113: Operations Research, Prescriptive Analytics, & Decision Science w/ Adam De Jans & Steven Stark

[00:00:00] Data jobs tend to pay higher than a lot of other jobs.

[00:00:02] You get to use cool technologies.

[00:00:03] At the end, in most cases, like it's worth it.

[00:00:06] Turns out having a strong network really plays a huge role in your

[00:00:10] ability to get new jobs, change roles.

[00:00:15] That just plays a huge impact on your career.

[00:00:17] Welcome to the Data Career Podcast, the podcast that helps aspiring data

[00:00:21] professionals land their next data job.

[00:00:24] Here's your host, Avery Smith.

[00:00:27] Welcome back to another episode of the Data Career Podcast.

[00:00:30] I'm joined today by a panel here.

[00:00:33] We're going to be talking about some very cool things about data

[00:00:36] tiles and operations research, which is a very interesting field

[00:00:41] in data science and analytics.

[00:00:42] We have Steven Stark and Adam Dijans with us.

[00:00:45] Guys, welcome to the Data Career Podcast.

[00:00:47] Thank you.

[00:00:48] Really excited to be here.

[00:00:51] Yeah, for sure.

[00:00:52] Steven, do you want to introduce yourself first and then we'll have Adam go next?

[00:00:55] Just mention what you currently do and maybe a little bit about your background.

[00:01:00] Sure.

[00:01:00] So hello everyone.

[00:01:01] My name is Steven Stark and I'm a senior data scientist with several years of

[00:01:04] experience currently working at American Tire Distributors.

[00:01:07] So currently I work on a few major projects involving operations research in

[00:01:12] terms of inventory decisions at distribution centers and my background's an

[00:01:16] industrial operations engineer.

[00:01:17] I have a master's degree in the field from the University of Michigan and I

[00:01:20] just really enjoy using mathematical programming to solve this problem.

[00:01:24] Perfect.

[00:01:24] I like it.

[00:01:25] How about you, Adam?

[00:01:25] So I'm in a decision science lead at Toyota North America and I work in the

[00:01:31] supply chain working in the digital transformation group.

[00:01:35] So we are trying to transform everything that we do within supply chain.

[00:01:40] So a lot of forecasting and optimization all in order to like everything from how

[00:01:47] do you get the resources you need to build the car, to building the car, what

[00:01:52] order you do it in, how do you schedule that?

[00:01:54] And then actually shipping that out to the customer as well.

[00:02:00] Very cool.

[00:02:00] We're going to talk about, you know, a little bit more about your

[00:02:03] guys' jobs here in a second.

[00:02:05] One of the things as I was kind of prepping for this interview that I

[00:02:09] noticed and I'll pull up your guys' LinkedIn here, the amount of titles

[00:02:13] that you guys have both held and that I've held is kind of insane.

[00:02:17] Basically like there's analytical modeler, research assistant, business

[00:02:21] analytics, engineer, statistical analyst, operations research scientist,

[00:02:27] senior data scientist.

[00:02:28] Adam, you have the let's see the AI consultant, which that seems crazy.

[00:02:33] Decision scientist.

[00:02:34] You have a data scientist, operations research scientist, software

[00:02:38] engineer, data analyst, dating mining research assistant and software developer.

[00:02:43] I have data scientist, optimization engineer,

[00:02:47] chemometrician and data analyst.

[00:02:49] And like, that's just like three of us.

[00:02:51] And it's like, we're not necessarily new to the fields, but we're not necessarily

[00:02:54] like I've been in the field for millions of years.

[00:02:56] And if you think about like a dentist holds one title their whole career,

[00:03:00] like they are a dentist, but just between like the three of us,

[00:03:03] there's like 14 different titles.

[00:03:05] So I wanted to just quickly, I'll go to Adam first, have your

[00:03:08] thoughts on like data titles.

[00:03:10] Like what do you think about them?

[00:03:11] And like you mentioned before we got on the call that you've had more than that.

[00:03:15] You just, those are the ones you just put on your LinkedIn.

[00:03:17] Yeah.

[00:03:18] So the titles are kind of, I'll call them mostly meaningless because every job you

[00:03:24] go to has a different title and you could be doing pretty much the same thing.

[00:03:29] Job to job might be a different title.

[00:03:31] It might be the same title.

[00:03:33] Nothing really aligns.

[00:03:35] A lot of this has to do with a lot of the companies are still trying to figure out

[00:03:39] what it means to be a tech company or a data company where they'll hire a data

[00:03:44] scientist, but you know, they're not ready to do data science.

[00:03:47] So you're doing data engineering, or maybe you're just doing some analytics piece.

[00:03:52] As for myself, I kind of swapped around from software development.

[00:03:57] So I was in like more of a traditional software role and then that progressed

[00:04:02] through more operations research mixed with data science.

[00:04:08] And then, like you said, I did that one year of consulting.

[00:04:11] That was a wild time.

[00:04:14] Consulting is a very interesting field.

[00:04:17] And then back to decision science, even at Toyota, I've had multiple internal

[00:04:23] titles that have changed things from like analytics consultant, which is weird

[00:04:28] to have it as an internal title.

[00:04:30] Principal, like principal engineer.

[00:04:34] I've had data scientists, decision scientists, a mix.

[00:04:38] So I just picked the most appropriate from what I think would fit in theory to match.

[00:04:44] So I don't have all these different titles.

[00:04:46] I like it.

[00:04:47] You choose the one that fits, but maybe it's the one that's the sounds, the sexiest.

[00:04:50] That's what you're going for.

[00:04:51] That's true too.

[00:04:52] That's true too.

[00:04:53] Which one will get me, what one helps you stand out as well?

[00:04:57] I love it.

