Interview with Scott Davis of Valorem on the World of Data Science

Hello everyone. My name is Ian Peterman, host of Conscious Design Podcast. And here today I'm talking with Scott Davis, who's worked as a data scientist and now manages data scientists and has worked in the data field for many years. Previously a civil engineer as well, but Data is a huge thing that we now can't escape. Every startup touches it. And Scott, you've worked with small and large businesses. So let's, you know, let's dive in. Like, what's— a startup looks at data and they're oftentimes scared. They go, okay, well, it's a lot. It's look at data. And then they see an ocean and then they turn around and walk another direction. So what's— tell us a little about your experience, like you've worked in it and what, what would you recommend?

I think in general, data has value, but like you said, data is scary because it's a whole nother host of decisions that you have to make. And there's a lot of people vying for your money. So what do you do with data that you either are currently collecting or thinking of collecting? And really the problem that Fortune 500 companies are running into is no different than what a startup's running into, which is really just Access to good clean data. So my one piece of advice is store your data. Storage of data in the cloud is nearly free. It's, it's literally nearly free. And as long as you're not doing anything with it, to store it doesn't really cost anything. So if you're a brand new startup and you have no idea what you might do with your data, don't waste it. Store it. Whatever data you have, just store it. Somewhere down the line, you may find a use for it. And if you have it, then you can use it. If you don't have it, then there's no value in data you didn't collect.

Right. Well, that's interesting you say that Fortune 500 is having the same, same issue. So what— I mean, with a startup, you can imagine like, oh, well, you deleted it, you know, or you stored it somewhere and, and you forgot about it. And, you know, somebody went in and did something, right? Because you're just, you're in startup mode, you're rushing through, uh, and you're trying to, you know, stay alive, raise some money, right? Your, your focus is usually somewhere else. But how, like, how is Fortune 500? Is it just nobody? Everybody is just too scared? Like, what's— how are startups and Fortune 500 sitting in the same problem boat?

It's, it's the thing that everybody's gonna understand. It's that we're all too busy, that going through the steps of storing your data is just another thing to do. You don't know when you might use it, and it's difficult to make those decisions. So how do you get it from wherever it's being generated into its storage location? Do I have to hire somebody to do that? Can I do that myself? Is it breaking along? There's so many questions. So that's really the challenge is just getting Getting off the ground, making that first baby step. Right. Fortune 500 companies are no different. They're busy. They're doing things that they've been doing for years that can't break. So the new data doesn't get prioritized because it's just another task that they have to pick up and do.

Right. It's an extra, extra cost too, especially if you're collecting a lot of data. I would imagine that's a much larger check. Check you're gonna have to write. So, you know, once you— let's say once, once we do store our data, what's— what are the next steps? You know, you're a startup, you're, you're storing it. Let's say you make that decision. All right, we're storing it, we're putting it somewhere. You know, there's a lot of different options. And, and like you said, there's the option, well, how, how do you even determine? Do you hire in? Do you outsource? Do you Right. It's kind of the same thing in the product design world. Well, do you build an engineering team? Do you hire it? Do you do both? Do you, do you, what do you, how do you even break that down into a logical conversation with yourself of what, which areas, what are pluses and minuses of each way?

Yeah, really it comes down to what's the value in your data. So I'll answer that question by telling you a story that happened recently. A startup came to me and they'd been doing exceptionally well. So in the healthcare space, making vaccines, and they came and said, hey, we want to get better at making vaccines. Our current process from going from initial all the way through to something that is ready to be shipped takes 3 years. And it's just not competitive enough. We have to be able to get that down faster. Our, our Our competitors are doing it in a year and a half.

Hmm.

Well, first of all, their main problem is they didn't store the data, so we didn't have anything to help them figure out how to make that process better. Let's assume you were doing what we just proposed, store your data. They then would say, hey, the value of going from 3 years to a year and a half at our current size and structure is about $1 million. So then you can say, hey, I could go hire A consulting firm, or I could go freelance and hire somebody and just have them help me. A single person freelancing at $100 an hour. Okay, that makes sense if it's on a million-dollar benefit.

