- AI & Data Customer Intelligence Powered By AI
We built MOOST — an AI-powered customer profiling engine
that segments, classifies, and predicts customer behavior
using machine learning, deployed for a leading Tunisian
insurance company.
59K
Customers Profiled
115K
Contracts Analyzed
103K
Claims Processed
69,7%
Model Precision
04
Customer Segments
- OVERVIEWIntelligent Customer Profiling At Scale
MOOST is our proprietary AI product that segments customers based on consumption data and transaction history. Using advanced clustering and classification algorithms, it identifies, classifies, and predicts customer actions — enabling data-driven decisions for marketing, risk assessment, and retention strategies.
- FEATURESWhat Moost Delivers
A plug-and-play AI engine that connects to your data sources and delivers actionable customer intelligence — scalable to integrate with any ERP or CRM.
CUSTOMER SEGMENTATION
K-Means and hierarchical clustering to group customers by behavior, with dynamic real-time segment updates.
RECOMMANDATIONS
Collaborative and content-based filtering to suggest relevant products and actions per customer context.
DATA VISUALIZATION
Interactive dashboards and custom reports summarizing segments, risk assessments, and insights.
WORKFLOW AUTOMATION
Apache Airflow DAGs for ETL orchestration, model training, and continuous retraining with MLflow tracking.
THE PROBLEM
Siloed Data, Blind Decisions
The insurance company managed over 59,000 customers with 115,000+ contracts and 103,000+ claims spanning 11 years of data. Customer information was scattered across multiple databases with no unified view. Claims decisions were made without understanding customer behavior patterns, leading to reactive rather than proactive risk management. There was no way to identify high-value customers at
risk of leaving, or to detect early signals of fraudulent claims.
THE SOLUTION
Customer intelligence
We deployed MOOST to extract data from SQL Server, clean and engineer features including financial ratios, then apply K-Means clustering to discover natural customer segments. A Support Vector Machine classifier was trained to automatically categorize new customers into segments. The entire pipeline — from data extraction to prediction — was automated using Apache Airflow, with MLflow tracking model performance and MinIO storing results. The output: four clear
customer segments with actionable insights for targeted decision-making.
- The processFrom Raw Data To Predictions
Data Collection & Preprocessing
Extracted data from SQL Server across people, contracts, vehicles, and claims tables. Cleaned anomalies, engineered features, and calculated financial ratios including cumulative S/P ratio.
AI Model Development
Applied K-Means clustering to discover natural customer segments, then trained an SVM classifier to automatically categorize new customers into one of the identified segments.
PROCESS AUTOMATION
Built Apache Airflow DAGs to automate the full pipeline. MLflow tracks model experiments and performance. MinIO stores all data artifacts and model outputs.
DEPLOYMENT MONITORING
Deployed the AI solution on the client's internal server. Continuous model retraining ensures segments stay current as new customer data flows in.
- DATA SCALE11 Years Of Customer Intelligence
Data spanning 2013–2024, cleaned and centralized into a unified dataset for model training.
59,543
PEOPLE
4,735
LEGAL ENTITIES
115,316
CONTRACTS
118,621
VEHICLES
MOOST transformed how we understand our
customers. For the first time, we can see clear
segments based on real behavior data, and the
automated pipeline means our insights stay fresh
without any manual effort. The accuracy of the
classification model exceeded our expectations.