The Data Chief | Episode 55

Swiss Re’s SVP of P&C R&D, Jerry Gupta on AI Innovation and Building Revenue-Enhacing Data Models

Jerry Gupta

Senior Vice President, P&C R&D

Swiss RE

Current EpisodeEP55: Swiss Re’s SVP of P&C R&D, Jerry Gupta on AI Innovation and Building Revenue-Enhacing Data Models
00:00
00:00

Episode Overview

Swiss RE’s SVP of P&C R&D, Jerry Gupta is a firm believer in the revenue-generating potential and AI and machine learning. On this episode of The Data Chief, Jerry offers his view on innovative data models, frameworks for building and operationalizing these models, what makes a good data scientist, and why it will always be more challenging for data teams to maintain existing models vs. innovate new ones.

Key Moments:

  • The difference between cost-cutting versus revenue-enhancing initiatives (3:20) 
  • Jerry describes Dimitris Bertsimas’ ”Data Model Framework” (07:50)
  • Why the quality of your output depends on the quality of your input (11:35) 
  • How to address the shortfall of data scientists in the job market (13:48) 
  • Evaluating the prominence and relevancy of data (20:01)
  • Jerry’s thoughts on why businesses aren’t operationalizing data models (22:32)
  • Balancing data model interpretability and accuracy (28:40)
  • Building vs. maintaining your data models (33:16)
  • Blockchain innovation (39:15)
  • What should inform regulation of ethical AI and machine learning? (45:33)

Key Takeaways:

  • What are the skills needed to be a sucessful data scientist? (13:48)
    “If you look at LinkedIn, look at job descriptions for data scientists…and you'll see 90% are overweight on the technical aspects. Yeah. So [businesses] will say, we want a PhD. They will say, we want Java, Jango, and this or that. And they will sort of discount the business expertise. Research has shown however that in the early stages, business savvy, generalists trump tech specialists.”

    According to Jerry, great data scientists needs to have a broad understanding of how to apply technical and mathematical knowledge to business, and that includes communicating clearly and simply to the rest of the stakeholders about what the numbers are saying. The problem with a lot of data scientists positions going unfilled is the underlying issue with seeking over-qualified statisticians, aka Ph.D. level, and missing the importance of business accumen and data storytelling.
  • How do we create a habit of setting clear value statements for models? (27:56)
    “... The most critical balance is between interpretability and accuracy. Which is, are you going to prioritize accuracy or are you going to prioritize interpretability?”

    Being clear and consistent with the value that you are seeking to gain from your model will help you to stay on track as the data and the outcomes evolve. Gupta mentions the need to find a balance between interpretability and accuracy and then one with accuracy and precision. According to Jerry, beyond that you need to reconcile accuracy and generalizability. Arbitrarily chasing one of these objectives at the expense of the others puts the accuracy of the model at risk. 
  • How can we better inform the regulation of AI and machine learning? (45:33)

    “When I look at data science any model you build, you have to look at it through two lenses. One is the lens of the data quality. The other one is human impact.”

    As the bounds of personal privacy are being questioned, the clamor for effective ethical standards and governance grows louder. Whose responsibility is it for the use of gathered personal data to be used responsibly? Whose ethics are we abiding by? The answer to all of this is multi-faceted and evolving as culture moves to catch up with technology.

Key Quotes:

Model maintenance is probably one of the most neglected, but also… one of the most expensive elements of a data science project over the long term.

The first balance you have to find, which I believe in my mind, the most critical balance is between interpretability and accuracy. Which is, are you going to prioritize? I almost always prioritize interpretability.

Data scientists are excited by building new things by moving the needle.

When I look at data science any model you build, you have to look at it through two lenses. One is the lens of the data quality. The other one is human impact.

Mentions:

Bio:

Jerry Gupta is a Senior Vice President at Swiss Re, one of the world's leading providers of reinsurance and insurance. He is an executive with experience leading data science and technology initiatives, and managing both business and technical teams. At Swiss Re, he leads Tech Enabled Data Driven Innovation with a goal of developing new products/solutions and taking them to market. Previously he was the Global Head of Program Management at Amazon.

Prior to that Jerry helped launch the Innovation and Venture groups at Liberty Mutual Insurance. Jerry is a technologist and data scientist, experienced in finding Product-Market fit and developing user experiences that delight customers. He has launched new businesses both as an entrepreneur and within Fortune 100 setting. He has conducted due-diligence on transactions worth over $3B in aggregate value and has raised over $25M in private placements. In addition, he has been on the board of or an advisor to several start-ups in the US, India and Europe.

Jerry has an MBA from MIT Sloan School of Management, and an MS in Data Science from Northwestern University. He also has an MS in Computer Sciences from Bentley University.


BACK TO THE DATA CHIEF