How FinTech is Evolving Risk and Compliance Through AI
Fintech interest runs strong in NYC and we had a full house of ~150 attendees for the October 10th event on evolving risk and compliance through AI at the Betaworks Studios offices. There was a great mix of innovative entrepreneurs in Fintech as well as Fortune 100 leaders in attendance.
Risk and compliance spending is in the tens of billions and still growing in the face of higher amounts and frequency of fines. Legacy techniques are not well suited to the challenges of the modern data set — challenges defined by dimensionality, sparsity, unstructured and unlabeled data. Machine learning leveraging the combination of unsupervised learning, supervised prediction and extreme transparency brings the promise of improving outcomes while lowering costs.
For example, Alex Baghdjian, Financial Services Strategy Lead at Ayasdi, spoke about an Ayasdi client who saved $56M by reducing false positives in money laundering prevention analysis by 26% using machine learning progressive learning AI models. Alex started as the lead speaker of the night. He shared the expecting a conservative compliance function to rip out existing systems to replace with new ones is an unlikely path to success. It’s best to insert functionality by augmenting existing pipelines. As far as approaches, a blend of unsupervised learning to generate segments and supervised learning to predict behavior and thresholds for segments in an explainable manner provided the best results.
Lastly, predictive insights can also come from the shape of data and a topological data analysis can enhance results compared to plain clustering. Members interested in learning more about the Topological Data Analysis technique can find more at http://www.ayasdi.com/social/TDAintroduction/.
Alexander Fleiss, CEO of Rebellion Research and Cornell Financial Engineering Instructor, followed up with thought-provoking insights on the application of machine learning to trading and portfolio construction. According to Rebellion Research’s models, currencies are more indicative of economic direction than commodities. Their initial models also tried to capture and model based on over 2,000 features but that led to much noise and model instability and much better results came from a more targeted approach with less factors.
Perhaps the most motivating part of the night was the candid response that most companies don’t have great data, especially at the genesis of a project, but it shouldn’t dissuade people from starting a machine learning journey and beginning to improve.
Conversations continued amongst the audience until closing time at Betaworks Studios.
The Bots and AI team is looking forward to the next Bots and AI event will be on using machine learning to build better teams!
