Obviously, intelligent or predictive underwriting makes sense, but how exactly ML in insurance can be useful?

ML-based use cases in underwriting

Automate submission triaging

An interesting use case of machine learning in insurance underwriting for insurers includes automated submission triaging, i.e. using ML algorithms for filtering applications and assigning them to underwriters in the most effective way. Predictive underwriting in this case contributes to creating a more efficient working model as ML helps to prioritize submissions based on different factors and make sure that a particular application will be handled faster.

For example, an ML-powered underwriting system can filter the applications regarding their complexity and assign them for review to the underwriter with the appropriate level of knowledge. Another example can include prioritizing submissions based on types of insurance coverages, such as life, health, travel, and property insurance, and then distributing applications to the insurance agents with the corresponding background.

Streamline your submission processing

Handling new underwriting submissions means going through volumes of documents in the shortest deadlines possible. As said, doing it manually takes not only lots of time but is inefficient and error-prone.

The application of ML and AI can yet bring lots of value to insurers in this case, enabling data extraction from DPFs, printed and handwritten documents, and emails automatically. As part of data science, technologies like optical character recognition (OCR) or natural language processing (NLP) are great to help insurers search for the necessary information in the document, retrieve it, and process it if used to build an ML-based underwriting system.

Data extraction
Automated data extraction: Optical character recognition

As a result, using ML-powered solutions gives underwriters a chance to process new submissions more efficiently. Instead of wasting time looking for underwriting information, such as an insured name and address, type of risk, broker info, etc., and entering it into the system manually, agents can rely on automated insurance underwriting systems.

Assess risks more accurately

After the application is filed, the next step is to calculate its risks. The insurance company has to determine whether a particular applicant poses an acceptable risk and approve or reject the submission.

ML-driven underwriting systems will be truly useful for analyzing the submission and calculating the related risks. Underwriters no longer need to rely on applicant-provided data only, which could have mistakes and omissions, intentional or unintentional. In contrast, ML systems can assist underwriting with getting information from alternative and more diverse sources and, thus, promote a 360-degree approach to underwriting risk analysis. ML can help underwriters check an application profile against loads of data points retrieved from sources like social media platforms, bank information, statistical data from public sources, and third parties.

This way, the risk analysis will be more accurate, but the insurance sector can also win on improved operations. Using text classification, ML-powered intelligent underwriting systems can automatically process all the submitted and researched data and contrast it to industry standards. Then, the system will present the results to underwriters for interpretation.

In one example, the insurance company used an ML algorithm for predicting the likelihood of flooding in the area based on historical and geospatial data. The adoption of ML technology into the operations allowed the insurer to increase the accuracy of underwriting services to 83%, generate a 10-fold reduction in throughput time, and increase case acceptance by 25%.

Optimize rates for premiums

Any risk assessment in underwriting ends with defining the coverage amount and premium that the insurer is ready to provide to the insured (on the condition they did agree to provide the insurance at all).

Optimize rates for premiums

However, here’s an interesting observation from the insurance industry: the McKinsey study tells about one small business owner looking for commercial insurance and receiving coverage amounts from five different providers. After comparing the premiums, it appeared that there was a staggering 233% difference in prices.

Getting premiums right in the insurance sector is challenging. Traditionally, industry leaders relied on the “cost-plus” method, which meant calculating the risk premium based on direct and indirect costs as well as a margin uplift.

Although the method is workable, it has its drawbacks, mainly excluding any non-technical pricing factors and being unable to quickly adapt to the market fluctuations. And when pricing rates do not align with market conditions, whether being too high or too low, the outcomes are the same — high underwriting expenses, lost revenue, and narrowing margins.

Now think about incorporating AI and ML technology into your underwriting analysis. In this scenario, pinpointing optimal pricing rates means relying on a broader array of data:

  • Geospatial Information Systems (GIS)

  • Projected real-estate market fluctuations

  • Geopolitical risks

  • Weather and climate forecasting

  • Real-time data from IoT networks

  • Social media

  • And many other data sources that your company can have access to

Aside from getting more fairly-priced premiums, your insurance business can offer more personalized quotes to end-users. For example, IBM informs that today’s most complex AI and ML underwriting systems can recalculate premiums based on specific customer input and circumstances like kilometers driven or goods transported. Thus, AI and ML technology give way to a more client-centric approach to underwriting risk pricing calculations.

