From Celent – Applying AI to Insurance Problems

The move to complex problem solving with data, analytics, and AI

Key Takeaway

Insurers are pushing the use of AI beyond the simple automation of rules to more complex, predictive applications.

In these excerpts from Oliver Wyman’s sister company Celent, Craig Beattie, Nicolas Michellod and Zao Wu provide a compelling overview of the kinds of problems AI (artificial intelligence) is being used to tackle in insurance. 

Increases in computing power, greater investments in algorithms amenable to parallelization, and growth in data availability are combining to have two important impacts on applying artificial intelligence:

  • Supercomputing capability in the hands of the chief data officer and data scientists.  Historically, only dedicated R&D departments and universities had access to the raw computing power needed to build, train, and test the most complex machine-learning approaches. Now these algorithms are available in commercial tools as well as open source libraries and can be executed on the cloud without the need to acquire super computers.
  • Democratization of machine learning and the rise of the citizen data scientist.  It is now possible to automate many of the data scientist’s tasks — even the selection and testing of appropriate algorithms, so that an individual with little or no formal data science training can deliver valuable machine learning models and AI tools.

When investing in artificial intelligence (AI), insurers try to address various types of problems with a natural progression that starts with the automation of basic principles, moves to patterns identification, and ends up predicting outcomes and acting with anticipation.

Our observations of initiatives insurers have launched in data, analytics, and artificial intelligence show they have focused on initiatives where they can get tangible results quickly. In general, they have used teams mixing various profiles and roles, including data architects, data engineers, data analyst, citizen data scientists, and also data scientists. Insurers are getting more ambitious in using data scientist teams to solve complex problems.

Definition, prioritization, and anticipation are the key words for insurers:

  • Insurers should define and prioritize the human and technical resource needs and plans to acquire them based on their artificial intelligence target objectives.
  • It is important that insurers anticipate how the market of human and technology resources could evolve to figure out how to acquire resources going forward.

What are the problems insurers face where AI can provide solutions?

  • Distribution and how to propose the best product to consumers.
  • Underwriting car insurance and how to optimize the underwriting margin in auto insurance.
  • Claims and how to better mitigate claims frauds.

The three cases depicted in Figure 4 help us understand how artificial intelligence can assist insurers in improving insurers’ decision-making process:

  • Rule-based automation: The initial task is trivial. It consists in automating decisions through the definition of rules. In our three examples, for instance, insurers can create simple rules to define basic motor-insurance underwriting principles. In claims frauds, insurers use various indicators to define a score to highlight cases that would require further investigation.
  • Pattern recognition: Following this initial step, insurers move away from pure rule-driven automation to analyze relevant data. To be able to position the best offers to consumers, they learn patterns from historical information. In motor insurance, a growing number of insurers are using telematic data to fine-tune their proposition. In claims fraud detection, insurers try to identify claims-fraud patterns to define specific claimant / fraud case profiles.
  • Predicting outcomes and acting with anticipation: The more complex problem consists in observing past data to reliably predict outcomes and act upon them. The time factor also plays a crucial role if insurers want to seize an opportunity. In terms of product offering, artificial intelligence will help insurers propose an optimal set of insurance products in real time to a specific consumer. In addition, insurers can make product recommendations using the now-famous algorithm, “People who purchased this product also purchased this one.” In motor insurance, the objective in the long run will be to go beyond risk underwriting to active risk avoidance. To do so, research in the automobile industry to market driverless cars represents a promising future. Finally, in claims fraud, insurers want to reliably predict frauds. The objective is to use this prediction to actually avoid underwriting a risk at the point of sale.

The complex problems requiring multiple analysis and a prospective approach are also the ones that have a bigger impact on the traditional insurance value chain. Many of them are geared toward shifting the insurance role from a traditional risk payer to risk prevention or avoidance.

Based on the initiatives insurers have launched in data, analytics, and artificial intelligence over the past three years, we observed that a majority of insurers have focused on initiatives where they can get tangible results quickly:

The bulk of initiatives employing data, analytics, and artificial intelligence focus on solving simple problems, as defined in Figure 5. In addition, it is important to observe that, although insurance companies are trying to get fast and easy results, in the last three years, Model Insurers and Asia Insurance Technology Awards programs have demonstrated that insurers are now more willing to tackle complex problems.

Read the complete report, Demystifying Artificial Intelligence in Insurance:  A Tale of Two Models at Celent.com.

Join our InsurTech Community

Learn More