On this week's episode I chat with Freddie Huynh, a veteran data scientist in the credit risk space. Most notably, Freddie oversaw the development of the FICO Score from 1996 to 2015.
It was really before analytics and data is what it is today. In a way, Freddie is one of the OG data scientists, one of the original guys out in that space. During our chat we discuss how the FICO Score works, we break down some of the urban myths about how scores are calculated, and we chat about the evolution of the data science industry. Enjoy.
Transcript (this is an automated transcript):
MPD: Freddy. Thanks for being on the show today.
Freddie Huynh: Hey Mark. Really appreciate you having me. Thank you.
MPD: Can we start off with you giving an overview
Freddie Huynh: of your background? Yeah, you bet. So I'm I'm a, credit risk expert, credit scoring expert.
I spent the bulk of my. Professional career at fi ICO 18 and a half years at FICO. I tell people the reason why I stayed there so long is that I had a very unique responsibility. I was basically in charge of the company's flagship product, the FICO score for a a really large chunk of my career.
So I felt I had the best job of the company. It's a product that impacts virtually almost, a very large portion of the us consumer popul. It's a ubiquitous in financial services. And quite frankly, I thought they were a little crazy at giving me the keys to the kingdom. But it was it was a responsibility which I really treasured.
One that was really valuable, loved it toward the end of my career. As much as I loved my job there, I felt the opportunity cost of staying at FICO was too was just too great. And so then I started I started putting my feelers out and I went to. Analytic consulting company for two years.
And before I latched onto a a small lending company called freedom financial asset manage. There I headed up credit risk for about three and a half years before I moved over to freedom debt relief, their sister company, focusing on the debt settlement industry and within the debt settlement industry.
What my main focus is now is heading up all the research that demystifies debt settlement you such that it'll lead to more productive. Conversations with consumer advocates, regulators creditors and whatnot. Yeah. So in a nutshell, that's my journey, mark. That's awesome.
MPD: So sounds
around management. Awesome. I wanna jump into FICO. Obviously that's the big name, everyone. It's a household name at this point. It's been around for a long. You mind giving a little overview of the company, what their in
Freddie Huynh: the world. Yeah. FICO is most is best known for their for the FCO score, which is an all, which is a.
General purpose credit risk assessment, right score that goes from 308, 850 higher the score the better, but basically it provides lenders a, a means to assess the credit risk of an individual. FICO is also. Produces a number of other products. They have a, a suite of very strong fraud solutions.
But by and large they're most well known for their F O score, the credit score.
MPD: And so what they're doing just in a nutshell, make sure I have it is they're gathering data on everybody. I guess at least in the states from a variety of sources, crunching the data, coming up with a score and then selling access to those scores.
To organizations that are lending money. Is that the main use case?
Freddie Huynh: Yeah. I would amend that as follow so F I O actually doesn't do any data aggregation. So they partner with the credit bureau. Experian, Equifax, and TransUnion and the bureaus provide F O with the data representative sample the us consumer population for which FCO can model on.
So F I scores are basically built solely on credit bureau information. So these are the credit histories that are compiled. The various credit bureaus. And then when a lender let's say like a bank of America or a chase is interested in pulling a F I O score. They work directly with the bureaus fi FICO licenses, their algorithms to the credit bureaus for them.
And basically the way they get compensated is through royalties through the sale of the scores.
MPD: OK. So I've been in all the different Experian and the credit bureau. I notice there's always a small discrepancy between my score and one to the other. What's the cause of that? They are they using different data sources.
What if they're all going through FICO? Is it different inputs into the same model that they're using?
Freddie Huynh: Where are they different? Yeah. So I think there's, yeah. So I think there's two sources. So one of the sources is that the algorithms are slightly different from bureau to bureau. And that's and that design is AC is intentional basically, when I was at FICO, our goal was we wanted to make sure that the That each algorithm that we developed for a given bureau was optimized for that particular bureau.
And to the extent that there are nuances in the data that are a little bit different it may reflect itself in a slightly different score. So that, that's one difference. But I would say far and away, the biggest driver of any type of cross bureau differences is going to be slightly different data across the bureau.
