The Role of Prompt Engineering in Pricing and Underwriting in Insurance and Banking

Introduction 

Generative artificial intelligence (GenAI) is revolutionizing the world of financial services. Its rapid evolution is enabling insurers and lenders to boost business efficiency and accuracy, deeply personalize their interactions with customers and prospects, and allow their teams to shift focus from mundane tasks to more strategic and impactful pursuits. 

The tools supporting GenAI are also evolving rapidly. For example, copilots and the large language models (LLMs) underlying them hold the promise of bringing GenAI capabilities to a broader swath of financial services teams, opening new GenAI avenues to technical and non-technical users alike.  

These GenAI-powered virtual agents can work alongside human experts to accelerate the use and usefulness of GenAI in solving real-world problems in a fraction of the time. Copilots can enhance decision-making to drive improved business growth and multiply the efforts of teams involved in such fundamental tasks as rating, pricing, underwriting, product formulation, and customer acquisition. 

Of course, working effectively with copilots also requires users to become skilled in prompt engineering. 

What Is Prompt Engineering?   

“Judge a man by his questions, rather than his answers.” 

-Voltaire  

Although Voltaire lived over 250 years ago, he understood the importance of asking the right questions if one hopes to get the right answers. This is where prompt engineering enters the GenAI picture. 

Prompt engineering is the practice of designing, composing, and entering prompts into AI models, particularly large language models (LLMs) to maximize the models’ effectiveness. The goal is to enter prompts that will produce outputs that are specific, accurate, and useful for the business effort at hand. Prompt engineering involves structuring questions, instructions, and/or specifying context in a way that helps the AI model correctly interpret the user's intent and deliver optimal results. 

Key Elements of Prompt Engineering 

There are several central tenets to good prompt engineering: 

• Clarity: Clearly specify what you want from the model 

• Context: Provide relevant background information when needed 

• Instructions: Use explicit instructions, such as "list," "explain," or "compare" 

• Constraints: Define limits under which the answer must operate 

• Iteration: Test and refine prompts for better responses 

Practitioners of prompt engineering strive to keep these goals in mind when crafting the questions they feed into their copilots. The reward is maximizing the effectiveness of the LLMs powering the copilots and discovering the shortest path between Point A (the questions input) and Point B (the answers/business directions output). 

For example, a very simple way to craft a prompt about climate change risks in property and casualty (P&C) insurance might be: 

"Tell me about climate change and its effect on insurance liability." 

While having the virtue of simplicity, it risks the possibility that the response may be too general, less focused, and less useful for what the user was hoping to learn. Using prompt engineering principles, a well-engineered prompt might look more like this: 

"Explain the main consequences of climate change and its effects on expected losses from hurricanes in the Gulf States of the US in less than 250 words." 

This prompt is more likely to yield a concise and informative response. (Thank you, Voltaire.) 

Prompt Engineering – Key Benefits 

Prompt engineering is key to getting the right answers to users’ questions. The overall benefit, of course, is greater business effectiveness and refocusing resources to focus on more strategic initiatives. Here are some of the contributing factors: 

• Greater Developer Control Over Model Building 

  Prompt engineering allows more control over how users interact with the AI. Good prompts provide context to the LLMs and ensure that outputs and presentation to users are concise and consistent. 

• A Better User Experience 

  One of the biggest factors in inefficient model development is the trial-and-error process involved. With good prompt engineering the time to model development is shortened, the outcomes are more consistent, and benefit more rapidly realized. 

• Enhanced Teamwork and Enterprise-Wide Leverage 

  Prompt engineering can result in prompts that aren’t tied to a specific system, and which can then be shared across the organization, reducing rework, and leveraging effort more effectively across the enterprise. 

• Bringing the Most Appropriate Resources to Bear 

  Prompt engineering is democratizing access to functionality for less technical users, so that they can apply their domain knowledge in areas such as rating, pricing, and underwriting without having to learn the ins and outs of the sophisticated technology they need to excel at their jobs. 

  At the same time, technical users can focus on driving ever more effective pricing and underwriting strategies without getting bogged down in the minutiae often required to transfer knowledge to those domain experts. 

Use Cases in Insurance and Lending 

The use cases for copilots and prompt engineering in insurance and banking are essentially limitless. While they are useful for such business processes as customer support automation, loan application support, and claims processing assistance, we will focus here on their application to pricing, rating, and underwriting.  

Some of these use cases are covered in our eBooks for insurance and banking on copilots, which provides a richer and more detailed summary of their benefits and several examples of their use.  

Here are just a few of the possibilities when combining copilots and prompt engineering:   

• Use Case: Leverage internal documentation to aid new users’ training 

  Example Prompt: 

  “How do I run a pricing simulation for homeowner’s insurance?” 

• Use Case: Allow new hires to navigate internal data more effectively

  Example Prompt:

  “What are the possible data types for demographic variables inside my system?” 

• Use Case: Use a copilot to generate Python code without requiring extensive coding knowledge 

  Example Prompt:

  “Generate code to create a GLM demand model for Illinois using vehicle age, repair history, and vehicle model type as predictors” 

• Use Case: Allow inexperienced staff to generate code more readily for repetitive tasks 

  Example Prompt: 

  “Generate code to import data into our models every hour and batch-schedule other tasks on the half-hour” 

Conclusion   

Generative AI enhances business growth and profitability in insurance and banking by helping to drive new business, streamlining business processes and reducing costs, and aiding solid decision-making through more rapid and thorough risk assessment and data analysis.  

These effects are further enhanced by adding copilots and prompt engineering to the toolkits of forward-looking teams in these financial services segments. In a rapidly evolving and highly competitive world, you owe it to yourself and your firm to learn and leverage these tools and skill up in your workforce in prompt engineering now. 

Learn how the Earnix Copilot can transform your business