Downloaded 11 times
Uploaded byDatabricks
Wizard Driven AI Anomaly Detection with Databricks in Azure
This document discusses the implementation of AI-driven anomaly detection using Databricks in Azure, focusing on fraud prevention and the role of citizen data scientists. It highlights the need for effective anomaly management through a holistic view and various detection methods, emphasizing the business benefits and successful case studies. The solution employs a no-code, serverless architecture that integrates into existing systems to optimize fraud detection and recovery processes.
More Related Content
Similar to Wizard Driven AI Anomaly Detection with Databricks in Azure
Wizard Driven AI Anomaly Detection with Databricks in Azure
- 1. Wizard Driven AI AnomalyDetection with Databricks in Azure Naomi Kaduwela Head of Kavi Labs Rajesh Inbasekaran CTO
- 2. Agenda Naomi ▪ Fraud PreventionOpportunity ▪ Why AI Audits ▪ Rise of Citizen Data Scientists ▪ Solution Approach ▪ Designing for Citizen Data Scientists ▪ Anomaly Lifecycle ▪ Deployment Options ▪ Success Stories Rajesh ▪ How the Solution Works ▪ Cloud Native, Serverless Architecture ▪ Databricks Integration
- 3. Billions of Dollarsof Opportunity $350 B Fraudulent Healthcare spending* * According to the National Health Care Anti-Fraud Association $25 B Spent annually by US Banks on anti-money laundering compliance* * According to Forbes $40 B Total annual cost of Insurance Fraud (excluding health insurance)* * According to the FBI
- 4. Ideal for AI! Why AI Audits DataVolume & Complex Patterns Need to Adapt to New Changes High Frequency Transactions Transaction Flagging Actor-to-Actor Flagging AI flags the root cause of Anomalies in a Scalable way!
- 5. Rise of theCitizen Data Scientist Thanks to technology abstraction Data Scientists can now focus on solving the business problem Accelerating time to value & maximizing their human potential!
- 6. Solution Approach 1. DifferentAnomaly Signatures (possible fraud) exist within same data 3. Despite different methods, a holistic view of anomaly is required for business 2. Different methods are efficient in detecting different Anomaly Signatures 4. Management of entire anomaly lifecycle management is critical for effectiveness and efficiency
- 7. Designing for CitizenData Scientists Business Benefits 04 ● Holistic and meaningful view ● Aggregate model into quantifiable business opportunity Evaluation & Visualizations 03 ● Collect and report model metrics ● In built visualizations aid understanding Portfolio of Algorithms 02 ● Diverse portfolio of algorithms available ● Ability to compare parameters across methods & combine multiple AI methods together Wizard Driven, No code ML 01 ● No programming required ● Enable Citizen Data Scientists Anomaly Lifecycle Management 05 ● Track from detection to actual recovery ● Human in the Loop for continuous improvement
- 8. Wizard Driven, No-CodeML Screenshot
- 9. Portfolio of Algorithms •Unsupervised • Supervised Distribution Clustering Association Sequencing Historical Occurrence Random Forest Neural Network
- 10.
- 11. Business Benefits 7,761,096 $768,408,624929,412 $21,263,307 26,452 $ 4,723,995 36,573 $ 5,263,785 119,079 $ 20,536,362 295,041 $ 262,482 21,099 $ 3,760,542 123,243 $ 4,062,246 308,025 $ 3,384,471 2019 Business Benefits Summary Anomaly Opportunity Breakdown By Method Billing Error Duplicate Repair Labor Overcharging Material Overcharging Over Repair Wrong Shop Wrong Repair Opportunity Records Savings
- 12. From Statistical Anomalyto Confirmed Fraud Raw Data Predicted Anomaly Possible Fraud Confirmed Fraud Actual Recovery
- 13. Human in theLoop Anomaly Lifecycle Model Building Update Feedback Anomaly Detection Recovery Process Anomaly Validation Citizen Data Scientist Business SME
- 14. Deployment Options Estimate Option 1:Prevention Real Time Scoring at Time of Estimate to Prevent Fraud Money is Exchanged Payment Option 2: Reclaim Batch Processing Post Invoicing to Reclaim Fraud Invoice
- 15. Enterprise Tech StackIntegration Digital Solutions Layer KPIs and Metrics, Descriptive Dashboards. AI Audits Data Services Layer Integration, Transformation, Governance, Security, Orchestration, Data Catalog Source Systems & Infrastructure Ingestion of Internal Systems, Industry Systems, Customer Systems. Storage & Compute
- 16. Success Stories ▪ $6.8Mof potential FW&A in prescription drug claims ▪ $7M of opportunity in Equipment Repair Bill Invoicing Audits • Transportation • Pharma & Healthcare ROI is High! Payment time is Short!
- 17. How the SolutionWorks
- 18. Cloud Native ServerlessArchitecture
- 19. Databricks Integration Batch • JobsAPI • 2.0/jobs/run-now • Python Task • Python Params Interactive • Notebook Task • Notebook Params
- 20. Wizard Driven AIAnomaly Detection Thank You! Please share your feedback! Feel free to reach out https://www.linkedin.com/in/naomikaduwela/ https://www.linkedin.com/in/rajeshin/