For example, if you apply for a car loan, and and a lender only pulls their inquiries from Equifax. That inquiry, that inquiry is only gonna show up on the Equifax report. It may not show up on the experiment report or the TransUnion report, or if a certain data certain.
Data furnitures only report to one of the three bureaus or two of the three bureaus that could also be another source of differences. But what I would say is that though there are differences on average, the differences tend to be small and I would say large differences tend to be more the exception in the rule, but certainly there are differences between barrels.
MPD: Now what is it used for outside of the barrels? Is it or is that the main crux of the FICO?
Freddie Huynh: I'd say. Yeah. So the most use of it is going to be from a from a credit scoring standpoint. But I would say that since late nineties, probably early two thousands, the the consumer score disclosure business has really has really grown as consumers have become more aware of credit scores and the importance of credit scores in their lives.
There is a. For consumers to know, and to monitor their credit scores. And and so that, that is a business, the BDC business direct to consumer business which not only FCO is very interested in, but also the bureaus as well.
MPD: like credit Carmen, that's based a little bit too. Exactly.
Freddie Huynh: Yeah, exactly.
MPD: OK. To give a sense of the significance of this. What would happen if the FICO score went down or if there was no FICO score today, can you kinda illuminate some of our dependencies on this number?
Freddie Huynh: Yeah.
I so these scores are well, so just to provide some context. Yeah. Roughly about 10 billion score, FICO scores were sold a. Which is, so just the sheer volume of scores which are used. And then those scores are used in a variety of contexts across the customer life cycle.
They're not only used in terms of pre-screening to help them target, Hey, who should I send this credit card solicitation to, or this personal loan solicitation to. So in the solicitation part, but also in the acquisition part. So when someone applies for a credit the score gets pulled and that, and that is a component.
It's not the most important component. But it is a very relevant piece of the Of the ultimate credit decision. And then on the back end of the credit life cycle, once someone is a consumer many lenders will pull that score just to monitor the overall credit health of the consumer.
And also it could be used to help in, cross selling across the life cycle as well. So now if that score were to disappear, that would certainly in the short term would create a lot of. Havoc a lot of destruction because because that type of, that general credit risk assessment is needed.
I think what but I think what gets really interesting is over the middle or the long term is then it creates a void like, and who can fill that void in terms of being able to provide this general, as this general credit risk assessment.
MPD: Does anyone else do it now or is FICO kinda the only show in.
Freddie Huynh: No, they, there's certainly other like the bureaus have their own score too. FICO's biggest competitor is something called vantage score. And and they are definitely, gunning for the market as well. Got it.
MPD: Here's the thing that I think credit karma tries to illuminate a little bit for the consumer.
They try to kinda. Atomize or distill down the inputs into the FICO score. They try to make it easier to figure out, Hey, your score is down. You wanna do something about it, pay off this credit card and give you tips. What go, what are the, what is in the FICO score? How is it calculated? How do you think about, someone gets a 600 versus a 700.
How do you get to that?
Freddie Huynh: Yeah. So I'm gonna first talk about a high level and then what I'm gonna do is provide something a little more prescriptive. At a high level, there's this there are main five main components of the fi O score, right? In order of importance. There are, one category that focuses on payment history, which shouldn't be a surprise cuz how likely someone's gonna pay their bills is probably reflective of how they have paid their bills in the.
Okay. There's also the second most important net that first category comprises roughly about 35% of the score. The second category is indebtedness that roughly captures about 30% of the score. And basically we're looking at how much money does a consumer owe within this category variables that are very predictive within the FICO score or things such as credit card utiliz.
Okay. The third, most important category is gonna be age of credit history where or length of credit history. And just because consumers who have established a longer credit history on average, tend to perform or repay better than those who have not. When it comes to the fourth category, there's something which we look at that's called pursuit of credit.
And and not surprisingly consumers who are more credit, hungry who are more aggressive for shopping for credit, they tend to be riskier than those consumers who don't shop for credit. Now, that being said, it it constitutes a, relatively small portion of the score roughly about 10.
