Tim Santos

Assistant Director, Ernst & Young LLP

Tim is an Assistant Director and ML Operations Lead in the Client Technology AI Practice at EY. He has been working on AI products in the financial services involving large scale predictive models, document and NLP solutions, and Automated ML. He is an experienced AI and Data Scientist focused on the end-to-end AI solutions from experimentation to industrialisation. His experience spans from bleeding edge R&D to nationwide-scale telecommunications ML models. His technical interests include speech processing, deep learning, NLP, Augmented Reality, and Digital Signal Processing. When he’s not developing models or doing technical work, Tim also facilitates workshops on how to use improvisational theatre principles in data storytelling, design thinking, and agile methodology.

Past sessions

EY helps clients establish their data- and AI-driven transformation strategies, operationalise their AI governance frameworks, as well as build and monitor AI solutions. In this presentation we discuss how we have approached the nuances of building AI solutions in financial services, and how a highly-regulated industry meets innovation with experiment-driven emerging technologies.

The adoption of AI as a critical component to the future of financial services has been widely recognised. AI does enable the creation of innovative financial products and personalised services. It also derives value from improving processes and services through intelligent automation. AI has made great strides, particularly in machine learning. However, these advanced methods require vast amounts of good quality data for models to learn from. This is a great challenge in financial sector due to a multitude of factors. We discuss these challenges and how we solved some of them.

The talk covers our experience in building models where data is scarce or highly restricted, our learnings from deploying models in multiple geographies and jurisdictions, and how we monitor models where data can drift because of changes in customer behaviour, degrading data quality, or new legislation. Good quality data is a big problem in many sectors, but it becomes more prevalent in the financial sector due to incomplete data sources, biases and imbalances, among others. With pressure from regulators, privacy concerns and restrictions, this often leads to very small samples of usable data. We tackle the above challenges with various methods approaches, such as synthetic data generation, data anonymisation, missing data prediction, and transfer learning, among others.

We also believe that domain expertise remains an integral part in maintaining a healthy and successful AI ecosystem, we will also discuss how we have embedded automated and human-in-the-loop guardrails to capture domain knowledge and ensure trust in the AI solutions we build for our clients.

Speakers: Tim Santos and Mustafa Somalya