Overview

Grantline Insurance Co. is a US-based home insurance company that handles thousands of property claims every month. The problem was its claims system, which was old and built for manual review, so every claim went through the same queue regardless of size, whether it was a small roof leak or a total-loss fire. Because of that, investigators spent just as much time on low-risk claims as on the few that really needed a closer look, and some fraudulent or inflated claims still got through. The warning signs were there, things like odd claim histories, questionable repair estimates, and suspicious timing around policy changes- but they sat in separate systems, so no adjuster could realistically check them all on every file. Grantline needed a way to catch high-risk claims early without slowing down its process or creating friction for honest policyholders. That’s where Bacancy Technology came in. We built an AI-based claim scoring engine that reads claims data in real time and scores each claim for fraud risk using a mix of machine learning and rule-based checks. High-risk claims now go straight to the investigation team, while low-risk claims move through as usual.

Technologies Used

Python
Scikit-learn
FastAPI
Apache Kafka
PostgreSQL

Project Highlights

checkmark

Real-time fraud scoring built directly into the existing claims workflow

checkmark

Hybrid engine combining machine learning with business-defined rules

checkmark

Explainable scores that give investigators and compliance a clear reasoning trail

checkmark

Seamless integration across legacy claims, policy, and repair-estimate systems

The Challenges

1

Fraud signals were scattered across claims history, policy records, and repair estimate systems with no unified view for scoring.

2

The existing claims workflow treated every claim identically, so investigators had no reliable way to prioritize high-risk cases.

3

A black-box scoring model risked regulatory pushback and investigator distrust without clear, explainable reasoning behind each score.

4

The scoring system had to run in real time on incoming claims without slowing down the existing claims intake process.

Solutions by Bacancy

1

As a provider of insurance IT services, we built a data pipeline that continuously pulls and normalizes data from the claims management system, the policy administration platform, and the repair estimate vendor portal into a single feature store. It computes fields like prior claim count, time-since-policy-change, coverage limit history, and estimate-versus-market-rate variance once and keeps them current, so the model always scores a claim against a complete, up-to-date picture instead of a partial snapshot from one source.

2

Our machine learning experts trained a gradient-boosted classifier (XGBoost) on Grantline’s historical claims data, including past confirmed-fraud cases, so it could learn the patterns that separate fraudulent claims from legitimate ones. We combined the model with a configurable rules layer for the red-flag conditions the business wanted enforced directly, such as claims filed soon after a coverage increase or estimates well above regional benchmarks. Both layers merge into a single risk score, catching the patterns the model learns statistically and the conditions the business wants flagged every time.

3

We added an explainability module on top of the model using feature-attribution techniques, so every risk score comes with a clear breakdown of which factors drove it and by how much. Instead of just a number, investigators can see, for example, that a claim scored high because of a recent policy change combined with an estimated outlier, with each factor weighted and visible. This gives investigators a solid starting point for their review and gives compliance teams a clear audit trail they can pull up if anyone ever questions a scoring decision.

4

As a full-stack AI development company, we deployed the scoring engine as an event-driven microservice connected to claims intake through Kafka, so the system scores a claim the moment someone creates it instead of waiting for a batch cycle. The scoring service runs alongside intake, so it never holds up the acknowledgment the policyholder sees. As soon as it computes a score and its risk factors, high-risk claims move straight into the investigation queue with priority flags, while low-risk claims continue through standard processing untouched.

Core Features

checkmark

Unified Fraud Risk Feature Store

checkmark

Hybrid ML and Rules Scoring Model

checkmark

Explainable Score Breakdown

checkmark

Real-Time Investigator Queue Routing

No. of Resource

05

No. of Resource

Time Frame

February 2026 - July 2026

Time Frame

Outcomes

30% reduction in fraud-related claim losses

100% of incoming claims scored in real time

2x faster identification of high-risk claims

0 added latency to claims intake processing

100% of flagged scores accompanied by explainable reasoning

24/7 automated risk scoring coverage

Technical Stack

AI / ML: Scikit-learn, XGBoost, SHAP
Backend: Python, FastAPI
Database: PostgreSQL
Streaming: Apache Kafka
Cloud Infrastructure: AWS (ECS, CloudWatch)
API Communication: REST APIs

Experience With Bacancy

Optimize Your E-Commerce Workflow for Maximum Efficiency

iconWatch the video

Fintech Platform Scalability & Growth Optimization

iconWatch the video

Legacy Code Upgrades for Modern User Experience and Enhanced Data Analytics

iconWatch the video

Empowering Retail Marketing With Seamless User Experiences

iconWatch the video

Migrating To Laravel From Core PHP With Twilio Integration

iconWatch the video

Upgrading PHP Laravel version while scaling the Admin Dashboard

iconWatch the video

Optimizing CRM Performance With Data Management and Salesforce Integration

iconWatch the video

AI Solution Engine for Improved Performance and Error Handling

iconWatch the video

Enhancing Managed IT Services With Streamline Automation Process

iconWatch the video

SEO and Bulk Data Upload Solutions For An E-Commerce Healthcare

iconWatch the video

Elevate Your E-Commerce with Scalable UI and Enhanced Calculators

iconWatch the video

Simplifying Annuity Rate Management for Complex Calculations

iconWatch the video

Addressing Null Safety, Code Structure Issues, and Manual Deployment Issues

iconWatch the video

Optimize Task Management With Advanced Tracking

iconWatch the video

Revolutionize Applicant Tracking: Advanced Image Optimization Techniques

iconWatch the video

Cost-Effective Media Management Solution and Resolving Data Upload Issues

iconWatch the video

Enhance Your Management Service with User-Centric Innovations

iconWatch the video

2500+ Projects Experienced Innovation with Bacancy!

Get access to an experienced team of developers and engineers from Bacancy, handpicked to ace your goals. Kickstart within 48 hours, no-risk trial.

Book a 30 min call

14+

Years of Business Experience

1458+

Happy Customers

12+

Countries with Happy Customers

1050+

Agile Enabled Employees