And the last category is a little catchall, it looks at types of credit seeing kind of the mix of credit that that consumers have really can provide additional information to, to help refine the the risk assessment there. So as an example in this day and age where it's so easy to get credit card, If you don't have a credit card, there's a little signal there.
And and so the score picks up on that as well. And and that last dimension really kinda only contributes, let's say about 10%. Okay. So I was gonna tell you, I was gonna explain it two ways. And so that's the first way at a high level. But one of the things that I used to always tell or always.
Really emphasize to our clients, particularly, when we would give these kind of scoring seminars, is that, I would tell them like, listen, if there is one thing that I want you to leave from the seminar there are mechanism within the score that actually make it very clear as to what's, to help explain what's co your score, cuz every consumer is different.
For example, though payment history is really important. There are some consumers who most consumers actually pay on time. And understanding what really drives their score is gonna be a little bit different. What I tell consumers is that with every FICOs score, there are up to five, what they call reason codes that are returned to the consumer that will help, that are designed to help explain the consumer.
Hey, why are you scoring? And each of those reasons is really relevant. And in terms of it's it helps explain why the consumer isn't scoring higher and it's ordered in in value of importance. So if the top reason is listen, your proportion of credit balances to credit limits is too high.
That's the number one reason. And the consumer has a very clear idea of what's causing. Probably the last thing I wanna say is that for consumers who have really high scores, those reason codes are probably gonna sound really picky. But at the end of the day, they need an explanation in terms of, this is why you're not scoring higher.
MPD: I have a feeling credit karma and others pick those codes up and that's, what's driving the recommendations. It seems like it's, I get some recommendations on there, pay down this, pay down that I'm sure that those codes being processed. Okay. So you said two things in there that I thought were interesting.
The payment history it's intuitive to me that someone who has historically paid on time for 20 years straight is likely to pay on time, more likely to pay on time tomorrow, where you guys, when you guys drew that conclusion is there I'm presuming there was a level of analysis to find behavioral kind of patterns.
Say, Hey, we know that this correlates with an X percent higher probability and then it was implemented in to the model. Is that the right way, was that I'm assuming that work was done, but you guys, did you go through and figure out this is how each of these effects. Future behavior, do a little bit of historical analysis
Freddie Huynh: or modeling.
Yeah. Yeah, absolutely. There, there's both kind of what you would call univer analysis, where you look at an individual variable and then there's multivariate analysis where you've got model. You wanna see how all these variables interact with each other. But bottom line is, is that it's a tried and true pattern, which we've seen, since, since the Donna time and.
People who've paid their bills in the past, more likely pay their bills in the future. Yeah. And and when you look at payment history, there's really three subcategories. There's severity of delinquency, frequency of delinquency, recency of delinquency. And so even within a given category, there's different flavors, which, which the algorithm is trying to pick up.
Because someone who is. Who's been recent delinquent. Who's been recently delinquent. It may maybe they haven't been severely delinquent yet, but that recent delinquency that may, that's a strong indication of, things that come on average.
MPD: Very interesting. I've always was funny that actually getting your score checked, affected the score itself.
Freddie Huynh: Ah, that is a that is an urban myth consumer. So yeah consumers who check their scores whether it be, through credit karma, my FICO, these, various sites like that that in kind of industry jargon, that leads to what they call a soft.
And soft inquiries are not included, are not factored into credit score calculations, but there needs to be a record that, Hey, your credit, your credit record was pulled for this. So the good news there, mark is that it does not that won't that won't hammer your your score and what's a hard.
A hard inquiry is basically any consumer initiated search for credit. So if a consumer let's say responds to a, let's say a credit card solicitation, or if a consumer let's say, goes to a car dealership and applies for a car loan. Or if a consumer let's say, starts looking for a mortgage, got it. That is they're initiating. A, a search for
MPD: credit, right? And so you're saying, Hey, there's a pursuit of credit here, which is a signal that they're taking on more liability, which is a signal that the person after them should think that they're less fundable because they've just taken, they're probably taking on some more liability elsewhere.