Right.

So you can do that math. Now, if you're much, much smaller and you're looking at something that might benefit you in the neighborhood of tens of thousands or hundreds of thousands, maybe there's something that you pick up and you do on your nights and weekends. Maybe there's somebody that you go to a local college and you ask if there's a use case that some of the, the new college students would like to play with some data, and you can get some, some free help that way. Uh, it really just depends on the value of that data.

Okay.

But as long as you have it, you can use and extract that value, whether it's yourself or a nephew that's learning computer science, right? Somebody's going to have access to the data and they'll be able to do something of value with it.

Right. That makes sense. It's all based on what the value, value of it is. Now you can store it, right? You're, you're storing it. You're— there's processing of it. There's using it in, in an end case. Um, and there's a lot of areas. Is that typically one person? Like, is one person going to be able to manage all those areas? Effectively? I mean, you, you can obviously start— there's, you know, startups, one-person startups. We all can do amazing things without sleep and a lot of caffeine, but there's a limit to that, right? So is it really a realistic thing to say, you know, one person— you mentioned a freelancer— one person be able to really actually be able to do that effectively, or are those kind of areas too different? For one person to effectively manage, really bringing, you know, getting all the value out of that data?

Yeah, and as is typical, the answer is, well, it depends.

Right, always.

So the higher your skill level, the more of that process you can do, but that means that you're most likely going to also be a little bit more expensive. So if you broke it apart, really you have simple storage. A data engineer could be freelanced to do that for very inexpensive because they're literally just taking your data, setting up a process, dropping it, storing it. Process is done. Now there's access to data. If you want to analyze it, you could hire an analyst, somebody that comes in and just looks at your data. You ask them questions, they give you answers. Moving up the tier, you move into data scientists. Data scientists can now run models, build predictions and really help you out. The complexity of that really depends on, on the data you have.

Right. And the data, data is so big, right? So we have everything from the number of people that show up to your website, you know, which is, you can easily set up Google to track and collect or any, any other system all the way to, you know, detailed information on your customers, or like you're talking about, the vaccine company, they're looking to accelerate their process. So that's not even external, has nothing to do with a customer. That's all internal data, which is, you know, whatever that, whatever that looks like for a vaccine company. I've never worked for one. So, you know, so, so we have such a broad range Now, a startup might ask the question, well, if I were going to prioritize one, you know, I don't have enough time to do all of it. And I don't know what the value is yet because I'm a startup or I'm a small business. You know, what was— is there any kind of tier system? You know, as a data scientist, if you're given a project, do you tier— are you able to even tier it into what would be most useful? Is it all just use case?

Yeah.

Like, how is there— is there even a logical process that you could go through if someone came to you and said, here, I have all these options, what do I do? What should I do first? Is that— or is it just a, you know, just the cloud of, well, it totally depends, and there's not really a way to, you know, there's not a formula to it, so to speak?

So what you're actually addressing is the difference between identifying the need for somebody to be internal Versus the ability to outsource. And I'm going to use the term subject matter expert here, because if you're running your company and it's just you and you're the subject matter expert, then you're the person that has to decide what you want to go do. So an example would be, I know I'm running an email campaign. It doesn't seem to be as effective as I would like. I need to ask the question how to make it more effective. Now I could hire somebody and have them analyze my data. And I'll pose the question. Now, if you get bigger and you're managing a group of 10 people or 12 people, and you now say, I don't really know what the best question to ask is anymore, but my team does. Now somebody within your team is a subject matter expert. They may also be your data scientist, and they would be able to make that decision for you. So as you go from an individual to a small team, to a larger managed team, you now have the ability to bring in those subject matter expertise to do some of the work for you.

Right. And it, with data, is it, you know, as a data scientist, do you need to be an expert, a subject matter expert in order to work with the data? Is that, you know, like in the engineering world, you don't have an electrical engineer do a do a structure because that's not their specialist. Is that the same case with data scientists? Is it still, you're wanting a specialist with that kind of background or is it more learning? You can learn and figure out. And as long as, as long as you know the goal that you're trying to reach, you're able to actually say, okay, well, here's how we can use that data to achieve that.

