About five years ago, I accidentally came across an article. In that article, they analyzed, from a cross-industry perspective, which sectors would be most affected by the advance of robotics, automation, and the emerging presence of artificial intelligence.

Seeing that ranking of “impacts” and its forecast made me—after all these years—ask myself how true or false it has proven in the present, and, in any case, even more interestingly, what the “conversation” is today that these (or should I say those) disruptive technologies are having with our industry in general and the underwriter’s work in particular, and what this “today” tells us about tomorrow.

That article (from 2016–2017) suggested that 99% of underwriters’ activities would be replaced by advances in robotics and automation over the following decade.

Para empezar mi primera reflexión es ¿por qué sería necesario justificar el valor y existencia del suscriptor como contraargumentación de que aquel pronostico resultó erróneo?

Because it is necessary to assume and acknowledge that the industry is facing a new challenge arising from unprecedented exponential digital growth and a new generation of “competitors” (people and algorithms; yes, algorithms).

The presence of these new technologies in the daily lives of people and businesses has reshaped our societies. Terms such as artificial intelligence, machine learning, the Internet of Things, autonomous vehicles, blockchain, etc., are increasingly common in people’s lives and in the dynamics and evolution of companies.

In the last 10 years, artificial intelligence has experienced exponential development; the 4.0 revolution is among us, and as a generation we are called upon to face the challenge that lies before our eyes.

In particular, and with respect to our industry, it is important to note that it has developed over more than 600 years within our societies. As a historical background, it is worth noting that the first insurance contract was maritime and dates back to the year 1347; at that time, insurance as a business was widely распространed in medieval Italian cities, and later emerged as an “industry” in 1688 with the first contract underwritten at Lloyd’s Coffee House in London. Thus, the insurance contract was born as an operation that allows the transfer to a business organization of the harmful consequences arising from the hypothesis of the realization of a risk that materializes in losses.

Now, the use of the law of large numbers and statistics has always been a fundamental pillar for insurance companies, since predicting the frequency, number, and severity of losses that will occur in relation to the insured risks can only be estimated through probability calculations. Likewise, this can be achieved through the use of data and statistics.

While it is true that data science has always accompanied the insurance industry, it is no less true that Big Data and new technologies have radically transformed the role of data in the insurance business.

It is enough to think about what we do on an ordinary day, from the moment we wake up until we go to bed. Throughout the day, we use our phones and their various apps, GPS systems, telematic devices, and/or wearables. Each of these applications becomes a source of data that leaves a digital footprint of our preferences, consumption habits, and behavior patterns over the twenty-four hours of the day. This digital footprint, accumulated over weeks, months, and years, crudely records our most intimate identity, forming our digital profile. With some nuances, this same description could be applied to commercial and business activities.

Thus, through this data management, Big Data allows those who need it to obtain precise, relevant, and connected information to identify business opportunities, carry out more efficient operations, reduce costs, or offer new and better products and services. This relevant information is what artificial intelligence–based systems use, leveraging this vast amount of data obtained in real time to continuously improve their processes.
Así es que el Big Data a través de esta gestión de datos permite a quien lo necesite obtener información precisa, adecuada y relacionada para identificar oportunidades de negocio, realizar operaciones más eficientes, reducir costos u ofrecer nuevos y mejores productos y servicios. Esta información relevante es la que utilizan los sistemas basados en inteligencia artificial que aprovechan esta enorme cantidad de datos obtenidos en tiempo real para mejorar permanentemente sus procesos.

It is a fact that the use of new technologies has arrived in the insurance industry to stay. In this context, today we can identify two sources of data in the field of insurance:

1. a) Traditional data sources
2. b) Data sources derived from new technologies

For the first, we have data on the insured risk obtained through surveys, forms, or personal interviews, including historical information, personal data, information about the insured asset, location of the insured risk, and so on.

