Clients behavior analysis for car insurance aggregator
A US car insurance aggregator needed a way to determine what content to show each visitor during the quote generation process. We developed an AI-powered prediction module that analyzes user intent in real time and dynamically adapts page layouts to maximize revenue from insurance commissions and advertising.
Industry
InsurTech
Client
Under NDA
Region
USA
Main challenges
Business Challenge: No way to determine the right user experience
The client could not determine which users were likely to purchase insurance and which primarily interacted with advertisements.
- Missed advertising revenue opportunities
- Lost insurance conversion potential
- Same experience shown to all users
Technical Challenge: Predicting behavior during quote generation
Predictions had to run inside the existing quote flow without increasing response times or affecting user experience.
- Inference required under ten seconds
- Revenue and intent predicted simultaneously
- Large dataset required extensive labeling
Delivery Challenge: Deploying AI into a live platform
The solution had to integrate into production quickly while maintaining stability and proving measurable business impact.
- Delivered within a 2.5-month timeline
- Integration could not disrupt live traffic
- Revenue uplift required validation testing
What we did
We built an AI/ML module that uses historical user data to predict how each visitor will behave on the platform. The system runs automatically for every user requesting a car insurance quote and determines whether to display insurance quotes, ads, or both. There was no prior solution in place - we developed the entire prediction capability from scratch.
We tested multiple machine learning models and selected an ensemble that combines purchase-intent classification with revenue forecasting. The module was deployed as a dedicated microservice and integrated into the existing platform. Automated training and retraining pipelines on AWS SageMaker keep the model accurate as new data accumulates.
Revenue growth validated through A/B testing
Records prepared and used for model training
Response time maintained in live quote generation
Revenue streams optimized by a single AI engine
Real-time prediction engine
An ML-powered prediction service evaluates customer behavior during the quote generation process.
- Predicts purchase intent in real time
- Runs automatically for every visitor
- Operates within the quote generation flow
- Delivers predictions under latency limits
Revenue optimization layer
A dedicated optimization engine forecasts expected visitor value and determines the most profitable experience.
- Predicts expected revenue per visitor
- Optimizes insurance commission revenue
- Optimizes advertising monetization revenue
- Supports automated layout decisions
Dynamic layout personalization
The platform adapts content presentation based on real-time prediction outputs.
- Displays quotes based on buying intent
- Displays ads for browsing-oriented users
- Combines ads and quotes dynamically
- Eliminates static page experiences
Automated training pipeline
Automated ML workflows maintain model quality as customer behavior evolves.
- Trains models on historical datasets
- Retrains using newly collected data
- Maintains prediction quality over time
- Supports continuous model improvement
A/B testing framework
A dedicated testing framework was introduced to measure business impact and validate model effectiveness.
- Compares AI-driven and control groups
- Measures revenue uplift automatically
- Validates model effectiveness in production
- Provides evidence for business adoption
Key results and business value
One-size-fits-all layout layers replaced by personalized user experiences
No visibility into customer intent turned into real-time behavioral predictions
Revenue loss across channels replaced by automated optimization
Unproven business impact proven by +10% revenue growth
Features Delivered
Key capabilities of the AI-powered prediction platform
- Real-time customer behavior prediction
- Dynamic page layout personalization
- Revenue optimization across two channels
- Automated model training and retraining
- A/B testing and impact measurement
AWS
Python
XGBoost
Scikit-learn
MongoDB
Docker
Technical Highlights
Weighted ensemble modeling
Unified classification and regression models into a single scoring system for reliable decisions
Low-latency inference service
Built a dedicated ML microservice capable of delivering predictions within quote generation constraints
Revenue optimization logic
Predicts purchase probability and value to maximize advertising and commission revenue
Production-safe data access
Implemented isolated database views that protect production systems from analytical workloads
Continuous retraining pipeline
Automated retraining workflows keep prediction models aligned with changing customer behavior
Cloud-native AWS architecture
SageMaker, ECS, Lambda, S3, and CloudWatch provide scalable infrastructure for ML operations
Client Feedback
"Before this project, we were essentially making the same decision for everyone. Now the platform does a much better job of adapting on its own, and that's something you notice over time rather than overnight. Teams spend less time debating what should be shown to users and more time focusing on growth opportunities. It feels like we've added another layer of intelligence to the business without making the product more complicated"
Chief Technology Officer
Client
Unlock new growth opportunities
Discover how custom software with AI-powered features can help your business move faster, improve workflows, and create new competitive advantages.