[00:04:58] How about you, Steven?

[00:04:58] Do you kind of agree that data titles are meaningless?

[00:05:01] So I wouldn't say they're meaningless.

[00:05:03] I think, I mean, Adam made a bunch of good points and I want to emphasize that.

[00:05:06] Yeah.

[00:05:06] When it comes to a job, whether it's in data or not, I mean, I think it's a lot

[00:05:10] more important that you enjoy the work and you feel like your comments say it

[00:05:13] well, and you know, it's a good work environment and center over like the title.

[00:05:16] The thing though is that like, I think titles have values in terms of when you're

[00:05:21] actually trying to change jobs in terms of perception, because let's just say

[00:05:24] recruiters on LinkedIn or hiring manager on LinkedIn, they're just

[00:05:27] like scrolling through people.

[00:05:28] Now I don't know everything they see, but I feel like the titles

[00:05:31] themselves will stand out.

[00:05:33] So someone might just see a tummy like, Oh, this person's a lead data

[00:05:36] scientist at company A while this other person's only a senior

[00:05:38] data scientist at company B.

[00:05:39] So like, well, person A might, must be buyer.

[00:05:41] That's not necessarily the case, but we know when the recruiter doesn't have

[00:05:45] time to sit there and read everybody's description and profile and resume

[00:05:48] and all that in depth, they might just jump to that conclusion.

[00:05:51] So I think that's probably like the most importance comes within titles.

[00:05:55] But then again, I think other components are more important, like the kind of

[00:05:58] work that you're doing and how you feel about your job and the compensation

[00:06:01] and whatnot, and going back to that too is like my current job that I'm in now,

[00:06:06] like I have the title of senior data scientist, but when I was applying to

[00:06:09] jobs, I had two offers, like the one that I'm currently in now.

[00:06:12] And another one that was called like staff data scientist.

[00:06:15] So even though most people would say, Oh, well, staff sounds better than senior.

[00:06:19] Well, I mean the compensation and like the work and whatnot, I felt like was

[00:06:23] better at my current job than the staff role that I turned down.

[00:06:27] So, I mean, yeah, there's definitely more to life than titles.

[00:06:29] You also bring a good point that like, even like the, I don't know what

[00:06:33] the correct term for this is.

[00:06:34] I'm going to call it the prefix, like the lead, the staff, the junior.

[00:06:39] Even that stuff is a bunch of bogus.

[00:06:41] I've had a company that we've actually put a lot of our students into out of our

[00:06:45] bootcamp that they've hired our students and they hire all of our new students who

[00:06:50] are brand new to the field, a lot of like transitioning teachers, some people didn't

[00:06:53] even have college degrees and they all become a senior analytics associate.

[00:06:57] I don't know what an analytics associate is and what makes you a senior one, but

[00:07:01] like even the prefix on things can be kind of wild and silly.

[00:07:06] I want to go back to what Adam said earlier, where like, you know, you try to

[00:07:10] choose the title when you put on your resume or on your LinkedIn that

[00:07:14] best represents what you did.

[00:07:16] Do you think it'd be okay for someone, like, do you think you have to stick to

[00:07:20] the official title that was given you?

[00:07:22] Or if you're given kind of a wacky title to kind of maybe more

[00:07:25] generalize it or dumb it down?

[00:07:28] I think it's okay to switch it.

[00:07:30] I like to switch it to fit what I actually do.

[00:07:33] So if it aligns closer to what I do, then that's what I'll choose.

[00:07:38] I think like right now, currently my title is like principal engineer, something,

[00:07:44] something.

[00:07:45] I don't really do engineering at all.

[00:07:47] Software engineering maybe, but if you just see principal engineer,

[00:07:51] what does that mean?

[00:07:53] So it's like one, I want to be more descriptive to let recruiters or other

[00:07:58] people know what I'm doing or what's closest to what I'm doing.

[00:08:02] So I think there's some wiggle room there.

[00:08:05] Now you can't totally, I'm not saying fabricate what you do, but if you even

[00:08:11] get a better, a different title that aligns more to what you're doing and

[00:08:14] it's not a big stretch, then I think that's acceptable, especially on LinkedIn.

[00:08:20] I agree.

[00:08:21] Like when I was a chemometrician, no one even knows what that is.

[00:08:25] And then I'm like, oh, I'm a data scientist who does stuff with chemistry.

[00:08:28] Oh, okay.

[00:08:29] That makes more sense.

[00:08:31] Steven, I was curious though, out of all these different jobs that you've had,

[00:08:34] is there one that you like more than the other?

[00:08:36] Like for instance, do you like being a data scientist more than doing like an

[00:08:40] operations scientist role or is it kind of all just been enjoyable one way or the

[00:08:44] other?

[00:08:44] So I guess this is going back to what Adam was saying before where it's like,

[00:08:47] you can have like a different title but still do the same kind of work.

[00:08:51] Because like, I mean, yeah, in general, in terms of like the kind of work I'd

[00:08:54] rather do like mathematical optimization, like prescriptive analytics over

[00:09:00] predictive analytics, like machine learning, trying to predict something.

[00:09:03] But I mean, at the end of the day, like all the roles, like it's more so focused

[00:09:07] on like, okay, here's what the business has, you know, this is what the business

[00:09:10] needs you to do now.

[00:09:11] And like, you know, it's like you don't always necessarily get the pick and

[00:09:15] choose like, oh, I get to work on this type of problem or that type of problem.

[00:09:17] It's like, well, here's what priorities like you need to do them.

[00:09:19] So like, you can't just be like a, I don't want to say one trick pony, but there's

[00:09:23] like a variety of skill sets you need where, especially whether you're like data

[00:09:26] scientists or like operations research scientists, you're going to be working

[00:09:29] with databases, so like you're going to need to know SQL.