Regarding data sources derived from the use of new technologies, these are the data extracted from telematic devices, the insured’s online activity in general, satellites, and the insured’s behavior displayed through the digital channels provided by the insurer (claims history, payment history, contracting/renewal history, etc.).

I want to focus on the intersection between the latter and artificial intelligence, and the role and challenge this implies for the underwriter. In other words, I aim to momentarily set aside the theoretical approach and pilot tests in the use of artificial intelligence (which we have read about extensively and in varied contexts) and instead focus on how to make its use possible for the benefit of the underwriter, directly from their own desk.

At the end of 2022, OpenAI launched ChatGPT, a chat interface for its Large Language Models (LLM) for the general public. Many refer to this as a turning point for artificial intelligence technology.

Direct interaction with ChatGPT amazed users when they realized its impressive artificial intelligence (AI) capabilities, sparking widespread debates about use cases, ethics, and how it would revolutionize industries across the board. Although ChatGPT has achieved immense public success, it is only a temporary milestone in the rapid development of increasingly advanced AI model versions.

Language models, such as ChatGPT, are approaching a level of sophistication where they could become a “killer technology” for the insurance industry. This refers to a radical innovation in a product or process that will quickly render the utility of the techniques or processes that preceded it obsolete.

Most activities of insurance companies revolve around working with unstructured text:

• Processing of applications and claims
• Customer support
• Document evaluation for underwriting and risk analysis

This is precisely the reason why it is impossible to compete with language models like ChatGPT in the long term. They can quickly read, process, and analyze large amounts of text (whether written or verbally synthesized) and convert it into simple human language such as chat messages, emails, summaries, or reports.

While many insurance companies have been busy developing their own artificial intelligence capabilities, the cost of cutting-edge language models is decreasing exponentially. At the same time, their performance is skyrocketing. Models are no longer confined to research and development but are being leveraged in real commercial applications.

Despite the hype and its impressive capabilities, it is important to have reasonable expectations regarding the current reliability of ChatGPT. It should be considered as an extremely sophisticated text predictor rather than a superintelligence with deep understanding of the world or complex problem-solving abilities. At least for now, LLMs like ChatGPT cannot extract and comprehend the meaning of words as humans do. Instead, they work with statistical correlations they have learned between words, sentences, and grammatical structures.

This makes it more successful in creative tasks where there are multiple correct ways to respond, such as generating text or summarizing a document. However, it remains prone to errors when it comes to concrete facts. While it can still be perfectly suitable for certain “internal” tasks or basic customer-oriented tasks, it should not be left unsupervised in high-risk business use cases.

Currently, one of the main limitations of ChatGPT is its tendency to confidently provide information that seems plausible but is objectively incorrect, which can mislead its interlocutor due to its generalist training and limited understanding of the world. However, with proper and domain-specific fine-tuning, it already offers potential for insurance underwriters’ tasks in various use cases.

What follows aims to “ground” some examples of real and concrete applications of AI in the management and tasks of the underwriter.

• ChatGPT

By using it in a simple and creative way, underwriters can easily enhance their efficiency and productivity, saving time and resources that can be better devoted to other, more important tasks (https://chat.openai.com/chat).

Summarize documents to facilitate understanding

Documents such as policies, claims, or other legal texts are known for being long, dense, and generally difficult to analyze. ChatGPT can make this everyday task much faster by summarizing the given text into a format more useful for the underwriter’s purpose. Beyond the obvious summary (“summarize it in 10 bullet points”), ChatGPT can also produce more sophisticated outputs. For example, it can be asked to create and populate a table with relevant insurance categories based on the input text, making complex writing very easy to scan.

However, never share confidential information (such as client data or sensitive business information) directly through the public ChatGPT interface, as OpenAI may have access to the input text and could use it to train future AI models with the shared data.

Generate responses to customer inquiries

ChatGPT’s ability to analyze text is exceptional, but another key differentiator is its capacity to produce (semi) creative, human-like writing. The goal is never to rely on ChatGPT for objective information, but to use it to draft responses. ChatGPT can help create professional and prompt email responses to customers, using simple terms.

