How Insurance Leaders Can Prepare for Artificial Intelligence Today

Raghav Bharadwaj

Raghav is serves as Analyst at Emerj, covering AI trends across major industry updates, and conducting qualitative and quantitative research. He previously worked for Frost & Sullivan and Infiniti Research.

How Insurance Leaders Can Prepare for Artificial Intelligence Today

There is a consensus among industry experts (both from our own insurance AI secondary research, and according to a 2017 Accenture survey report) that AI is going to be a key driver in making insurance products “smarter” in the coming 2-3 years.

With the advent of artificial intelligence in the insurance industry and its inherent pace of innovation, an opportunity for disruption of business process by AI companies in the industry is slowly taking shape.

In partnership with Insurance Nexus and some of their associated thought leaders, we explore the answers to the following questions:

  • What do leaders in the insurance industry need to know in order to be capable of making an informed decision about AI applications?
  • What’s possible with artificial intelligence and machine learning in insurance today, and which AI-based capabilities will be important for insurance companies to leverage in the near future?

AI in Insurance – Future Trends

According to Mariana Dumont, head of new projects at Insurance Nexus:

“In conversations I’m having with insurance executives, I’ve noticed that we are all very excited about where AI is headed in the insurance industry, but there’s a lot of uncertainty, we don’t know how it will change the core business model, and people’s jobs.”

The latest research report by Insurance Nexus – titled Insurance AI – A Road to Impact highlights three important upcoming AI trends that have a likelihood of changing the industry in the near term:

  • New types of insurance products – Insurance firms will adopt machine learning techniques to derive insights from client data (from IoT sensors for homes, vehicles etc) enabling personalized insurance product development.
  • Pattern recognition for identifying business opportunities and fraud risks – Machine learning will be used to identify patterns in user data (like driver performance monitoring using data from the vehicles sensors and maintenance history) in order to identify new business opportunities and help reduce likelihood of fraud in cases of identity theft, data security breaches and payment / transaction fraud.
  • Improving customer experience – Chatbots will enable an increasingly automated buying experience using customers social data for personalized interactions. This would also enable ‘on-demand’ insurance for specific events.

New Types of Insurance Products

Industry experts agree that in the near future there will be an influx of new product development from insurance firms. This opportunity is heralded by the acceptance of smart home technologies and wearable electronics. The data collected by the sensors in these devices can be input to ML platforms and gain insights about specific customers.

For example in a health insurance application, the data collected from telematic and wearable sensors, can help create a personalized pricing policy with customers paying lesser for leading less ‘risky’ lives (based on a predefined set of data features for risk)…

Another example would be in auto insurance where the data from automotive sensors and maintenance history can be fed to an ML platform in order to develop personalized insurance products with lower premiums for relatively ‘safe driving’…

Pattern Recognition in Insurance

ML based pattern recognition for identifying fraud risks and business opportunities will likely become ubiquitous in the future. One company already applying AI to identify business opportunities is Progressive, in the auto insurance space. The company is using ML to interpret driver data to track market trends and identify business opportunities.

According to Pavan Divakarla, Data and Analytics Business Leader at Progressive, machine learning algorithms are starting to help the company better understand customer data in making predictions about what will happen in the insurance marketplace.

He added that earlier since this was a bottleneck, AI-enhanced pattern recognition is enabling a much faster throughput of the company’s insurance models and improving the generated business value.

Insurance companies file fraudulent claims worth over $80 billion a year across all forms of insurance. Fraud detection is one of the most common and fastest growing AI applications in the insurance space.

There are already instance of startups, like Shift Technology which use AI-enhanced platforms to help insurance companies prevent fraud. Experts agree that most insurance firms will be looking to process hard copies and digitise all customer information (thereby aiding AI integration) in the near future.

Improving Customer Experience

Chatbots have been in existence for quite sometime now, and with continued improvement in  Natural Language processing (NLP) technology, are on the track to becoming inevitable for companies with large customer bases.

In the insurance industry chatbots can help to enable a completely automated buying experience for users by offering services like identity verification and personalized products.

AI in Insurance – How to Prepare for the AI-Based Market Disruption?

With AI being a hype word in most industries now, a lot of insurance companies are interested in AI-based capabilities and trends. There is also a consensus among industry players that AI will be a vastly disruptive technology For example chief of Berkshire Hathaway, Warren Buffet, stated in an interview with news outlet CNBC that autonomous vehicles could eventually become safer than human drivers, implying that insurance premiums will be much lower and in the long run, auto insurers could take a dent in revenues.

Mariana from Insurance Nexus adds that “There is uncertainty of how these changes will happen. Most likely in the long run, AI will bring efficiencies and give companies more resources to invest in other projects.” She mentions that the uncertainty around AI’s impact leaves employers wondering if some work will be automated entirely, and what that means for their teams.