Freddie Huynh: Is that the logic, why the hard inquiry
affects it. Yeah. Again, what we've seen time and time again, is that people who are searching for credit on average tend to be riskier than people who don't. What's really interesting, with that is that if you look at that variable through like across the economic cycle, it, it becomes more relevant, more predictive during an economic downturn than it is.
Let's say during good economic times. And, the logic there is that, during an economic downturn, struggling consumers use credit as like this financial lifeboat and and that search for credit, becomes you becomes a stronger signal.
Another way to look at it is again, is like during times of economic uncertainty, really good credit risk. They know, it's you know what, it's not a good time to apply for credit. Things are pretty crazy right now. Yeah. That's how we would look at that type of information,
MPD: super helpful.
On the VC side, I've seen a lot of companies over the years trying to disrupt the FICO score, trying to find new ways to come at the analysis. And it seems like none of them have really set roots. Why is it so hard for new entrants in this market?
Freddie Huynh: Ooh yeah. I would say for a, a number of reasons, so the where I think the value of FCO comes is that, if FICO nowadays.
Really kinda serves as what I would call like lingua franca in the financial services industry. Virtually everyone within financial services knows what a FCO is. And also it means something to them. If you talk to a lot of investors or you talk to a lot of bankers, and they're trying to just at a high level, understand the quality of a certain portfolio.
They'll, they go, what's your weighted average. And so it's become a common currency, for that and to, and for for a new company to to be able to disrupt, to achieve that very difficult. I would say that there are really two events that allowed FICO to achieve that status.
And I would also find that that today it would be a lot harder for, If there was no FICO score today. And if someone had an ideas like, okay, I've got this great idea for one credit score to rule them. All right. I would argue it probably wouldn't happen. And and I can elaborate why, but but I think the two big events that that come to mind for me is I believe it was in 1995 or maybe 1996, where the GSEs essentially endorsed F I.
They basically said, Hey within if you're underwriting this mortgage we're gonna need a FICO score for that. And that really cemented F I O really put it on the map. I would say prior to that, one of the other things that allowed FICO to take a foothold is the fact that FCO is available from.
At all of the three credit reporting agencies. And I, so any new player would need to make, have a score that's available, available widespread and and would be accepted by all players. I think that, I think that's a really tough bar in today's age.
I think one of the other things that I alluded to is this notion just Hey, in, in today's day and age, I would think it'd be really hard to create a brand new FICO score. And I think the reason why is when fi FICO was originally conceived, it was at a time when ALA where analytics really wasn't at the forefront.
And, back then, the bureaus focused on the data and a company like FICOs, just focused on the analytics. Over time, everybody understands the value of data. Everybody understands the value analytics now, and there is no way that the bureaus are going to kinda give up their revenue.
For something that they can do on their own now, they've invested heavily in developing their own analytic capabilities. They're not gonna wanna split the revenue with someone else. And so I think one of the things that would need to happen in order for kind of for FICO to be supplanted is that you're gonna need to have a data source that's better richer, more comprehensive, more reliable than the existing set.
It's going to need to be it's gonna need to also meet a lot of scrutiny from the lenders, at the end of the day lenders pull FICO because they trust it. They've used it over time and, FICO doesn't cut corners and it's something which they can rely on.
And I think that's gonna be, another burden. Or a barrier for entry for a new participant.
MPD: And you mentioned the credit bureaus have developed their own scoring systems and done their own analysis. Are those competitive with FICO in the sense that, the bureaus may stop using F I O at some point, are they built
Freddie Huynh: their own tool inside.
Yeah. So I, I would answer it this way, mark is that the bureaus definitely have their own FICO competitor. And, if it, in a vacuum, it was up to them, that's what they would sell. They basically get, keep more of the revenue share there.
But at the end of the day, it's the lenders that drive the demand. And so if lenders continu to ask for FICO score, Fi scores are gonna need to be, sold. Got
MPD: it. It's a bit of a brand level network effect since everyone's used to the name and trust.
Freddie Huynh: It's a, yeah, exactly that.