[00:09:31] And then also to like, you know, being able to like work with business

[00:09:33] stakeholders, like ask clarifying questions, you know, being in the loop and

[00:09:37] having good communication sales, like those are all going to be relevant.

[00:09:40] So, but then like, I guess like in terms of the, when you're asking like the jobs,

[00:09:45] like I've been fortunate enough that like all my jobs have been like pretty good

[00:09:48] experiences for the most part.

[00:09:50] I, one thing I will acknowledge that in terms of like the product slash service

[00:09:53] itself is Adam at my previous company, when I worked at NBC Sports Next, which

[00:09:57] is a part of NBC Universal, I just thought that was, I mean, I thought it

[00:10:00] was really cool, like everything that's behind the NBC brand, like, you know,

[00:10:03] like the parks and the movies and TV shows and whatnot.

[00:10:06] So in terms of like coolest brand, I'd say that one so far, not that the other

[00:10:10] brands I've worked for were bad or anything.

[00:10:11] Like I just thought the entertainment industry is pretty interesting.

[00:10:15] I would agree with that.

[00:10:16] And I'm, I'm a basketball fan and the NBA just bought back the rights to the

[00:10:19] NBA or sorry, NBC just bought back the rights to the NBA.

[00:10:24] So I think that would be an exciting time as well.

[00:10:27] Now I pride myself, one of the things I try to do on this podcast is there's

[00:10:32] quite a few data podcasts out there.

[00:10:34] I always try to make this podcast as beginner friendly as I can.

[00:10:39] And we've said a lot of fancy terms that I would love your guys' help in defining

[00:10:43] and talking about because I don't think a lot of people know them necessarily.

[00:10:48] So maybe I'll start with you, Adam.

[00:10:50] Can we've said the phrase operations research a lot.

[00:10:53] Can you like give like an explain, like I'm five definition

[00:10:57] of what operations research is?

[00:10:59] Sure.

[00:10:59] So when I think of operations research, I think of there's some decision that

[00:11:05] needs to be made and you're trying to figure out what is the best decision to

[00:11:11] be made so it's easier to compare it to something like data science or machine

[00:11:15] learning that people are used to.

[00:11:17] So with machine learning, you're usually doing something

[00:11:20] that's more predictive analytics.

[00:11:23] So you're building models around, you know, trying to predict something,

[00:11:28] maybe trying to forecast something.

[00:11:30] How many sales will I have?

[00:11:32] How do I recommend like maybe I'm building a recommendation engine or I'm trying

[00:11:37] to see like what story does the data tell, but you're not focused

[00:11:42] on an actual final decision.

[00:11:44] So I think that's one of the key differences in operations research.

[00:11:48] You'll hear a mathematical optimization.

[00:11:51] And if you're unfamiliar with that, it's there's some objective that

[00:11:56] you're trying to maximize or minimize.

[00:12:00] There's some decision variables, like what can you change

[00:12:03] in the course of this problem?

[00:12:05] And then there's some constraints on it as well.

[00:12:08] So for example, if you're staffing people, maybe nobody

[00:12:12] can work more than 40 hours.

[00:12:16] So I think the key focus there in operations research is there's

[00:12:19] something you're trying to decide.

[00:12:22] So you'll use things like I said, mathematical optimization or

[00:12:26] mathematical programming simulations.

[00:12:29] And also you focus on some of the psychology that goes

[00:12:34] on with your stakeholders.

[00:12:37] So those are really the big differences between what is traditionally

[00:12:43] considered data science versus operations research.

[00:12:47] I like it.

[00:12:48] Steven, anything to add to that?

[00:12:49] I guess if I had to describe it in one sentence, which might sound

[00:12:55] like pretty nebulous, I'd say it's the use of math technology and logical

[00:13:00] reasoning to solve business problems intelligently.

[00:13:03] I like it.

[00:13:03] That's a nice sentence right there.

[00:13:05] Steven's very passionate about it.

[00:13:07] By the way, we'll have Steven and Adam's LinkedIn URLs in the show notes down

[00:13:11] below and you guys can go follow them on LinkedIn, they create a lot of content.

[00:13:14] Steven is very passionate.

[00:13:16] In fact, he made a LinkedIn post that got a lot of traction here and I want to read it.

[00:13:20] He said, unfortunately, the most interesting part of data science

[00:13:23] doesn't get the coverage it deserves.

[00:13:25] I don't mean machine learning, although machine learning does depend on this thing.

[00:13:29] I don't mean SQL either.

[00:13:30] SQL gets more coverage than it needs.

[00:13:31] Yes, SQL is important, but it's not as exciting as people make it out to be.

[00:13:35] What I'm talking about is mathematical optimization.

[00:13:38] And if you guys just heard that, we heard, I heard, I don't

[00:13:40] know about you, Steven and Adam.

[00:13:41] I just heard the whole audience yawn as they heard that.

[00:13:44] That's okay.

[00:13:45] He says, I believe the reason why people don't get enough coverage is simply

[00:13:48] because not enough people know about it.

[00:13:49] Mathematical optimization is typically used or is typically covered, sorry, in

[00:13:54] the following undergraduate slash graduate programs, industrial engineering,

[00:13:58] operations research, mathematics, and computer science.

[00:14:00] And it may be touched on the following subjects as chemical

[00:14:03] engineering, economics, and business.

[00:14:05] And here are some use cases.

[00:14:07] So I wanted to go through some of these use cases one by one and talk about

[00:14:10] what the examples are.

[00:14:11] So one of the ones, actually, we should probably talk about the ones

[00:14:14] that you guys can talk about more.