So what you're, what you're talking about now is the next step in the complication of that process of I stored the data. Now I need to get the knowledge that I have to somebody else so they can help solve it. So there's always going to have to be a knowledge transfer because you could hire a data scientist and they know how to solve problems, but they don't know how to solve your problem. And they see your data, they don't know what your data looks like. So when you've stored it, you essentially have to tell them, okay, this is how my data is being stored, this is what it means. And once they understand, then they can say, got it. I understand it. I can go make a model.

Right. And then let's say you're, you're using customer data. Do you, as a data scientist, need to have experience? Like, is it, is that required or is it just a set of problem solving? And once, once you understand how to use data, it doesn't matter whether it's a vaccine manufacturer trying to speed up a process or somebody making an email campaign better. It's still just data with an— with a start point and an end goal, and the processes, you know, are basically the same.

I would say in large part any problem you get is similar to another problem that you've already solved. So the vaccine question is very similar to the email question. From a data science perspective, you have to understand what the question is. what you're trying to solve, and you put in place the same scientific rigor in solving each of those problems. Now, you might use a different solution set to do it, but you're really just trying to predict something that improves your process. So what I say is most all problems look the same to a data scientist, but the level of experience Means that you approach it without as many mistakes or without as many missteps because you already know how to solve it from a prior example.

Right. So it's more just not to simplify it too far, but really it's problem solving. So it's—

It's really what it is. Yeah.

It's really problem solving with data and it's not— so if I were a startup, I wouldn't have to worry about You know, if I were going to hire you to work with my data, I would— the question wouldn't be, have you worked with email data before? The question is, how much experience have you had with data? Because it all builds. It's, it's so related that it actually all builds on each other and you don't need to, you don't need to specialize and become, oh, I'm a, I'm an email data scientist. And then there's somebody else that's a, you know, I do, I do Facebook data and I'm a Facebook data scientist.

Right.

You're able to problem solve. It's a problem-solving process and it really applies across the board.

If you're interviewing a data scientist, really what you want to talk about is features. So think of that email data. You're going to have the email, a name, a timestamp, and hopefully you can connect that to some sort of success rate. Did they contact me? Did I contact them? Did we get a contract from them? Some sort of Right, this worked or it didn't work. So first of all, you have to have the ability to connect those. That's a pretty basic thing that just about everybody can do. The next one is where features come in. So if you have a very brand new data scientist, it's like, all right, I know how to build models. They're going to build a model, and building models is incredibly easy, and they're going to run the model and say, here's the answer, the model gave me an answer.

Okay.

Which it did, but it may not be that good. Where you get more experienced data scientists coming in saying, I'm going to think about this problem and start building features. What that means is I'm going to create new data from the data I have that gives me more information about the data. So a feature might be something like, hey, this email went out in December. Well, I'm going to create a new feature that's called holiday, or I'm going to create a new feature called December or winter and start categorizing my email messages by time of day, time zone, a whole bunch of aspects that say, is this a weekday, a weekend? Am I emailing somebody that I've already contacted or not? And now you have a whole bunch of robustness to your data that actually gives you the ability to solve it. If you just took the data and ran it, your answer might be 20% accurate. But once you put the time into all of those features and you run it, it might be 80% accurate. So the level of experience, and this is where the subject matter expertise comes in, allows you to create the features that make your data really valuable.

Right, because there's a lot you can do. Like what you just said is if you have a timestamp and an email and a name, you can still extrapolate what you just said, like, well, well, at what time of day, where Is it a weekday or weekend? And so there's a lot of— there can be that data within data, right, that you could pull out, or should— you should pull out if you're going to work with the data properly. Awesome. Now, beyond that, is there another— you know, if you're going through the process and you're trying to start actually using your data, you know, you've got— you're working with a data scientist, you've given them the data.

Right.