MPD: The one most common complaints I've heard from people trying to disrupt the space is that using historical financial data is not the most optimal way to actually predict future behavior. And I think they're grabbing onto use cases and scenarios where. People have gone through challenging times, but their times are no longer challenging and there's other data out there like income or other things that might indicate that they're financially stable in a way that they hadn't been prior and therefore more likely to pay bills on time.
What is your response to that? How effective is historical data in particular future?
Freddie Huynh: Yeah, so I would say that in general Within just the the field of kind of credit risk assessment more data as long as it's accurate data can only help. Now so what, so one thing that you mentioned is Freddy, it's pretty obvious that income, is relevant.
How come the fi Pcore doesn't, use income, and so I'm gonna first say, I agree. Income is. But I would also say, but guess what, the, the bureaus don't, store don't have a, a reliable source of income that can be used in the model, but, certainly having an opportunity, to work at a small lender as well, right?
When we are evaluating a consumer, both in terms of their propensity to repay as well as their capacity to repay, having verified. Absolutely critical. Cause that's such an important part of their cash flow. And knowing what their, knowing what their balance sheet is. So I would say again, to that question is yeah, there are other data elements that are incredibly relevant.
And having access to them can really help your decision making. Even when I was when I headed up F OCO development, I never pretended that fi that the F O score was like, the end all be all. It's just, it's one component of the decision making process.
But there are certainly other pieces of information out there, which you want to include in your decision making process as well.
MPD: Another thing I've heard from folks same vein different argument is that there's all new types of data that are emerging through the internet in particular, that could be a good signal.
And one of the ones that's come up that I thought was less obvious to me when I heard it, but there was some arguments for, it was the idea of looking at social relationship, data, people with larger communities of friends and other things from the, some of these social graph. Has anyone in any real earnest, he explored that as a corollary or kinda a, an indicator of ability
Freddie Huynh: to pay.
Yeah. I'm sure that there are some people out there who have explored it. I would say that when it comes to. Evaluating new data sources. There's a number of criteria which I go through in terms of, Hey, is this a legitimate data source for me to really to focus on, right?
And so one of the, one of the criteria that kind of that I focus on is that one thing that happens in credit risk, which is very different than a lot of other kind of data science applications, is that in credit risk? If you deny someone's with credit and you use the credit score. And that was like to really kinda drive the basis of that denial, right?
You are legally obligated to tell the consumer why this score DEC decline them. And as a result, you need to tell them, I think it's like up to four reasons that will explain to them why they didn't why they weren't extended credit now. I think what's really important in, in that type of context is it's not only do you have to explain them, why they didn't get credit, but it also needs to be, within that context, it's gonna need to be defense.
And so are you gonna feel comfortable as a lender telling someone, Hey, you know what, I'm not gonna give you this credit card because you don't have enough friends on Facebook. Or because, we often refer to that as palatability . And it's just that's just not gonna fly.
But if you tell someone, listen, we're not gonna give you credit because guess what, you are currently delinquent on, this wall's Fargo credit. That is defensible, right? Yeah, and again, there's a lot of other criteria out there in terms of why I would consider or not consider certain types of data sources.
But but as an example that you know, that we all use when it comes to things such as, Hey, why don't you wanna use social media data, you just told me more data can only help. Why wouldn't you wanna use it,
MPD: It's just too soft of an argument. One wants to hear that. Yeah. Yeah.
I get it. Even if the signal is in there, you guys wouldn't use it for that reason. Is that right?
Freddie Huynh: Yeah. So again, I
MPD: dunno the signal is in there by the way, but even if it isn't,
Freddie Huynh: even if it's yeah. I where there's gonna be other criteria as well, as an example, it's one of the things that the.
Is if you wanna comply, if you are a data furniture, if you wanna comply with a fair credit reporting act you need to have a dispute mechanism in place. And if and if a lender is citing a piece of data And the consumer doesn't agree with that piece of data. There needs to be a mechanism in place for that consumer to dispute it.
And so do I'll do these other data sources, so the bureaus do have processes to kinda deal with that. Do these other kind of new fangled up and coming, data providers, are they ready to deal with that? So it, yeah, it's again, it's it's a different ballgame.