[00:14:16] So one of the easiest ones to talk about is like, is like inventory.

[00:14:21] You know, how much inventory should a company hold?

[00:14:24] Like let's say we're selling, I guess the easiest one that I always talk

[00:14:28] about is like ice creams.

[00:14:31] If you're like an ice cream vendor in like New York city out on the street,

[00:14:34] how many little like little like cups of ice cream should you have?

[00:14:38] Because if you don't have enough, right.

[00:14:40] If you don't have enough ice creams there, you're going to potentially

[00:14:43] miss out on business because if you sell out, Oh, well, crap.

[00:14:45] So, you know, you sell out maybe at like noon, you're going to miss, you

[00:14:49] know, noon to 5 PM is business and you're not going to have enough money.

[00:14:53] Let's say you come back the next day and you packed a lot of ice creams.

[00:14:56] Well, if you have a lot of ice creams and you don't sell them all, let's say

[00:14:58] it's 7 PM, you're going home for the day.

[00:15:01] And you have 400 ice creams leftover.

[00:15:03] Those ice creams might melt.

[00:15:05] They might expire.

[00:15:05] So you might lose money there.

[00:15:07] So basically one of the things that operations research does is take a look

[00:15:10] at this problem and basically tell you what's the optimal amount of ice creams

[00:15:15] to have at your little cart that day.

[00:15:18] Adam, is that the type of problems you've solved in the past?

[00:15:20] Does that kind of make sense?

[00:15:21] Can you maybe relate to us what you've worked on in operations research?

[00:15:24] Yeah, I've worked on inventory optimization like that.

[00:15:27] And one thing I want to point out is nowadays it's not just

[00:15:31] the optimization by itself.

[00:15:33] A lot of the predictive analytics feed into optimization.

[00:15:38] So for example, you want to find the optimal inventory, but how

[00:15:43] do you come to that conclusion?

[00:15:44] You need that predictive analytics piece to forecast, you know, how many ice

[00:15:50] creams you think you're going to sell maybe, and then there's some statistics

[00:15:54] probability thrown in there, but another really good example that is really

[00:15:58] easy for anybody to understand that we've both, I've worked with Stephen

[00:16:03] on this as well is the vehicle routing problems.

[00:16:06] So they're very intuitive.

[00:16:07] You just want, you have a bunch of people, you need to pick them

[00:16:10] up and drop them off somewhere.

[00:16:12] Every car can only hold so many people.

[00:16:14] Like what's the best route to go to do this?

[00:16:17] And when we say best, what do we mean?

[00:16:20] We could mean the number of, we want to minimize number of miles driven.

[00:16:24] So when I worked in autonomous vehicles, that was really important

[00:16:27] because the more miles you drive, the more wear it has on the car and

[00:16:31] autonomous vehicles, you do not want to have to repair them.

[00:16:35] Like they're very costly.

[00:16:37] Or if you're looking for a really good customer service, maybe you're

[00:16:40] trying to minimize the time it takes from them to put in their request to

[00:16:44] be picked up to get where they're going.

[00:16:46] Or maybe you want to minimize the costs for the customer.

[00:16:49] So then you can factor in something like ride sharing, but

[00:16:53] that's a very easy example, right?

[00:16:55] You just picking people up and dropping them off, but coming up with that optimal

[00:16:59] sequence of pickups and drop-offs turns out to be quite challenging.

[00:17:05] And maybe people have heard of the traveling salesman problem.

[00:17:08] It's very similar, but it's like traveling salesmen on steroids, right?

[00:17:12] Because each vehicle is doing like its own, almost like it's solving its

[00:17:16] own little traveling salesman problem.

[00:17:19] Not exactly, but it's quite similar.

[00:17:22] That's a great example.

[00:17:23] You know, for instance, like Uber, how does Uber decide where

[00:17:26] like people get picked up?

[00:17:28] Or I guess they don't decide where people get picked up and dropped off, but like

[00:17:30] how are the cars allocated in that situation?

[00:17:33] And like you said, with autonomous driving, kind of the same deal.

[00:17:35] Thank you for that example.

[00:17:37] I was going to say there's a huge change management factor in there as well.

[00:17:41] Cause like a lot of people think, oh, just go to the closest vehicle.

[00:17:45] Right.

[00:17:45] But that's what your intuition thinks, like pick up the closest person,

[00:17:50] but that's not always the case.

[00:17:51] So you need to also work with stakeholders on that.

[00:17:54] And that's more of where like the operations research, when I said the

[00:17:58] psychology piece also plays a role.

[00:18:01] Cause you need to be able to explain why am I not picking, like somebody

[00:18:05] put in a request 20 feet away.

[00:18:07] There's a car right there.

[00:18:08] Why didn't he send that car?

[00:18:09] Why'd you send this other car?

[00:18:11] That's a big one because I guess one of the differences a lot of the

[00:18:14] times, not always, but with operations research is, and I think Stephen's

[00:18:19] definition really captured this well, is you're trying to make informed

[00:18:22] business decisions, but at the end of the day, the humans still

[00:18:26] making a lot of the decisions.

[00:18:27] When I worked at ExxonMobil, the operations research problem I worked

[00:18:31] with is what crude oil do we buy from different parts of the world?

[00:18:34] We're trying to figure out if we're going to make all these different

[00:18:36] products with the current prices of all these different crude oils and

[00:18:39] all their chemical properties, you know, what allocation of crude oil

[00:18:42] should we buy to maximize profit, minimize emissions, stuff like that.

[00:18:47] And at the end of the day, the computers, it's not like an algorithm

[00:18:50] that's automatically making the trades for, you know, the calls

[00:18:54] for purchasing the crude oil.

[00:18:56] It's still a human being.

