Today, underwriting is tedious, time-consuming and inconsistent. With AI, we can realise a better tomorrow.

Leveraging AI to revolutionise underwriting

Underwriting is at the core of the insurance industry. Today, however, the process is tedious, inefficient and leads to inconsistent results.

Human intervention in underwriting is a significant factor, such as with life insurance. Of course, where people are involved, inconsistencies are inevitable.  Many insurers are further hindered by old technology and procedures that prevent them from leveraging Internet of Things (IoT) tools like smartwatches and telematics. These devices provide real-time data on the activity level and driving performance of clients.

The solution to these problems is underwriting powered by artificial intelligence (AI). With AI-powered underwriting, decisions can be reached in days, not weeks. Since AI can derive insights from historical data, decisions can also become more consistent and reliable.

A few insurance companies are using AI in select areas, but there is yet to be widespread adoption of AI solutions across the underwriting landscape.

There are four techniques under the broad spectrum of AI that would provide underwriters with a comprehensive solution and a competitive advantage:

  • Machine learning
  • Natural language processing
  • Deep learning
  • Behaviour data models

Machine Learning

Insurance providers have vast amounts of historical data on applicants – medical history, family history, job data – as well as the risk assessments underwriters assigned to applicants based on the available data and a static set of rules.

A machine learning algorithm is capable of ingesting that historical information and creating a model that will imitate the decisions underwriters have made to that point.

For simpler cases, allowing the model to assign the risk classification would move the industry towards zero-touch underwriting, where decisions are made with zero manual intervention from the underwriter.

For more complex cases, the underwriter’s decision can be made quicker and with sounder judgment because of the underlying model that is there to support them.

As the model matures and the organisation becomes more comfortable with AI, the percentage of applications underwritten with the zero-touch concept can increase. Eventually, insurers could reach an end state where manual or assisted underwriting could comprise 5-10% of their cases, with the rest zero- touch underwritten.

Deep learning

With the advent of deep learning techniques, AI is becoming more sophisticated and better able to use reasoning and problem-solving skills in the way a human brain does. This will help in transforming the underwriting of various commercial risks that are currently being underwritten manually using traditional methods by expert underwriters. It will also enhance decision making and reduce costs (including underwriting losses).

Drones, equipped with infrared cameras, lasers, and sensors help gather data relating to weather, temperature, and environmental conditions, including radiation. They are the game changers that are disrupting the way risk is being assessed in the commercial sector. According to the Federal Aviation Administration (FAA), 7 million small drones could fill the sky by 2020, and as many as 2.7 million will be used for commercial purposes. Here are some examples of the ways drones can be used:

  • Property insurance: inspecting high rise buildings or weather damaged rooftops
  • Agriculture insurance: verifying crops across acres
  • Boiler and machinery insurance: for equipment break down insurance, drones can help collect data from high-pressure machines and unsafe boilers

Natural language processing

Natural language processing (NLP) has two clear uses within insurance. The first is that applicants can communicate with a bot to gather the data needed for an underwriter to make a risk assessment. The bot is able to understand the human it is communicating with due to the natural language processing technique.

The other application is with text- based data mining – currently a manual method for most insurance companies. If a doctor sends in an evaluation of an applicant, that data must be taken from the text document and coded into the system by someone within the underwriting department. With natural language processing, that information can be fed into a bot that instantly creates an electronic form. If machine learning is being utilised, that form can immediately be fed into the algorithm.

In commercial insurance, NLP can be used to better support the underwriters by being their virtual assistants. Apart from the usual data entry, NLP can help the underwriters to pull up relevant information on the risk they are writing using search- based analytics to speed up data access. For underwriting a mine, the underwriters would traditionally look at the location, depth and breadth, safety procedures, number of resources and other relevant factors pertaining to mine including past claims. In the coming days, the underwriter can dictate commands to the virtual assistant which using NLP can intuitively pull up not just the traditional data but also analysis on current developments, literature on environmental risk (or climate changes) and the economic conditions affecting the mine, and an insight into future risk trends giving the underwriter a broader and a complete picture of the mine and the mining industry. Mining insurance being specialty heavy

industry portfolio, the search analytic tool can also be used to provide data needed to make a decision on reinsurance placement.

Behaviour data models

Behavioural data models can be used to analyse the real-time customer data from IoT devices for more precise risk classification and product innovation. Using that data, insurance companies can launch new products that incentivise life insurance customers to lead healthier lives, or auto insurance customers to drive safer. When insurance companies leverage behaviour data to combine a wellness product with traditional insurance, they can tap into a new market segment, a new revenue stream.

Overcoming barriers to adoption

Insurance companies that utilise these AI techniques will see reduced costs due to less time needed for underwriting, as well as fewer mistakes and more consistent decisions. Being able to issue policies faster will allow larger, more established companies to compete with technology-driven start-ups that are taking away their market share.

However, most companies face barriers to adopting these solutions.

First barrier

The first being data capability. Historical insurance data is normally spread across disparate systems or applications, making it difficult for companies to group together true and consistent data for a machine learning algorithm to begin mining that data for insights.

Second barrier

The second barrier is an ecosystem of legacy technology. The existing technology used by most insurance companies is outdated enough that implementing AI techniques would require a significant financial investment, as well as a cultural shift within the organisation in terms of how technology is used in decision making.

Given these barriers, a ‘big bang’ approach is not recommended for companies looking to implement AI techniques within their underwriting department. Instead, companies should opt for a ‘start small and fail fast’ approach to achieve immediate wins and reduce the risks both financially and in terms of their time investment.

Lemonade, founded in 2015, demonstrates this approach by utilising AI with a simple product like renter’s insurance, which applicants can attain within a few minutes using the company’s mobile app.

Ladder is another company that leverages AI to get applicants a quick quote for a fully underwritten term- life insurance policy.

In starting with more targeted use cases, insurance companies can begin to engage with AI technologies to gauge effective use techniques. In time, these use cases will compound, generating sustained efficiencies for underwriters across all sectors of the insurance industry.