You're giving them the model, you're giving them, you know, as much backend information as you can, the contextual information. They talk about features. So that's a good thing to know. If the data scientist talks about features, then they're probably a little bit more knowledgeable. Is there anything beyond that, but before you get into, you know, Fortune 500 level of projects that a startup should be Or small builders should be aware of or be thinking about as they try to start actually extracting value from their data?

I think of it this way, and this applies to every single company no matter size. The goal of using your data— and I think of it as you're the king of your castle and you're sitting in your throne, and at your feet are a whole bunch of levers. The goal of the data is to tell you as the owner who's running this business If I pull this lever, what does it do? If I push this lever, what does it do? What if I push it halfway, what does it do? Now if I pull these both in opposite directions, what does that do? So what, what I say for my data scientists is no matter what your model does, it has to be explainable. So you have to be able to take it back to the business to say 80% of my solution is based on The time the email was sent. Really, that's all that matters. You stop your analysis right there and you say, great, now that I know something new about my business, which times are effective? Oh, looks like when I send them after 6 PM, people aren't checking them. They then sit and they get missed and it goes down through the list. Monday mornings don't seem to work either because people are swamped with all of their emails. But it really seems like Tuesday, Wednesday, Thursday at about 1 o'clock seems like it's the best time. That's an explanation of your business, one of these levers. So your main goal when you hire anybody to look at your data is just to explain how your business runs. And I hope that's not insulting to people that built a business.

Right.

But, but that's again, the case for any company I've ever worked for is they They fundamentally have access to information. They need to process it to understand how to make better decisions in the future.

Right. Well, it's, it's very true though. I mean, you can't, you can't watch everything all the time and solve everything. So that's, and I think that's where, you know, when I talk to startups, like save your data is one of the biggest things I say, and I'm not even a data person. I just know it's useful later. And, you know, talking to you and other, other people that work in data have those stories of, oh, well, I can't help you. You didn't save anything. There's nothing there. And, and working with, with small businesses and large ones, you know, they, they don't know what they don't know. And that's, that's one of the keys. And then you figure out once, once they figure out what they don't know, then it takes someone looking at all that data and going, okay, well, actually we have it. We do know it. Just no one ever looked in the right place with the right, with the right lens to see that we actually do have, have everything right here. Just needed a little processing to get there.

So I want to bring the barrier to entry down a bit with a very practical example from my own real life, just to show how valuable this is and how easily it can be implemented at a very low cost. And again, sort of with the presumption of once you know, Right. So personal story, I wanted to buy a car and I didn't want to haggle. I didn't want to try to figure any of this out. And I basically said, I don't really care what kind of car I buy. I want to buy a car that has these features. So I went and I downloaded a list of every car I could find, put it into a database. And then I built a machine learning model that said, go find these vehicles with these qualities within this price range. And predict for me which car I should go buy. And I did all this on a free basis, so you can get credits when you sign up to try things out. I think this entire process cost me $12, but ultimately what I did is I went in with my phone. I had productionized an API that I could call from my phone. I walked onto the dealer's lot. And I typed in that car and the machine learning model went through all of the cars. It spit out an estimate. And I said, hey, I'm not here to haggle. I'd buy the car for this amount. Knew nothing about it. And the guy is like, yeah, no. So I was like, okay, no problem. I'll find somebody else that'll sell for that price. And I walked off the lot and he called me 5 minutes later and said, if you buy it right now, like, are you seriously going to buy it? I was like, yeah. He's like, okay, I just didn't figure you were serious because you were on the lot for 3 minutes and you were gonna buy a car. That doesn't make sense. But it gives you the confidence that your data is leading you to the right choice. So I got a car that fit all of my needs for a very fair price, and I didn't spend any time doing it. And it, it took me an hour to collect the data, build the machine learning model and less than an hour to buy the car. And that entire process cost me $12 and I probably saved $2,000 or $3,000 in the process.

Well, now I know how a data scientist buys a car. That's awesome. I will have to think about that the next time I look for a car and see what it, see what it spits out. So $12 and it saved you thousands, and that's just a car purchase. So extrapolate that to any business, you obviously are going to multiply that by 10 plus easily. Um, and yeah, that makes it so easy. Now, if— how much coding was involved?

