Understanding Algorithmic Asset Valuations with Luke Arrigoni

Luke Arrigoni has spent the last fifteen years working with some of the top brands in the United States to build machine learning and data science programs, even before the term was coined. He has worked with a variety of companies, including UPS, J&J, Getty, AT&T, Stryker, Goldman Sachs, CAA, FOX, and Sephora.

Luke has media experience explaining to audiences at Bloomberg, WIRED, and other venues the impact of data on politics, machine learning, blockchain, and other ways that technology intersects in everyone’s day-to-day life. He previously built and sold an econometrics company and has knowledge of Sino-American trade relations. He worked for several companies, including J&J and Samsung, where he learned how to do business with technology and raise products in China.

Luke enjoys energizing people about otherwise dull topics like data by weaving a meaningful narrative around how it applies to everyone, even if you don’t realize it. Luke would be a great guest to talk about technology and its impact on industries.

Episode Highlights Here:

Luke Arrigoni:

So what you do is you build an ML program that takes all of your past data. Then you would basically create a category of these prototypes of people I want to target

Pierce York:

How do we use this to our advantage let’s say in real estate or a business? We’re trying to grow the business, so we can kind of like sell it. Putting in these systems and processes. What’s the best way to leverage AI and social media in order to generate revenue?

Luke Arrigoni:

I actually used to do machine learning for real estate brokerage. The first thing I did like, right out of high school was I figured out how to do basically comps, like how much is the house worth, so the house would come off the MLS, and we’d value it right away. This is like pre-Zillow, and then we’d be able to go in and make an offer on it before anyone else saw it. So there are two different kinds of categories for like real estate or other things that you just mentioned. The first is like a back-end one, which is what I just talked about, can you build a good strategy for acquiring assets? And there’s a second one is a front-end strategy, which is how do you get clients, right? Or how do you interact with clients?

Pierce York:

Okay, so let’s do two streams right here.

Luke Arrigoni:

You can think about the back-end and front-end. Most programs will have a back-end than the front end. How do I make my process better? How do I gather more people to throw in my funnel or once they’re in my funnel? So here’s a good example, the front end, the cooler one, which is, let’s say that I had $100,000, to spend on pay per click or some kind of marketing campaign, you could just say, here’s my intuition around who I think is going to be clicking on these things. Or you could build a machine learning algorithm that says, based on all of the experience I’ve had in the past, and the people that do buy from me, I’m going to filter it out and say, these are the people I want to target. So what you do is you build an ML program that takes all of your past data. Then you would basically create a category of these prototypes of people I want to target, then you’d go on Facebook or Google and you’d buy from those categories.

Pierce York:

Essentially, how complicated is it to build a machine learning system?

Luke Arrigoni:

It’s really hard at scale. So if you have billions of rows, like some of the people I’ve mentioned, my clients, they have enormous systems. On that scale, nothing is trivial. At scale, I can’t do anything on a single machine. You see everything about everything in terms of like 100 machines. But if you were to go on a tutorial today and learn how to do it, and you knew Python already, you’d actually be surprised by how well you can interact with the frameworks. But then if you wanted to get below the tip of that iceberg it’s years of math to understand it, but not a lot of people are operating at that level. So there’s like an operator’s for machine learning that would take you maybe a year or so. Then there’s like, I want to be an expert in it. That would take you a decade.