There are several significant hurdles around data management for insurance companies aiming to apply AI, including: Preparing and structuring data, lack of access to skilled labor, and longer research and development cycles for platform development. We consulted the experts from Insurance Nexus’ latest report to garner some perspective on how insurance leaders can overcome these challenges.

Understanding the Use-Case

A large number of current real-life AI applications in the insurance industry have revolved around customer experience enhancement. Yet, due to the current unavailability of comprehensive case studies in the sector, adapting AI-processes to other applications has been difficult for enterprises.

A good starting point for insurance leaders interested in AI would be in improving internal business cost efficiencies. This would enable companies to better understand the intricacies of how to define the objective of AIs in other applications like customer experience.

George Hansen, Group P&C Claims Actuary, Zurich Insurance Group feels that the best ideas of how AI can impact the insurance process will come from outside of insurance. As for advice to business leaders in insurance he adds:

“Pay attention to how other industries are using AI and determine if there is applicability to the insurance process.”

AI Skilled Labor Management

Along with the longer R&D cycles, another huge challenge for companies in this space is the lack of availability of AI-skilled labor. Moreover Constructive communication among different ‘types’ of executives is an added layer to the current talent dearth. Mariana adds:

“There are two types of executives who are leading the adoption of AI in the insurance industry, one are the analytics executives who come from the industry and are most likely very proficient with the technology but do not have familiarity with the insurance industry. On the other side we have veteran marketing leaders in the insurance industry, who have a partial understanding of what AI can do for them”

Any AI technology integration would need technically skilled professionals in an organization to train these AI systems. For example, a company looking to find patterns in customer data to produce personalized insurance products would not only need an (expensive and hard-to-find) AI team, they would also need knowledge that they may not have today, including:

  • How to systematically cleanse and centralize data sources from around the company, so that this information can be used for AI-related initiatives
  • How to hand project to a data science team (a task requiring knowledge of what kinds of problems machine learning can solve, and how data scientists work)
  • How to bear the (sometimes lengthy) process of machine learning experimentation with patience
  • How to integrate machine learning products once a team creates them (a process that often varies greatly from integrating other out-of-the-box software)

Training data scientists to develop subject-matter expertise in the insurance industry is one thing – but training other employees for the types of skills and knowledge listed above is by itself a nuanced endeavor requiring companies to part with even more time and resources.

Matt Marino, AVP, Operational Effectiveness at the Unum Group has a few words of caution and advice for business leaders in this space:

“There is a great deal to learn, the landscape from a technology perspective is rapidly changing, staying aware of these new capabilities and continuously revisiting past challenges and constantly striving to improve is important.  Problems that might not be able to be easily addressed today, may be very simple to fix months from now.

As we see the technology advancing, we also have to recognize the skill sets required to maximize the potential of these tools may be different than the traditional skill sets we’ve relied most heavily upon in our past.  Recognizing this shifting landscape, understanding how competitive the market place can be on these new skills, and being proactive in gently shifting hiring focus to be prepared in a proactive way will help set organizations up for success both short and long term.”

Data Organization and Long R&D Cycles

One of the bigger challenges in AI adoption for enterprises is the long process involved in their development and ‘training’. AI based systems in most cases are not a plug-and-play solutions. Machine learning systems requires data (which needs to be aggregated, cleansed, and organized) and need long calibration cycles to improve the system accuracy to acceptable levels.

For other kinds of software development, a manager (in marketing or finance) might say “Can we have this done by end of June?” and get a straight answer “Yes” or “No.” With machine learning, there’s no telling how long the experimentation and iteration process will take in order to produce an application with real business value, and some projects are unable to come to fruition altogether.

Added to the steep costs involved in developing large AI solutions, the difficulty in integration means that firms need to know if they are financially equipped to adopt such a solution.

Yet, there seems to be a ray of sunshine here according to George Hansen, who agrees that in the future, text mining and natural language processing will allow the industry to leverage data that has historically been more difficult to use. He also adds that, heuristic processes (platforms which can learn to fill in data) will fill gaps in the data which should reduce concerns about data quality not supporting AI.

Organizations that proactively lead with an understanding of their data requirements and what the data can be leveraged for will find that applying new capabilities in the automation space is achievable.

Matt Marino adds:

“I believe we will start to see organizations that lead with an understand of what data can be leveraged for, how it can be most easily digested, and by looking at their data requirements and structures in this manner proactively will ease the challenges associated with applying new capabilities in the automation space.

It is also an interesting proposition when we think about natural language processing (NLP) and optical character recognition (OCR) plays that, in theory, the lack of structured data may be less of a challenge than what some of the earliest efforts encountered.”


This article was written in partnership with FC Business Intelligence Ltd. For more information about advertising and promotional services at Emerj, visit the Emerj Advertising page.

Header image credit: TechNode