Artificial Intelligence and Analytic Ops to Continuously Improve Business Outcomes
Download as PPTX, PDF6 likes1,013 views
Analytic Ops is an approach that focuses on continuously improving business outcomes through artificial intelligence by getting AI solutions into production quickly while ensuring regulatory compliance. It addresses typical challenges where only 15% of advanced analytics projects reach production due to underestimating complexity and lack of agility. Analytic Ops prioritizes production, focuses on business value, and allows for iterative changes through agile processes and best practices from software development. This enables the creation of sustainable data products and models in a fraction of the usual time.
Artificial Intelligence and Analytic Ops to Continuously Improve Business Outcomes
- 1. Artificial Intelligence and AnalyticOps Continuously Improve Business Outcomes Nick Switanek, PhD Marketing Director for Artificial Intelligence 4/18/2018
- 2. 2 Why Analytic Ops Analytic Ops Case Study Key Features of Analytic Ops Solutions Agenda © 2018 Teradata2
- 3. 3 The Race is On: AI Is Happening in the Enterprise PayPal WalmartJohn Deere Lowe’s Wells FargoJP Morgan
- 4. 4 Typical Scenario: The Value Challenge Value from Advanced Analytics is not realized until you reach production © 2018 Teradata Years 15% 6 Months Time-to-market of a new data product in typical large- scale organization. By time product is finalized, it’s already obsolete. Only 15% of Big Data projects reach production (Gartner 2015) Underestimated complexity, lack of agile iteration with business requirements. Without work a static data product will lose value completely within a typical period of 6 months
- 5. 5 Analytic Ops: Overcoming The Value Challenge Prioritize production and focus on business value from the start. © 2018 Teradata Focus Iterate Sustain Build for well-chosen business use case. Build for production. Use best practices. Follow agile to bring typical project from 1 year to 3 months. Anticipate and integrate change. Inflexible projects are doomed to failure. Allowing for changes in requirements and understanding ensures success. Best practices from application/software development enable the creation of sustainable data products quickly. Think Big. Start Smart. Scale Fast.
- 6. 6 Analytic Ops: Overcoming The Value Challenge © 2018 Teradata
- 7. 7 Value into Production Quickly: Robust and Compliant Benefits of Analytic Ops • Reduce time to market • Improve quality of product • Reduce maintenance overhead • Ensure auditability and regulatory compliance Which departments benefit? • Business end users: receive a better product faster • Data systems owners: governance & clearer processes • Data science teams: optimization & automation • IT support: reduced operational overhead Toolkits Processes Best Practices Accelerators Contents of Analytic Ops
- 8. 8 © 2018 Teradata Analytics Ops Case Study: Fraud Detection
- 9. 9 Fraud Types: Customer Initiated © 2018 Teradata
- 10. 10 Fraud Types: Fraudster Initiated © 2018 Teradata
- 11. 11 Challenges for Fraud Detection © 2018 Teradata Legacy systems can’t keep up on their own
- 12. 12 Banking Anti-Fraud Solution © 2018 Teradata • Real-time data integration • Security and protocol: follow existing bank procedures • Organization and integration of silos of data Data Modeling, Pipeline & Ingestion • Operationalize insights quickly and continually • Enhance collaboration among data scientists • Improve intelligibility of complex DL models Machine Learning & Artificial Intelligence • Monitor and manage many models running in production at same time • Integrate traditional ML and deep learning into tried & true rules engines Model Management Framework
- 13. 13 A Framework to Enable Analytic Ops Capabilities © 2018 Teradata
- 14. 14 Analytic Ops Enhances Existing Detection Systems © 2018 Teradata
- 15. 15 Deep Learning: Results on fraud verification dataset © 2018 Teradata • Ensemble model (AUC 0.89) • ConvNets (AUC 0.95) • LSTM (AUC 0.90) • ResNet (AUC 0.94) >30% increase in detection >40% reduction in false positives Massive operational cost reduction
- 16. 16 Key Requirement: Model Interpretability © 2018 Teradata • We have deployed LIME (Locally Interpretable Model-agnostic Explanation) for customers – Improves trust in the model results – Complies with EU’s General Data Protection Regulation (GDPR)
- 17. 17 Lessons Learned from Analytics Ops at Danske Bank © 2018 Teradata
- 18. 18 © 2018 Teradata Key Features of Analytic Ops
- 19. 19 Analytic Ops Accelerator © 2018 Teradata Platform Analytic Ops DS Lab ProductionQA DeploymentValidation DS Workbench Training Feature Extraction Model Creation Feature Generation Visualization / Explanation Orchestration Execution Model Management, Data Set Management, Versioning Incremental training Check-In … End-to-end framework to facilitate the training, deployment, and management of traditional and deep learning models at scale
- 20. 20 Analytic Ops enhances Model Training Simplifies and automates continuous model training and iterative innovation • Configurable training options (e.g. optimization algorithm, learning rate) • Visibility into model vitals © 2018 Teradata
- 21. 21 Model Deployment Automates production deployment of models • Supports hybrid infrastructure, across cloud and on-premises • Supports Champion/Challenger model management: – Winning model pushed to production with 1 click © 2018 Teradata
- 22. 22 Facilitates collaboration on analytic assets • Version control for both models and data sets • Change approval logging • Version roll back © 2018 Teradata
- 23. 23 Value into Production Quickly: Robust and Compliant Benefits of Analytic Ops • Reduce time to market • Improve quality of product • Reduce maintenance overhead • Ensure auditability and regulatory compliance Which departments benefit? • Business end users: receive a better product faster • Data systems owners: governance & clearer processes • Data science teams: optimization & automation • IT support: reduced operational overhead Toolkits Processes Best Practices Accelerators Contents of Analytic Ops
- 24. 2424 © 2018 Teradata
Editor's Notes
- #4: https://www.technologyreview.com/s/545631/how-paypal-boosts-security-with-artificial-intelligence/ http://www.eweek.com/innovation/john-deere-adds-ai-iot-to-farm-equipment http://www.lowesinnovationlabs.com/updates/2017/1/3/lowes-in-adweek http://www.disruptivefinance.co.uk/2017/01/31/artificial-intelligence-jp-morgan-is-showing-the-way/ http://fortune.com/2017/02/10/wells-fargo-artificial-intelligence/ https://venturebeat.com/2017/07/11/how-walmart-uses-ai-to-serve-140-million-customers-a-week/