This is exceptionally useful for non-native English speakers, as ChatGPT can correct grammatical errors, provide stylistic assistance, and craft the response based on the desired style or length specified in the prompt. Specifying the tone of voice in the prompt can also help produce more refined results (“write the response in a more polite yet persuasive tone”).

Taking a step further in the use of GPT

• Simple integrations

While ChatGPT is a powerful tool, there are many more ways to leverage GPT’s advanced language processing technologies to enhance processes and tools within the insurance industry. For example, by simply integrating GPT with company systems and providing specific training to achieve a particular objective.

What “simple integrations” means:

• Using GPT for a specific task
• Integration of a GPT model into a single company system
• Train the model on a specific data source for a particular goal or task

• Claims analysis and triage

What makes GPT a useful tool in insurance is its ability to efficiently handle unstructured text. Fortunately, there is no shortage of this in the insurance industry. An insurance company can receive a large volume of claims daily. Without a system to prioritize these claims, it can be difficult for the company to manage the workload effectively and ensure that the most urgent claims are addressed first.

To address this problem, the company could use GPT to classify and prioritize claims as they are submitted. GPT can be used to analyze each claim, including the type of policy, the nature of the loss, and the potential impact on the policyholder. This analysis can then be used to assign a priority level to the submitted claim.

Using GPT to automate the claims classification process can help create a faster and more positive customer experience by reducing the time it takes to process claims and addressing, for example, the most serious ones first.

• GPT-powered chatbot

Chatbots are becoming an integral part of many companies’ websites. There are also several practical examples of GPT-powered chatbots being used in insurance, and doing it well can be a differentiating factor when it comes to customer service and engagement.

Implementing a chatbot assistant is becoming radically easier, with some companies offering GPT-enabled bots that can be used on a website (or WhatsApp and Facebook Messenger) with minimal effort. However, it is inevitable to provide a certain level of training on the company’s specific policies, insurance products, and general frequently asked questions.

• Complex integrations with GPT

As in the previous examples, fine-tuning and training the GPT language model for specific tasks can help enhance and automate a wide range of internal applications and customer-facing processes.

Arguably, the most critical and demanding process within insurance operations is underwriting: assessing risks, setting premiums based on those risks, and determining the terms and conditions to be established in the policy.

This involves processing and analyzing a variety of documents, including demographic data, background and historical records, financial records, and claims behavior, among—likely—many others, depending on the characteristics of the risk to be underwritten.

Conclusions

We are experiencing exponential technological growth that is changing the way we operate and run our businesses; therefore, we must adapt to these changes and understand the new exposure they bring to the underwriting process.

For their part, there are many things insurers can consider to help them become more efficient in the way they operate, keep up with technological changes, and, above all, respond to new customer behaviors and emerging risks.

Data analysis, artificial intelligence, blockchain, and chatbots are methods we can use to make underwriters more efficient. Consequently, the underwriter of the future will need to adapt to the new world we face, as service platforms will change, new distribution channels will emerge, existing roles will evolve, and new roles will arise.

Clearly, human judgment, sensitivity, and reasoning will never be replaced, so that alarming 99% mentioned at the beginning will never become a reality. However, it is clear that the industry and its players must change rapidly.

Finally, as the great Peter Drucker said, “The greatest danger in times of turbulence and change is not the turbulence or the change itself, but acting with yesterday’s logic.”

____________________________________________________________________________________________________________________

Sources consulted:

• The Future of Underwriting – IIS Mentorship Program
• ChatGPT what it holds for the insurance industry – Supercharge
• AI’s Promise & Challenges for insurers – Gradient AI / Origami Risk
• The Pivot role of AI and Insurance Underwriting
• Will AI revolutionize marine insurance? – DWF
• Nuevas tendencias tecnológicas para el sector asegurador – onLygal Corp.