[00:18:57] And a lot of times the human being that was doing this had

[00:18:59] been doing it for 20 years.

[00:19:01] And so you have to deal with, you know, well, my gut's telling me this.

[00:19:05] Why is the computer telling me that?

[00:19:06] And so that's a lot of operations research, like you talked about

[00:19:09] is like being able to have some sort of explainability and like, this is why

[00:19:13] the computer thinks this is a good idea.

[00:19:15] Do you approve as a human?

[00:19:17] So that definitely is a part of it is like explaining.

[00:19:19] Steven, I wanted to go to you and just hear like what you've kind of

[00:19:22] worked on in operations research.

[00:19:24] Sure.

[00:19:25] So I'll use an example similar to what you were just talking about

[00:19:29] where the explainability part.

[00:19:31] So when I was at Walmart, I worked with a team and we had a problem of as new

[00:19:37] items would come into a warehouse from, you know, Walmart suppliers, we have

[00:19:40] to find shelving for them in the distribution center and the idea is that,

[00:19:45] you know, you know, different there's different shelving and the different

[00:19:47] shelving has different sizes and there's certain constraints, like really heavy

[00:19:50] items can't be on the top shelf because, you know, someone could get hurt

[00:19:53] trying to take them off the top shelf.

[00:19:55] You can't have very similar looking items next to each other because if you do,

[00:19:58] the person that takes the item off the shelf and puts it in the box, some of

[00:20:01] the sort of, they might accidentally take the wrong item.

[00:20:03] So how do we place the items such that we avoid those issues?

[00:20:06] And then also too, where it's like, you might have to take one item out of

[00:20:10] its current spot and then move it to a different place and it's like, that's

[00:20:13] going to, you know, results in a lot of walking and, you know, work hours.

[00:20:16] Like, do we want to do that?

[00:20:17] We only want to do if it's necessary.

[00:20:18] And with that, the, like the press that we were taking over was that like one

[00:20:23] person would just look at a report and make, okay, here are new items that

[00:20:25] would come in and this manager, she'd just like manually walk to each, you

[00:20:29] know, open shelf and then it's like, you know, get some measurements

[00:20:32] and try to put some in there.

[00:20:33] It's like, this process is like really hard because, you know, a human being

[00:20:37] can't process all this information at once.

[00:20:39] They can't keep track of like all these hundreds of new items that are coming

[00:20:42] in, so it'd need to, you know, not optimal solutions, but there's like a

[00:20:45] psychology as part as I'm brought mentioned before, because we had to,

[00:20:50] you know, explain to this person where it's like, okay, like this technique

[00:20:53] that we're using, it's not going to take your job away, like you're

[00:20:55] still going to have your role.

[00:20:56] It's just, it's going to help streamline your role so that you can focus on

[00:20:59] like more important problems at the distribution side.

[00:21:01] Cause like once we tackled it, once we all made this problem, there's other

[00:21:04] problems that we currently can't handle yet, but like you'll be able to handle

[00:21:07] those instead.

[00:21:07] The big thing too is that like explain to people that there might be a decision

[00:21:12] It looks like it's not good for this one, like small portion, but it's

[00:21:16] actually best for the whole system.

[00:21:18] I think that's like what's very challenging to describe.

[00:21:20] Cause like some people might just think like, oh, well the absolute best.

[00:21:24] And when I say like for the whole system, it's like, let's just say for instance,

[00:21:27] when it comes like the packing, you know, you're putting items on a shelf, you

[00:21:30] might have like one item that's like you put on a shelf, but there's like a lot of

[00:21:33] room left over and you might be like, well, that item should go on a better

[00:21:36] fitting shelf.

[00:21:37] Well, it turns out that by putting that item there, all the other shelving

[00:21:41] is like more efficient.

[00:21:42] It's like as a whole, it's better than, you know, trying to get that one

[00:21:45] particular item.

[00:21:45] It's like best fitting shelf.

[00:21:47] Yeah.

[00:21:48] It's, it's making tough decisions basically.

[00:21:51] It's making these decisions across, you know, thousands or maybe even

[00:21:54] millions of other decisions.

[00:21:56] I can't believe we're like 20 minutes into this and I know Steven mentioned it

[00:22:00] earlier, but quickly there's four different types of analytics.

[00:22:04] Descriptive is what happened in the past.

[00:22:06] So like for instance, if we take Steven's example, well, maybe Steven's example's

[00:22:09] hard, but like, well, like what did customers buy?

[00:22:11] For example, what's popular at the store that's descriptive analytics.

[00:22:15] And then there is a diagnostic, which is like talking about like maybe why

[00:22:18] customers bought those things like or how then there's predictive of what will

[00:22:22] they buy in the future.

[00:22:24] And then there's prescriptive, which is like how to get stuff done.

[00:22:27] Like where should we put this stuff inside of the store?

[00:22:31] So that's something that's noticed and that prescriptive analytics is really a

[00:22:34] big part of operations research.

[00:22:37] That's kind of where operations research kind of lives.

[00:22:40] Now these jobs that you do inside of operations research, they're great jobs.

[00:22:44] You're solving interesting problems.

[00:22:46] Like for instance, what oil should we buy?

[00:22:47] Where should we put stuff on the shelf?

[00:22:48] What parts do we need at our manufacturing company?

[00:22:52] Where do we source them?

[00:22:52] Stuff like that.

[00:22:53] But one of the things I asked Steven and Adam, you know, that we try to be

[00:22:56] user-friendly, beginner-friendly at the podcast is like, could someone who's

[00:23:01] transitioning from like a completely different career has a non-technical

[00:23:05] background, could they potentially land in operations research as their first data job?

