Data Science Meetup - ATB - May 16, 2018.pptx
- 1. Collaborating with University of Alberta to Build a Real-Time Artificial Intelligence System Gunjan Kaur Enterprise Data Science ATB Financial May 16, 2018, Lunchalytics
- 2. Background Gunjan (Senior Analytics Manager) ● Economics and Finance (UofA) ● 4 Years (ATB - Enterprise Data Science, Credit risk) ● Other experience - FX modelling 2 “Our role is simple - to transform the world of banking, exploit the power of AI and machine learning to create new insights that will drive better solutions for our customers and create an exceptional understanding of their needs.”
- 3. 3 Partnership Between ATB Financial and University of Alberta ● ATB signed a multi-year research partnership with UAlberta ● Goal: To push and test the boundaries of research by connecting a community of diverse backgrounds across science, mathematics, and economics and bring this research to production quickly to solve problems for ATB customers https://ai.atb.com/
- 4. 4 Strategy - Why? ● ATB’s vision statement ● Create happiness ● Provide exceptional banking services - something you do for customers, not to customers ● Leverage new tools and technology to deliver outstanding experiences for team members and customers ● To lead the way in applications of AI and machine learning to make banking, and life, better for Albertans
- 5. 5 Current Research: Churn Early Warning ● Goal: apply reinforcement learning to customer behaviour data to detect unhappy customers and remedy the situation before they leave ATB ● Comprehensive, non black box churn model that gives understanding to the underlying mechanism of churn at ATB ● Dr. Chris Frei (prof. Math. and Stats. Sciences UofA), Bohdan Horak, Joshua Mitra (grad students UofA) ● Mark Sebestyen (product manager, Data Scientist, ATB)
- 6. 6 ● Goal: use state of the art reinforcement learning models to create the best possible online customer experience ● Borrow game theoretic RL advancements and solutions to the banking context, become the decision engine of atb.com and ATB mobile banking platforms, e.g. NextBestConversation ● Dustin Morrill (Graduate student, Comp.Sci. UofA) ● Mark Sebestyen (Product Manager, Data Scientist, ATB) Current Research: Customer Experience RL Lab
- 7. Current Research: Real-Time Artificial Intelligence System 7 ● Goal: Utilize cutting edge techniques to analyze data streams in real time, identify and act upon changes and anomalies in customer behaviour in real time ● U of A Researchers: Dr. Omid Ardakanian, Dr. Hamzeh Khazaei ● ATB staff: Team of ATB data scientists and developers What is Normal ?
- 8. Problem Statement: Why ? ● Current state: Advanced analytics based on batch updated/static data ● Derivation of greater insights from stream data vs summarized data/calculations $%^^^^^^^^$##$$$$$$$@@@@^*&&&&&& vs average spend = $$$ ● Reusable components of real time analytics infrastructure between different use cases: Fraud/System monitoring Churn/customer servicing needs Account management ● Establish a fusion platform - set of interfaces that will enable ATB and UofA team members created components to co-exist and interact seamlessly 8
- 9. Early Prototype in Google Cloud Platform ● Goal: Provision of a flexible, portable and adaptable point of departure for UofA team research ● Why Google Cloud Platform ? Cloud based infrastructure with auto scaling capabilities - virtual machines, tensorflow for ML, storage, querying and streaming capabilities 9 Simulated Data Data Preprocessing, Model Deployment Push/Pull, Subscribers Querying, Final output, Real Time Dashboard
- 10. Next Steps and Key Learnings ● Connect to real-time data streams from point-of-sale systems ● Microservices - containerization ● Microservices - creation of seamless interface for interaction between different components ● Amend the prototype to various use cases: Account Monitoring, Customer Servicing (different data streams, ML algo) ● Build out research network further - strengthen partnerships with academia ● Cost savings from quick prototyping of PoC, rapid deployment of solutions 10
- 12. Questions 12 Please connect with us if you’re interested in learning more about Data Science at ATB ! Dmitriy Volinskiy - [email protected] Tyler Dauphinee - [email protected] Gunjan Kaur - [email protected]