Help Claims Fraud Investigators better identify and quantify claim fraud with the power of AI and ML.
Company
Industry
Insurance
App Type
Enterprise Application
Date
2022 - present
Role
Lead Designer
Activities
Our Target Buyers are the Claims Organizations in Insurance Carriers. These organizations are typically trying to solve the following problems by identifying fraudulent claims and preventing fraudulent payouts.
Fraud Increases Cost per Claim
Money paid out to claims that are fraudulent increases overall costs due to payouts that aren't accurate and the time spent to payout someone you shouldn't.
Fraud Increases Claim Duration
Fraudulent claims, and investigation fraudulent claims, takes time for the team to investigate, thus taking away man-hours that could be spent helping real customers.
Investigating Fraud Causes Poor NPS
Investigating a claim for fraud when it doesn’t exist adds more time between them reporting the claim and sending a payment, creating a terrible customer experience.
I synthesized roles and responsibilities found in Claims Organizations into three archetypes.
When speaking with customers and prospects, the following items surfaced as consistent challenges across all roles and organizations, regardless of scope and scale.
Fraud is Hard to Spot
Understanding where the risks are in a claim relies on human experience and is not aided by many tools today. Additionally, there is a lack of visibility to understand changes in a claim over time.
Redundant & Repetitive Tasks
Many tasks at the start and end of investigation are the same amongst all investigations and involve running the same reports, uploading the same reports for each respective case, and updating internal and external systems when the investigation is concluded.
Systems Aren’t Well Connected
Getting data in and out of applications is an incredibly tedious process today. It usually involves some form of manual download and upload between local files and various online web portals.
The application has several tools to assist users in their investigation and assessment process. This includes things like AI-generated risks and recommendations, an AI chatbot which you can ask questions about the case (similar to ChatGPT), a machine learning model that will rank the risk percentage of fraud based on previous fraud found, and the ability reinforce the model and dataset based on user feedback.
All of these features help reduce the cognitive load for the user and help them make better decisions, faster.
Several configurable automations are built into the application. These include three types of automation categories: automatically assigning claims to investigators, automatically ingesting reports from tools outside of the application, and automatically updating internal and external systems when the investigation is closed.
All of these features help reduce menial admin tasks and help users focus on more challenging, impactful work.
The application can be setup as a standalone product or configured to work in a variety of ecosystems. We can configure integrations with external systems through APIs, as well as connect data points between objects in the application itself. Additionally, all of the data from internal systems can be brought into the application so users no longer need to have multiple tabs open to get the full picture.
All of these features help keep data in sync and up to date.
A potential fraud case gets referred to the investigation team and needs to be accepted by the Gatekeeper before an Investigator can begin working on it.
The Investigator conducts the actual investigation to determine the prevalence and severity of fraud. Phone calls, public reports, photographs, and other pieces of evidence are gathered and annotated before a final report is written and submitted to internal and external databases.
Asynchronously, the team members and managers can track cases, ensuring investigations do not go stale or remain in one state for too long.
I defined key fundamentals of the design system in the form of semantic tokens.
These fundamentals were applied to every component and supported custom overrides.
I made sure to use variables on all components so that themes could be changed on the fly, including dark mode and custom themes for enterprise clients.