[00:23:11] And so we kind of like had a quick conversation about this earlier, but Adam,

[00:23:14] I'd love to kind of hear your thoughts of like, let's say there's a high school

[00:23:17] math teacher who wants to get out of the classroom.

[00:23:19] Could they potentially land in operations research as their first data job?

[00:23:23] Let's say that it is a math teacher.

[00:23:25] They have a math background.

[00:23:26] They have the foundation that's necessary to do this role.

[00:23:32] They're probably missing a few operations research courses.

[00:23:35] Let's just say they go and study that on their own.

[00:23:38] They're still going to face the issue of having that piece of paper.

[00:23:43] So particularly with operations research, there's a huge emphasis on having a PhD.

[00:23:51] Even just having a master's is barely enough a lot of the times.

[00:23:55] I don't think I've seen really any role outside of government roles where only

[00:23:59] a bachelor's is required.

[00:24:02] Usually there's some type of graduate degree with a very strong

[00:24:05] preference toward doctorate degree.

[00:24:07] So there's like the, could they do it?

[00:24:11] Do they have the skill to do it versus the practical piece

[00:24:13] of getting past the HR gate?

[00:24:15] Steven, thoughts?

[00:24:16] I think it was like 12 people total.

[00:24:19] There were three masters, me and Ambien, two of them, and then another,

[00:24:22] a good friend of ours, and then like nine PhDs that my other teams have

[00:24:26] joined at least master's and above.

[00:24:28] So yeah, I mean, there is a big emphasis on graduate degrees now.

[00:24:32] If it was like a high school math teacher, if they already had

[00:24:35] like a master's degree in math, I think that would like, I mean, that

[00:24:39] probably satisfied the degree requirement, but then again, it's just

[00:24:42] brushing up on certain topics that they might not have studied or it

[00:24:44] might not be as fresh in their head.

[00:24:46] That team that he's referring to, like almost all of them went to really

[00:24:50] big well-known universities as well.

[00:24:53] So it wasn't like some random college.

[00:24:56] I wouldn't want to like have the graduate degree be like a turn off

[00:24:59] because, and it's not easy, but like there are places where they have

[00:25:03] like one-year masters or ways to make it more affordable.

[00:25:06] So like for instance, a lot of my friends that did master's with me at

[00:25:10] the University of Michigan, they had multiple semesters of being like a

[00:25:13] research assistant or a graduate student instructor, and in those cases,

[00:25:17] like in exchange for working, it's like a graduate student instructor.

[00:25:20] People probably know it's like, it's like the graduate students that

[00:25:22] like helps out the professor in teaching the course, they grade papers,

[00:25:24] things like that, they can have their tuition waived and also get paid on top of that.

[00:25:28] So those are opportunities too.

[00:25:29] Now again, like I understand too that like you might have a family,

[00:25:32] you might have other commitments and you might have a, you know, but

[00:25:35] you know, other commitments.

[00:25:36] So it's like, it might be hard, but like there's different avenues and like,

[00:25:39] because like there are sometimes I'm like, I just feel like things like,

[00:25:41] Oh, graduate degree.

[00:25:42] No, I can't do it.

[00:25:43] I feel like it's more attainable than some people make it out to be.

[00:25:46] Like, it's not as, I don't think the work is necessarily as bad as

[00:25:48] somebody who made it out to be.

[00:25:49] Like a PhD, like that might be a lot, but even like a one year or like

[00:25:52] one and a half year masters, I think are for the most part,

[00:25:55] like pretty doable and flexible.

[00:25:56] And even like allowing you to take it online or, you know, do half

[00:26:00] part-time or something.

[00:26:01] So like you can still, you know, have time for your job and like

[00:26:04] other commitments and whatnot.

[00:26:05] I agree with you.

[00:26:06] I mean, my master's was two years part-time online.

[00:26:09] It wasn't like the hardest thing on planet earth.

[00:26:13] And you guys are totally right that like a lot of operations research,

[00:26:16] they're pretty stingy with needing at least a master's degree.

[00:26:19] I will say that if we go back to some of the programs that Stephen

[00:26:23] mentioned in his LinkedIn post, for example, if you studied industrial

[00:26:26] engineering, if you studied operations research, I don't see that degree too

[00:26:30] often.

[00:26:31] If you studied math, computer science, chemical engineering, economics, maybe

[00:26:35] like a more technical business degree, there's a chance that you could

[00:26:39] potentially as an undergrad.

[00:26:40] So for example, after my undergrad, I landed my job at ExxonMobil in

[00:26:44] operations research, and I only had an undergrad at the time.

[00:26:47] I ended up getting my master's while I was there.

[00:26:49] I mean, I landed a job, an operations research job without the master's

[00:26:53] degree, but I wouldn't have been able to technically climb the corporate

[00:26:57] ladder as high in that team because I didn't have like a master's or a PhD.

[00:27:02] If you already have a master's, then you're set.

[00:27:04] If you have a PhD and you're trying to break into data, this is probably one

[00:27:07] of the best places for you to be.

[00:27:10] If you're like an elementary school teacher or you're an Uber driver or

[00:27:13] like you're a cashier or a bank teller, probably not necessarily the best

[00:27:17] place to look for your first day to job.

[00:27:19] Maybe your second day to job based off of experience.

[00:27:22] But one of the interesting things that we talked about earlier, Adam, was

[00:27:25] that like in the end, I don't know about you guys, one of the jobs

[00:27:29] that did operations research, we used proprietary third-party tool to do

[00:27:33] a lot of optimization and thinking.

[00:27:35] So a lot of the times I was just using Excel to put stuff into

[00:27:38] that proprietary software.

[00:27:40] I've been at a company where we've used Python to do our optimization.

[00:27:45] That was a little bit more, more involved.