When it comes to credit risk,
MPD: I love this, this. We never made a bet in the space in any of these companies try to disrupt FICO. But with the stuff you're saying right now proves something that we always talk about internally. It's this idea that you can take a vacuum of a situation of a problem and present a solution.
But if you don't go to the nuances understandings of the industry, the requirements for, filing complaints, Reconciling things and that you would need Twitter or whoever else to have an entire department they don't have, and they're not gonna do. Just, it just speaks to the important practicality of actually understanding these industries and getting into the weeds of it,
Freddie Huynh: Absolutely. One of the things that I think separates like a good data scientist from a great data scientist is domain expertise. You have to know. The ins and outs of how you know, of how your solution is gonna be implemented. And if you know the rules of the game then, then you have to, you have to innovate around that.
But I think that's like such a big part of of successful data science. I love that.
MPD: I actually wanna dive deep into the data science space. But I wanna ask one more FICO question before we move. Another complaint that comes up a lot is bias because of the FCO system and the argument.
I think if I'm gonna distill it properly is that historical data is hard for people who have come up from disadvantaged positions because Hey if they grew up poor or they had dynamics it's this lagging indicator that continues to hold them in their. I understand the points before you're saying, Hey, it's a significant indicator about future behavior.
How do you respond to the folks that are waving a flag that the FICO score is part of a, like a socioeconomic bias in the country.
Freddie Huynh: Yeah. So I think that's really a, it's a very complicated, a question's really thorough question. But it's something that, that comes up a lot. And so the way that I like to think about it is that at the end of the day, what I would want the FICO score to be, if I was a consumer was an impartial evaluator credit risk, now. There is nothing in the score doesn't, the score only listed credit information, there's nothing in there that would be defined as overtly discriminatory. Doesn't look at factors such as race, sex, marital status, religion, all that jazz.
But it doesn't mean that, Hey is there something about is there something in the data which provides, let's say which disproportionately impacts certain consumer. But but at the end of the day, the way that I would, focus on it is like you want the algorithm to be to be as objective and as unbiased as possible.
And we're gonna get back into the biased conversation. And it, you cannot let it overtly use these, prohibited, bases or categories. But I always like to give an analogy, is let's say that. Let's say that I'm a math teacher and I create a math test.
And there are certain students who just don't perform well in the math test. My job isn't to give them okay. What make 'em feel better and give 'em all A's, that test needs to mean. Where that test is designed to basically assess their aptitude in mathematics, whatever they're studying.
And I wouldn't necessarily, want to to engineer that score a little bit different in order to, in order to quote unquote make them feel better. And the same thing with the goes to credit scores, there may be certain groups who, Variety of socioeconomic reasons aren't are maybe scoring lower, but that doesn't mean you wanna socially engineer the score and change different again.
So being like a, a quant or, a data scientist, I always wanna figure understand what's the root of the problem, right? And so the root of the problem, isn't the algorithm. In my unit standpoint, it is what are the true barriers that are preventing these consumers to have access to credit, to that will allow them to eventually create a credit history that would put them along the happy path.
And again it's easier said than. These are things and, I know a lot of, probably a lot of companies that, that you've interacted with are really focused on, Hey, how do we improve financial inclusion? How do we promote financial inclusion? I think one thing where there's still a tremendous amount of opportunity is financial literacy.
And I don't know about you, but like when I look back upon how I developed my own credit history, it was really kinda. Dumb luck. It wasn't, out of anything, I just remember my, my mom told me once she goes, when you get to school, get a credit card and she goes, you better pay it back every time that was it.
I think there are for a lot of people out there, there needs to be things that are a little more prescriptive where you are providing consumers, not only with the product for credit, but also with the education in terms of this is how you use it properly. And if you use it properly, guess what?
You will have an opportunity to be afforded more credit down the road which becomes very relevant. If one day you're interested in, let's say buying a home.
MPD: Thank you, Fred. Moving on the data science side just coming back to this, you've been a data science scientist for a long time.