Like, That's actually a good thing to bring up. I feel like there's probably 3 paradigms we should probably discuss. So the first one is considered maybe traditional code, which when you would hire a data scientist, they're going to come in and write raw Python code for you, something that most people would look at and go, oh gosh, I just hope it's right.

Right.

And have no idea how to use it. It's going to be in an environment where they might not even have access to it, and you just say, Thanks. Right, like if they say, hey, set your email to go out between, you know, 1 and 2 o'clock on Tuesday, Thursday, Friday, you're like, okay, I'll do it. And then the code disappears and the process is done. So that's, that's kind of the most standard if you're going to outsource it. Most every data scientist that comes out of school is going to know Python. There are newer versions that are a lot easier to use, and you can download a lot of pre-existing, pre-set-up things that just about everybody could try.

Okay.

And those are similar but a little bit more advanced. Probably won't get into that. But then there's really simple solutions, and that CAR version is what I did with this. So I think all the platforms have them. In this case, I used Azure's machine learning platform, which is just called Azure ML. And it's like drag and drop. And you can go on to their like app store and say, I want to find a car buying app. And it'll give you 4 versions and you literally click load and the entire machine learning model is built for you. You just connect it to your data and you might have to configure a couple of things, but you essentially say connect to my data, run. And then you just check it out.

So not any harder than uploading an email list to MailChimp. If you're running a couple of things and tell it where to, which data is which, and then it does its thing.

Exactly. Now, is it the best model? Absolutely not. Right. But is it good enough? And is it good enough for an hour's worth of time? So the main question I would ask is how much money do you have to spend and what are you looking for? You could on your own go into your Azure subscription and click on the model and see it run and see the other versions and pretty quickly piece together what it's doing. If you're going to hire somebody and they're writing Python code, you won't really understand what it is.

Right. Well, and you could always, you could always do that cheap version. Off of a pre-made one and then compare it with your data scientist and see, see what, what the difference is too. So that could always be an interesting comparison.

Well, in all honesty, the way I oftentimes develop will be to do the super fast and dirty first, just to get an idea of what it looks like.

Just to baseline it.

And then go into the code and build one that's a lot more robust. and refinable that you have more control over. You can create a lot more of those features on the fly. And then you can say, hey, the quick and dirty that I spent 20 minutes on gave me a 50% accuracy, but this version is 90% accurate. Okay, it was worth the money. Or you come up with a solution that's 50% accurate and you're like, oh, well, that's no better.

Right. Awesome. Perfect. Well, I think this has been really eye-opening, and hopefully lots of startups are listening to this and saving their data so that they can come to someone like you and say, hey, I got data. I actually saved it for the last 2 years. And now, now we need some help. And you can actually do something with that. For people who want to get ahold of you and You know, either pick your brain or, or work with you, what's the best way to get ahold of or find you?

I would probably start with LinkedIn. You can get ahold of me on my LinkedIn. If you look up Scott Davis, I work for a company called Valorem Reply based out of the West Coast in Seattle. And I would be more than happy just to Have a conversation, discuss where you're going, and see where things go from there.

Amazing. Well, I really appreciate your time, Scott. It's been good having you here and talking about this, and looking forward to hopefully have people reaching out to you and learning, learning from this so that they can, they cannot be the person that comes to you with no data, and we can avoid that. In the future.

It'd be fantastic. And I'll end this on a very practical note. There's really 3 places. They're all almost identical. You can go to Google, you can go to AWS, you can go to Azure, go to their cloud storage. And again, you can upload your data. The storage is almost free. I use all of them. Honestly, they're all nearly identical. So pick one, store your data, and we can go from there.

All right, well, you all, you all heard from a data scientist. They're all the same. Just, just pick one. Pick your favorite color.

Salesmen will be cringing right now, but they all work well.

They'll work well. So pick your color, save your data, and be happy in a couple of years when you need it. Awesome. Thanks, Scott.

Thank you.