ClaimGuard: Turning Insurance Data into Actionable Decisions
Inspiration
Insurance companies manage hundreds of claims across different teams, but it isn't always easy to identify which claims are taking too long or which teams are handling heavier workloads.
For Millennium Mutual, we wanted to build something that could help managers spot these problems early and make informed decisions.
Instead of creating a dashboard that only displays numbers, we wanted to build a tool that explains what the data means and recommends practical next steps.
That's how ClaimGuard started.
What It Does
ClaimGuard is an interactive insurance claims monitoring and decision-support application.
It helps insurance managers analyze claim performance, detect operational risks, prioritize unresolved claims, and explore possible workload improvements.
Our application includes four main capabilities:
1. Insurance Analytics Dashboard
Managers can view total claims, unresolved claims, overdue cases, and team workloads.
Interactive visualizations highlight workload differences and show the relationship between claim processing times and policy renewal outcomes.
2. Automated Review Queue
ClaimGuard uses rule-based detection to identify claims that have remained unresolved beyond a configurable threshold.
Claims older than 60 days are marked Critical and recommended for supervisor escalation. Other overdue claims are assigned High priority.
The system automatically generates a review queue that managers can filter, search, and export.
3. Smart Workload Rebalancing Simulator
Our simulator allows managers to explore how unresolved claim workloads might change if claims were transferred between teams.
For example, Team C currently holds 38 of the 48 unresolved claims in our dataset.
By simulating a transfer of 10 claims from Team C to Team A, managers can compare the workload distribution before and after the proposed change.
The simulator also identifies which unresolved claims would be considered for transfer, starting with those that have been waiting the longest.
It is a planning tool and does not modify actual claim assignments.
4. Smart Alert Center
ClaimGuard automatically generates alerts using three business rules:
- Overdue claim detection
- Overloaded adjuster team detection
- Senior-review verification for unresolved claims worth CAD $10,000 or more
Every alert includes a severity level, an explanation, and a recommended action.
Managers can filter alerts by type or severity and download the results.
How We Built It
We developed ClaimGuard using Python, Pandas, Streamlit, and Plotly.
We started by cleaning and validating the supplied insurance datasets. This included preparing dates, claim statuses, financial values, and related policy records.
Next, we used Pandas to analyze unresolved claims, team workloads, claim processing times, and policy renewal outcomes.
We implemented transparent Python business rules to detect overdue claims and excessive unresolved workloads.
We then built the automated review queue, the workload rebalancing simulator, and the Smart Alert Center.
Streamlit provides the interactive interface, while Plotly displays the visualizations.
We also created automated Python tests to validate the calculations, detection rules, simulation results, and data integrity.
Finally, we deployed the application using Streamlit Community Cloud and used GitHub for version control and collaboration.
What We Discovered
Our analysis identified several findings:
- The cleaned dataset contains 4,684 policy records and 847 unique claims.
- There are 48 unresolved claims.
- Team C holds 38 unresolved claims, representing approximately 79% of the unresolved workload.
- Fifteen claims have been unresolved for more than 30 days.
- Eight of these overdue claims have been unresolved for more than 60 days.
- Completed claims linked to policies that were not renewed took an average of 26.55 days to process, compared with 16.76 days for renewed policies.
The difference in processing times suggests an important relationship worth investigating, but it does not prove that slower processing caused non-renewal.
Using our default detection thresholds, ClaimGuard generated 27 business alerts, including eight Critical alerts.
Challenges We Faced
One challenge was turning raw insurance records into reliable information. We needed to handle inconsistent or missing data while making sure the calculations were accurate.
Another challenge was deciding how much automation was appropriate.
Insurance claims involve important financial and customer decisions, so we didn't want the application to automatically approve, deny, or reassign claims.
Instead, we focused on explainable detection rules and recommendations that support human review.
We also had to integrate backend analytics with an interactive frontend within the limited hackathon development time.
Accomplishments We're Proud Of
We're proud that ClaimGuard goes beyond a traditional reporting dashboard.
It not only identifies problems but also provides a prioritized review queue, explains why alerts were generated, and allows managers to explore possible workload changes.
We also built automated tests and deployed a working application that judges can explore directly in their browsers.
Our system follows the challenge's DATA → DETECT → DECIDE → ACT approach using transparent business rules.
What We Learned
This project showed us that useful automation doesn't always require a complex machine learning model.
Clear business rules, reliable data, and a practical user interface can provide meaningful decision support.
We also learned how important it is to communicate limitations and distinguish between identifying a risk and proving that a problem has occurred.
What's Next for ClaimGuard
With more development time, we would like to add:
- Real-time insurance system integration
- Email or in-app notifications
- Persistent claim-review tracking
- Role-based access for managers and adjusters
- Workload recommendations that account for team capacity and claim complexity
- More detailed financial and operational reporting
Our long-term goal would be to make ClaimGuard a practical tool that helps insurance teams identify delays earlier, prioritize their work, and make better decisions.
Try ClaimGuard
Live Application: https://claimguard-insurance.streamlit.app/
GitHub Repository: https://github.com/krishpatel9362-kp/ClaimGuard
ClaimGuard is a decision-support prototype. It does not automatically approve, deny, pay, or reassign insurance claims. Results are based on a historical dataset snapshot dated October 31, 2026.