I. I would assume, and the industry better than I do that you were a data scientist, frankly, before that was a highly popularized phrase
Freddie Huynh: before Zin Vogue, for sure. Absolutely. Yeah. Were you
MPD: using it in your own world or was it used internally or was just that phrase? Not even coined yet. Yeah.
Freddie Huynh: When I saw analytics or yeah.
It was. I started off as a project analyst. There you go. And then, everything focused on analytics and, back then, the company tagline was better decisions through data, and virtually, any company nowadays, if they wanna be successful, that's gotta be a, core competency of this.
MPD: feature somewhere with this. This was FICO was the specialized data science. I was outsourcing for folks at some level in this particular function. That's interesting for sure. For sure. When did data science kind of become a phrase and a thing? How did the industry and the career path evolve for you over the last, decades?
Freddie Huynh: How does this look? Yeah. I'm first of all, I'm trying to remember, like when did data science kinda Was coined. And I'm not sure, I don't know if it was like like mid two thousands or so. But but it, maybe it's like with the emergence of. The wild success of companies Google, Facebook, apple, eBay, PayPal, those types of companies that really, you know these retailers that really brought up to, the forefront and, so I imagined that was a, a key development that kinda spurred the growth of the industry.
But for me, it is. I'm just blown away by all the advancements. Like nowadays, hiring new data scientists. Yeah. The ki I say kids, but I shouldn't call 'em kids, but the, but the, but the people who I've hired. Outta school. They're amazing, the amount that they know certainly dwarfed, what I knew when I, when I graduated and I'm just blown away by what they're capable of.
And and what the industry has really has developed, over, let's say the, the last decade are
MPD: there schools that are really well known for training up data science?
Freddie Huynh: Yeah. I would say that there is there's probably, different flavors.
It, cause data science is more kind of generic term . And within data science, I would say or the way that I approach it, like if I. When I put together my team at freedom financial, I mainly focus my recruiting on people with mathematics degrees, statistics, degrees, operations, research, computer science, economics, but basically the common thread to all that right.
Is like mathemat. All of those disciplines the unifying language is mathematics. And that was really essential for what we're looking for. But one of the things that I, that I think is super, super important is that hiring a, team that comes from a diverse background.
So I went out of my way to make sure, okay, I've got my B and statistic. I need my machine learning walk. I need my AI guy. Where am I gonna get that? You know what? I would love to have an economist, would love to bring that on board. Oh, I love physicists. How does the diverse, yeah.
MPD: How does the diversity of knowledge bases of disciplines play together? So well, why is that important?
Freddie Huynh: And, ah, I just think that if you are faced with a really hard problem the, the more diverse your intellectual gene pool, the better chance you're gonna have to succeed. Cause if you have a problem, but everyone thinks the same way, get up, it really limits. And so if you have a bunch of different players who thinks about problems a little bit differently, then things can get really interesting cuz then they start riffing off of each other.
And so to me, I, again, when I think of building a, a data science team, that is kind what I really, that's one of my guiding principles. Got it.
MPD: If someone out there is trying to get into this field of data science, What should they be studying?
It sounds like there's a lot of answers to that. What's the tried and true most direct path into this
Freddie Huynh: game. Yeah. So what I would say is it would need to be some type of, one of those disciplines that I mentioned earlier, whether it be math, stack com, sci physics kind, but first of all the first thing I would tell him was like, you gotta.
You really gotta want it. You really gotta love it, cuz it is a competitive field. And also I think good data science practitioners, really good coders, so they can't be afraid to roll up their, roll their sleeves up and really dig into the coding. Really getting, really get into the data.
And one of the things that I look for are people who are. AEPT at managing data, as opposed to there are certain people who goes I'm not gonna do that. Someone's gonna help pull the data for me. I want people, I want, I think a really good practitioner.
Likes getting their hands dirty with the data, because the more you work with the data, the more you can familiarize yourself, the more you can make observations. And ultimately the more you can start making generating hypotheses and, understanding things. But bottom line is I think, any there's a number of disciplines I do think.