Improve coverage recommendations

In the insurance industry, the submission review ends by presenting the applicant with the best coverage recommendation. AI and ML technology have a lot to offer in this case, too.

In our experience, the Intelliarts team has developed ML-based recommendation engines that could help insurers with coverage judgments. Fed with historical dats and previous underwriting submissions, such underwriting solutions could make sense of the most appropriate coverage for the client, with limits and deductibles in each particular case.

The applicant gets a more personalized offer that aligns with their needs and risk profiles, so an insurance company can expect an increased conversion rate in sales and probably more loyal clients in the future. More importantly, the IBM report states about a 2 to 3% increase in premiums as the result of this personalization, which means increased revenue for the insurance company.

Machine Learning for Insurance Business
White Paper
Machine Learning for Insurance Business
Download now

How else ML can be useful to insurers in underwriting

1. More accuracy

The distribution chain in underwriting can be lengthy, with one submission reviewed by several employees. The more people involved, and the more manual work, the lower accuracy becomes in the submission review. People get tired and bored, so they make mistakes — but ML algorithms don’t. That’s why implementing an ML-based underwriting system for insurers will improve the accuracy of your risk assessment. Besides, ML models handle incorrect and superfluous data much better, being trained on historical data sources.

2. Less pressure on employees

Traditionally, underwriting has relied on human expertise, and this is understandable and will remain as it is. Still, the adoption of AI and machine learning in insurance underwriting could “unload” underwriters, taking the most tiresome and repetitive work from them. Instead, employees could focus on more interesting and complex cases of risk assessment.

Sofya Pogreb, the insurance industry leader and the CEO at Next Insurance, comments on the current situation in the following way,

“We believe with technology and machine learning, a lot of [human underwriting] can be done away with. The percentage of insurance applications that require human touch will go down dramatically, maybe 80% to 90%, and even to low single digits.”

3. Better customer service

Customer service is of the utmost importance in any industry, including the insurance sector. As mentioned, using machine learning for underwriting will benefit insurers’ clients with the services delivered faster and more efficiently. Implementing machine learning for insurance companies also brings more precision and personalization to underwriting.

Here are some other examples of how ML can improve customer service in underwriting:

  • Offering personalized car insurance premiums derived from various driving data points and personal driving behaviors
  • Updating the questions and their number that the underwriter should ask when filling out the application forms
  • Delivering more accurate prices in property insurance by using object recognition to check the physical condition, materials, and so on.

Challenges of machine learning underwriting

Despite the benefits we mentioned, we cannot ignore the challenges that machine learning can bring to insurers as well. If your company is going to implement ML-based underwriting, think about:

Challenges of ML-based underwriting
  • The lack of data: The effectiveness of ML algorithms depends on the availability of data. And this must be quality data, without duplicates, null, and missing values. Data fuels ML not only as the “training material” but a way to make the models smarter in the future.
  • Data security: Whenever it comes to massive data amounts, businesses have to consider data security and safety. Make sure your company safeguarded itself from data breaches and leakages.
  • Proper infrastructure: Implementing machine learning also means having proper data infrastructure, as another important element for ML performance. Think in advance (or better discuss with ML experts) where to store and how to manage data in a productive way.
  • Expertise: Implementing an ML model is easier under the guidance of ML professionals, especially if they have relevant domain knowledge. Intelliarts has a proven record of successful AI and ML projects in insurance, and we’ll be glad to help you.

Empowering underwriters with machine learning

With the insurance industry being inundated with more data and the ongoing technology revolution happening in the sector, more insurance companies see the benefit of implementing AI and ML solutions. Underwriters specifically can take advantage of ML-powered intelligent systems by:

  • Prioritizing submissions and assigning them to the most appropriate specialists

  • Processing documents during the application review faster and more efficiently

  • Calculating risks more accurately

  • Pinpointing more optimized rates for premiums

  • And giving personalized offers to their clients

Getting started with ML can be challenging, though. In case your insurance company is looking for a professional team for the full-scale development of an ML-based underwriting system, we as the Intelliarts team are ready to assist. Working according to the CRISP-DM methodology, we take an all-around approach to solution development. Also, If you want to reduce and predict customer churn, we are here to help you.

To get the most efficient results, we invest time in business understanding and data analysis before building the solution itself and deploying the ML model(s) into your existing system.

Get started with ML
Let's talk


Alexander Barinov
Managing Partner
Rate this article
3 ratings
Related Posts