[00:27:46] But a lot of the times we're using pretty simple models.

[00:27:49] Adam, like what type of tools have you used in your different companies?

[00:27:52] Yeah.

[00:27:52] As far as operations research, a lot of the tools we use are those solver.

[00:27:59] Well, if you're doing a linear integer programming problem, like you're not

[00:28:03] going to write your own solver that beats the leaders, right?

[00:28:06] Like Groby, C-Plex.

[00:28:09] But they're, and a lot of the problems that you, that you work on are quite simple.

[00:28:15] Stephen will tell you stories about that.

[00:28:18] Like a lot, a lot of the, a lot of the optimization problems

[00:28:20] are very quite simple.

[00:28:22] I've had one though, the vehicle routing problem where it was real time,

[00:28:26] a real time vehicle routing problem.

[00:28:28] So I had to have a super fast response time that required custom algorithms.

[00:28:35] And that took a lot of, there's a lot of computer science work.

[00:28:39] We had to write it in Java, right?

[00:28:41] Python might be too slow for that.

[00:28:43] So it was like a custom meta heuristic written in Java to solve that.

[00:28:49] But outside of that, a lot of the problems you work on are not too

[00:28:54] complicated and a lot of, even when they are complicated, it goes into the

[00:28:58] modeling piece, not really the, like the mathematical model, if you can write

[00:29:03] it down on paper, you can use some third party solvers to solve it.

[00:29:07] It's kind of like when you do an interview, you do some crazy technical

[00:29:11] screen and then you get the job and you're making a crud app, you just like

[00:29:16] reading and updating a database.

[00:29:19] So that's what happens.

[00:29:20] Yeah.

[00:29:21] I felt kind of, kind of the same.

[00:29:23] Steven, you're using a lot of third party solver stuff as well?

[00:29:27] In my current job, no.

[00:29:30] But, well actually, yeah.

[00:29:32] Oh, okay.

[00:29:32] Well, sorry.

[00:29:33] I was thinking of like open source for like commercial solvers.

[00:29:36] Like, but I mean, yeah, mostly commercial solvers.

[00:29:39] As Adam mentioned, things like CPlex and Groby I've used in past jobs.

[00:29:42] And then other ones I've also used open source ones such as, I know the

[00:29:49] Python, like Python has like a variety of packages, like Pulp and PyMo where

[00:29:54] you can form the problem in them, but then they can call solvers.

[00:29:58] A popular open source one is Higgs.

[00:30:00] I forgot what it, I feel bad because I forgot off the time I had like

[00:30:03] what it's short for, but that's a nice open source one.

[00:30:06] But then like, in addition to like the solving tools, like there's

[00:30:08] also, you know, we are also going to use like Excel, Python, SQL, like

[00:30:13] cloud computing services like GCP and AWS.

[00:30:17] Yeah.

[00:30:17] Those are just some common ones.

[00:30:19] But then also too, like another thing is that going back to some of these

[00:30:22] problems, like sometimes you can't just like use a solver to solve because

[00:30:27] you might need a solution like very quickly.

[00:30:29] So sometimes you might need to write your own like heuristic to solve problem.

[00:30:33] And like to use like simple terms, like a heuristic is like, Oh, like, like a smart

[00:30:39] way that's like not guaranteed to find the optimal solution, but like the idea

[00:30:42] is that, you know, you write an algorithm that like find like good solution quickly.

[00:30:47] So like having to write my own stuff in Python like that too.

[00:30:50] But as far as simple problems, can you bring up like, like some of the problems

[00:30:54] that we've seen in practice are like a few decision variables, right?

[00:30:58] Like it's not anything complicated.

[00:31:00] I mean, yeah, there was one at a past job where it's like the complicated stuff

[00:31:05] was, was not like the optimist.

[00:31:07] Like it was like pricing for golf courses and like the actual model itself didn't

[00:31:11] have like hundreds, you know, tens of thousands of decision variables, but more

[00:31:15] so complicated stuff was just like making sure we were having like the right

[00:31:19] sensitivity of pricing where it's like, okay, well if we up the price by this

[00:31:23] much, like how, how much will that actually affect demand?

[00:31:26] So like, yeah, some of the stuff outside of the optimization was actually, it

[00:31:28] was a lot harder than like the actual mathematical model itself.

[00:31:33] It's pretty cool because after you've created this model, which once again, a

[00:31:37] lot of the times like Adam kind of hinted at earlier, you have to use predictive

[00:31:41] modeling to create a mathematical model.

[00:31:43] For example, like when I was at the refinery, right.

[00:31:47] It's not like you can just like magically have like the exact rules of a refinery

[00:31:52] with like literally tens of thousands of variables and stuff.

[00:31:55] Like you have to use mathematical, let's just say predictive analytics to try to

[00:32:00] figure out like how to predict how the refiner is going to act.

[00:32:04] And then you use prescriptive analytics to try to figure out how to work it

[00:32:07] optimally, but it's pretty cool because you basically create like, I don't know,

[00:32:12] like the right term, like a digital version of something that exists in real

[00:32:17] life.

[00:32:17] So for example, what Steven's talking about is like you create like a digital

[00:32:21] version of this golf course, but it's almost like you play Sims and you like

[00:32:25] try to like, Oh, well, what happens if I crank up the price?

[00:32:28] Do I make more money or less money?

[00:32:30] You know, what happens if we do, I don't know.

[00:32:32] What do you do at a golf course that doesn't change the price?

[00:32:34] But like, you can like basically like create almost like a Sims version of your

[00:32:38] business and you can kind of hand it to business leaders and be like, Hey, play

[00:32:42] with this lever, you know, move this lever up and down and we'll try to tell you

[00:32:46] what will happen.

