Designing a SaaS tool to improve task efficiency of Insurance Brokers and reduce overall TAT: Case Study

Let me take you through my journey of designing a web application for insurance brokers in the US market, with the goal of streamlining their daily tasks and enhancing productivity.

11 min readDec 23, 2023

What is Blue Horizon?

Blue Horizon is an AI-based Insurtech SaaS tool that assists commercial insurance brokers in the US markets by reducing their manual work of data entry, data organisation, and quote and policy comparison while also empowering them with useful market insights and trends.

The features in the tool enable agents to streamline their daily tasks with the help of AI Copilot’s (Broker Copilot) recommendations, allowing them to make more informed decisions. Broker Copilot is a Large Language Model(LLM) designed for use within the Blue Horizon.

The product is currently in the Pre-MVP stage and was designed by me during a brief six-week timeframe at Velocita, a creative brand consultancy.

Why does it exist?

Understanding context

Commercial Insurance helps protect businesses by providing them coverage against various risks and reducing the financial losses they might face. This includes customising policies for different aspects like property insurance, general liability, and commercial auto insurance.

Businesses hire insurance broker agencies who act as intermediaries between the business and the insurance carriers. Brokers assess the business’s needs, obtain quotes from different insurance carriers, and analyse quotes to gain industry insights.

Understanding the workflow

Data from the quotes is manually entered, compared, and then studied for market insights. Finally, all the findings are compiled into a document which is called a Recommendation Report but we will call it a submission document for simplification here.

The suggestions in the document need to be tailored to meet specific business requirements and in-depth industry trends and analysis. The process takes days or even a week for each client because it’s done manually.

The goal is to ensure that the clients receive a comprehensive coverage at competitive terms with adherence to their specific requirements.

This is what different insurance quote documents look like:

Manual processes increase error risks and handling non-standardized insurance quote data poses challenges. Meeting deadlines becomes difficult, slowing brokers down and hindering business scalability and client attraction. This detracts from their primary role of offering industry insights and recommendations.

Brokers face errors, delays, and diversion of time due to manual processes in synthesizing non-standardized insurance data, impacting deadlines, hindering business growth, and impeding the study of crucial industry insights.

An in-depth discussion was held with 5 subject matter experts, to better understand the issues and workflow of the brokers. Following were the insights drawn from the discussion:

Non-standardized quote and policy documents require more time for brokers to label and compare.
Coordinating manual tasks across different teams results in inconsistent workflows, creating challenges in collaboration, especially with varying timelines and multiple clients.
The reliance on multiple, non-integrated tools for daily tasks creates inefficiencies; as brokers have to constantly update and cross-reference information across platforms.
Difficulty staying updated with market trends hinders brokers from providing insights; impacting the formation of comprehensive business insights and improved coverages.
Brokers find it challenging to update insights across various sheets, hindering efficient data consumption and insight formulation.
Lack of a monitoring system makes it challenging to assess team performance, track account outcomes, earnings, and evaluate SLA success.

Blue Horizon aimed to integrate AI into the broker agency’s workflow, aiming to replace manual data entry and quote comparison with AI-driven data labeling and advanced NLP capabilities. This enables efficient data extraction from documents and automated comparisons.

Competitive Analysis

To explore the new product space, we conducted extensive competitive research on direct and indirect competitors, analyzing available products in the market. Some of the competitors were —

Groundspeed (Similar tool but for Insurance Underwriters)
Chisel.ai (Only for quote compare and policy check)
Draftable (Document comparison tool)
ChatGPT and Bard (LLM Chatbots)
To-Doist and Basecamp (for project management)

Given the limited number of direct competitors in the market for this product, We focused our research on indirect competitors to identify relevant patterns applicable to the application.

The insights from the research helped us better understand the needs of the users, and design a better solution. So our objective was —

Develop a user-friendly platform integrating NLP for data labeling, workflow facilitation, and AI-driven insights and recommendations. This enhances productivity in broker agencies by reducing manual work, improving team collaboration, and simplifying data analysis.

The goal was to enhance the following metrics -

Improve Task Efficiency: Measure the percentage reduction in time spent on manual data entry and document management, indicating improved task efficiency.
Reduce Turnaround Time: Track the decrease in the time taken to complete key processes, reflecting the product’s impact on enhancing workflow speed and responsiveness.
Error Reduction: Quantify the decrease in errors and inconsistencies in data processing, showcasing the product’s impact on improving data accuracy and reliability.

Let’s move to the solution part of the problem.

A. Restructuring the workflow

The implementation of AI in the workflow will not only enhance system efficiency but also intelligence, addressing the issues identified in the research. Mundane manual tasks will be replaced by AI. Here’s how the Blue Horizon AI will operate:

After multiple iterations, we developed a new workflow for brokers. It integrates user-friendly features to facilitate task completion while leveraging AI capabilities to optimize output.

Let’s break down the design by major task flow and key components.

Our design principle aimed for a clutter-free and intuitive design, intending to address the problems identified during the research.

A1. Task Card

The list format can’t accommodate all the necessary details with proper visual hierarchy. Information consumption in lists with more than 14 data points is not easy for both the brain and the eyes.

Hence, we chose to use task cards to organize all necessary information with a clear visual hierarchy, reducing cognitive load for users and facilitating easy information scanning. Additionally, we integrated an accordion feature to provide task details and related next actions.

The key factor in designing the task cards is Information Architecture(IA), like any other feature. Below is the break down of how I designed the IA to make sure that the data points are scannable:

I compiled a list of the essential data points and actions required on this card.
I categorized the data points into different groups based on their context.
I prioritized the data groups and further ranked the pointers within each data group.