And, I think different people may approach it differently. I would say that I have a strong preference for candidates with advanced degrees, mainly because the field is so competitive, that's one easy way for me to differentiate, candidates. But but yeah,
MPD: so there, there's an interesting parallel I'm seeing in this where it looks in some ways like the computer scientist develop.
Market right when it became super competitive in the us, what we started to see in the developer side was an increasing look offshore, right? Prices came up so high in the us. There was a shortage of talent and people started looking for offshore solutions. We've got one company in the stack, a company called spoke that does BPO business process outsourcing.
And one of the areas that's been in really high demand for them. And they've been growing in very rapidly is data science, largely with staff and labor in the Philippines at a much lower rate and cost structured than in the us. Is that on the radar for you yet? Are you seeing offshore providers scale up into this market?
Has that become culturally
Freddie Huynh: accepted? Whether or not it's accepted is really boils down to the, the preference of an individual company. There are a lot of companies that basically want to keep things or, they wanna generate their own IP. They wanna keep their own IP.
They don't wanna outsource it. Now, there are certain companies which view that maybe there are certain functions which are lower value, which may be better served by, an offshore, source. So I think there's a spectrum there. There's certainly there's certainly talented there's talent internationally for.
And it's again, just like anything. You're, you need to find good leadership. And you also can't be an absentee landlord when when you're working with them, you need to be providing the right direction in
MPD: the right guidance.
This is the exact same narrative I was hearing in 2006, 2010 to 10 around development, there was a spectrum of tolerance for off. A lot of venture firms at the time when I first started this business were extremely allergic. If a company showed up and they had any out, outsource talent on the dev team, it was unfundable.
And now it's all the commonplace in startups, right? I think there's a consensus that you need to have a CTO, VP engineering, someone sitting in your team to oversee it, to make sure everyone's speaking the same language and there's leadership to your point. But the. The pattern has been an increasing percentage of outsourced resources.
I wonder if we're gonna see the same thing here?
Freddie Huynh: Yeah. Cool. All right. In all likelihood. Yeah. We're probably seeing it to a certain degree and again, the the revenue pressures almost make it mandatory to explore it. Yeah, you have to at one. Okay. What
MPD: does the data science industry need?
For folks listening that might wanna roll up their sleeves and start something, how could this market
Freddie Huynh: be improved? First of all, just the the stuff that they're, what the industry has come up with is, something which continues to impress me lo me way but one of the things.
It, if I were to be an armchair, critic of this, I would say, with all the, the speed and advancements that have been introduced in data science, I'd like to see a greater degree of self policing in the industry around what just ensure as just because you can do something, should you do something.
And and so I wouldn't consider myself an AI expert, but when I read about these conversations about about the potential dangers of AI, right? What is, what safeguards do we really have in place to to prevent these kind of horror stories, from actually even happening, and just from a business perspective, right? The way that I look at is. Change is inevitable. I get that, but you always want to change in your own terms. And I think one of the worst things that can for any industry, or one of the challenges it's probably a more diplomatic way of saying it is that if if they start to get regulated if you regulate yourself and it's not just lip service.
The likelihood that you're gonna have to face, those type of heads probably gonna be less. And at the end of the day you wanna do it cuz it's probably the right thing to do. But it's always hard because I know, all these companies wanna focus on hyper growth and and sometimes, these may be competing objectives.
MPD: Look Freddy. Hey, it's been a joy. Thank you for shedding. So much light on the space. And being on the show.
Freddie Huynh: Yeah, absolutely. Really appreciate having me, mark. Yeah, that's good. That's a great time. Thank you.
MPD: Cycle scores might not be rock and roll. That was super helpful for a tech nerd venture investor. Like me. We see a lot of stuff in and around the space, understanding the underpinnings of the logic and how to think about the industry. Barriers is very helpful. Hopefully if you're interested in data science, it was useful for you as well.
If you like what you heard, please look us up with a like, or a five star review. You'd better share with a friend. You can find me on Twitter at M P D. And to hear more of my conversations with innovators, subscribe on YouTube or any major podcast platform, just search for innovation. Mark
Freddie Huynh: Peter Davis.