4. After several iterations, I opted to display detailed process information through steppers and progress indicators within the accordion — ‘Details’.

A2. The Progress Detail accordion

The primary purpose of this accordion is to provide a detailed task progress overview, featuring a status label for quick reference on the task card.

As the copilot now operates in four steps, I mirrored this mental model in the details accordion. I incorporated steppers for all the steps with the addition of a loading bar in the first stepper as this task is system-driven, providing clear progress indication to the user.

The tasks include Initiating the Copilot process, Analyzing the Copilot Score, Human Review, and Submitting the Recommendation Report. Necessary actions and information were added in each step to aid the user in taking the next action.

I followed a consistent visual language across all the steppers in ‘Details’ accordion. You can explore the designs of these cards below.

B. Putting all together on the homepage

The header section of the homepage welcomes the user, showcasing buttons for creating a new task, checking notifications, and accessing Broker Copilot — the AI assistant.

Below this section, the content is divided into 3 major sections —Task List, Copilot Performance and Business Insights.

B1. Task Cards List

The task list occupies a significant portion of the homepage as it is a pivotal list for the users.

Sorting and Filtering System
Creating a robust sorting and filtering system was crucial to prevent user confusion when searching for relevant tasks.

The key filters included segregating the task list into quotes and policies, further categorized by a specific line of business and time frame of the task list.

An effective sorting system simplifies the task of finding relevant items. I incorporated essential sorting options with clear affordances, ensuring users can easily identify their selections. The options included sorting the Super Rush, Past Due, New, Archived tasks, and by Task Progress.

B2. Insights Overview

The left section of the homepage is designated for concise insights that users prefer upon landing on the page. These insights encompass two types:

Copilot Performance
As a recently developed language model specialized in commercial insurance, the model undergoes a learning process to accurately perform data labeling and document comparison.

Copilot continuously evolves with daily human inputs, generating a confidence score. The aim is to enhance accuracy and decrease turnaround time, ultimately optimizing efficiency.

Business Insights
Copilot navigates through vast volumes of unstructured data, comprehends trends, and leverages web access to deliver impactful industry and business insights.

Brief insights are dynamically displayed through a scrolling slider, emphasizing the latest information for users.

A deeper exploration into these insights and performance metrics are available in the Part 2 of the case study.

C. Broker Copilot — The AI Assistant

Broker agencies heavily rely on data analysis and understanding business and risks for specific industries, a task challenging for a single human to master entirely. Currently, each agency has some set of fixed industry experience and knowledge, hence the fixed areas of business.

However, large language models (LLMs) like Broker Copilot can transcend these limitations. These excel in comprehending large datasets, reducing the burden of manual work for brokers, enhancing workforce efficiency making them invaluable decision-making tools.

Come aboard for it’s design journey with me:

C1. Home Screen

When brokers click on the Broker Copilot button on the homepage, they are directed to a personalized LLM designed specifically for professionals in the commercial insurance sector.

This area consists of three sections: a history of previous chats with Copilot, recently completed Copilot tasks, and new business insights generated by Copilot, offering users engagement with the latest insights.

The design adopts a modern and futuristic aesthetic, drawing inspiration from user-friendly tools like ChatGPT and Bard. This familiarity ensures users face no learning curve when utilizing the feature.

C2. Chat Interface

After the user clicks the ‘Broker Copilot’ button on the task card, the copilot sets the chat window knowing the context of the task. If the copilot process is complete, the copilot generates the copilot score of the recent analysis with alerts for human review if any.

The chat screen includes prompts with an input field at the bottom and a button for file uploads.
It combines specific actions and prompt-based interactions to delve deeper into account information.
Brokers have the option to resolve issues one-by-one or together in the tabular comparison.
After resolving the data labeling issues, users can access a comprehensive overview and smart recommendations by clicking the “Copilot Summary” action button.

After each response, users are presented with the next set of actions or prompts for thorough exploration.
The next major task involves compiling insights and learnings into the submission document.
Users have the option to add selected text and charts to the submission doc editor, similar to Kindle.
Text added to the submission document trigger the opening of a side overlay. Users can view the data being pasted in the document editor within this overlay.

Prototype of text being added to the document editor

The document editor can be opened side-to-side, up and down, or in dual monitor configurations on the screen. The document editor offers all the basic features of a rich text editor for editing and formatting

Text editor image

C3. Comparison Table

After Copilot labels the data, it compiles data points from all comparison documents into a table. This table serves as a tool for agents to compare against fields, with distinct differences highlighted.
Users can easily navigate to different areas in the comparison using specific filters.

The fields are thoughtfully segregated into sections, enhancing information consumption. Each section can be collapsed, reducing the need for excessive scrolling.

Moreover, the columns are designed to be draggable, providing a convenient way to compare them with the first column i.e. the Quote Ask document.

Let’s see it in action to see how this works-

Prototype of the comparison screen

C4. Source Document

Reviewing source documents with different file extensions in emails and other communication platforms can be cumbersome.

To ease this, a feature was introduced that allows users to view source documents within the application. Users can examine the original documents and make comparisons by selecting names from the top bar.

Similar to the comparison screen, fields were included here. When a user selects a field, the corresponding area in the document gets highlighted. If multiple documents are compared, the text is highlighted in all documents based on the selected field on the left.

Additionally, users can navigate through the document using standard zoom in and out functionalities. Here if the specific fields are selected, they can be added as a highlighted snippet in the submission document.

Strategic Impact

In Retrospect

Learnings and challenges