[00:32:46] It's almost like predicting the future or at least allowing you to like almost run

[00:32:50] a simulation of being like, this is probably what will happen if you have this

[00:32:54] here, you know, this here and that there.

[00:32:57] It's a pretty cool tool that like Steven said in his LinkedIn post is pretty

[00:33:01] under appreciated in the data world.

[00:33:04] They'll call those digital twins, by the way.

[00:33:07] Ah, that's the term I was looking for.

[00:33:09] That's the term you're trying to, you almost had it.

[00:33:12] It's been a while.

[00:33:13] Yeah.

[00:33:13] A digital twin.

[00:33:14] Yeah, exactly.

[00:33:15] Which is pretty sweet.

[00:33:16] Anything else that the people should know about operations research and

[00:33:20] prescriptive analytics?

[00:33:22] Sure.

[00:33:22] I have something.

[00:33:23] So you actually made like a YouTube video, like a while ago about this, but

[00:33:26] like a good introduction is the diet problem.

[00:33:29] So the idea is, is that like each day and like there might be like small variants

[00:33:33] of it, so I might like confuse, I might change some of the details, but the

[00:33:37] idea is like, okay, it's like each day you have to eat, you know, obviously we

[00:33:41] all have to eat food because, you know, we need to eat food to live and you know,

[00:33:43] there's certain nutrients you want and there's certain food and you know, there's

[00:33:47] a variety of food items that you can pick which food items you eat such that it

[00:33:51] costs the least amount of money, but you satisfy, you know, calorie and vitamin

[00:33:55] requirements, et cetera.

[00:33:56] That's like a very nice, easy to understand optimization problem.

[00:34:00] That's something where you can just like Google like the diet problem.

[00:34:03] And there's like a lot of like resources and videos on including Avery, your own

[00:34:07] interesting video where you actually went to McDonald's, like use the, you know,

[00:34:11] use a diet problem model to determine what you're going to eat there.

[00:34:15] Yeah.

[00:34:15] That's well, I'll have a link to that, but yeah, I tried to find the McHealthiest

[00:34:19] combo meal that there possibly could have been using operations research.

[00:34:23] The answer was like, not very exciting.

[00:34:25] It was like five, like Dr.

[00:34:26] Peppers.

[00:34:27] I don't think water was an option.

[00:34:28] So diet Dr.

[00:34:30] Peppers, like two salads in apple, like a couple of apple slices.

[00:34:35] That was like it or something.

[00:34:36] So it was not, it was not very exciting, but yeah, operations

[00:34:39] research is very powerful.

[00:34:41] I want to ask you guys one last question.

[00:34:43] And if you can go back and tell yourself something, when you guys were first

[00:34:46] starting analytics, what would you tell yourself?

[00:34:48] Like, how would you like, what advice would you give your younger self in the

[00:34:51] field, Adam will go first and then we'll go to Steve.

[00:34:53] Not even data related, but one would just be, one is to network actually

[00:34:58] like meeting people and networking more.

[00:35:01] So when I was in school studying, I probably would have studied

[00:35:05] less and talked more to others.

[00:35:08] Turns out having a strong network really plays a huge role in your ability to get

[00:35:15] new jobs, change roles.

[00:35:19] A lot of the openings that you see are, I'm not based, this is not a real

[00:35:25] statistic, but it seems like half the time that half the time there's a role

[00:35:30] that's usually filled by an internal reference.

[00:35:36] So that plays, that just plays a huge impact on your career.

[00:35:41] As far as the data side, when I studied math, nobody told me that

[00:35:46] I had to go to graduate school.

[00:35:47] So I didn't know what I was doing.

[00:35:50] I just showed up and went to school.

[00:35:53] But yeah, I think as you progress throughout your career, like the people

[00:35:58] side of things makes a huge difference.

[00:36:00] So I'd probably focus more on that and less on the technical side.

[00:36:06] I love that.

[00:36:07] I don't know how familiar you are with our podcast, but we talk about

[00:36:10] the SBN method, skills, portfolio, network.

[00:36:13] And they're basically a third of the equation of landing good jobs.

[00:36:16] So I love that.

[00:36:18] Your advice is not even the skills part.

[00:36:20] It's not even the data part.

[00:36:21] It's all about the networking.

[00:36:22] I totally agree.

[00:36:24] Steven, how about you?

[00:36:26] Okay.

[00:36:27] It's hard for me to give my younger self advice.

[00:36:30] So I guess I would change it to give advice to like folks more so

[00:36:33] just folks breaking into data.

[00:36:35] Because there are times people might see something, they might've heard some

[00:36:38] of the content they were talking about, might be intimidated or scared about it.

[00:36:42] But I just think it's like data, whether it's like operations research

[00:36:46] focus, or even like visualization, engineering, machine learning, et cetera.

[00:36:50] I think it's like pretty fun, interesting, exciting that

[00:36:52] there's like a lot of good roles.

[00:36:54] And for the most part, I have trouble seeing myself doing any other kind of job.

[00:36:59] Like I couldn't see myself just being like a project manager and something.

[00:37:03] Now other people want to do that.

[00:37:03] It's completely fine.

[00:37:04] So like my thing would be like, don't get discouraged if you're

[00:37:07] trying to learn something.

[00:37:08] You're finding it hard.

[00:37:09] We all start from the beginning at some point, or we all had to.

[00:37:12] So like if you're like struggling with like SQL at first, you want to like,

[00:37:15] you know, just keep up, just keep at it.

[00:37:17] Realize that, you know, it's going to take time to learn.

[00:37:19] A lot similar like Python or some of these other math max topics in that

[00:37:23] there's a lot of like people you can reach out to on LinkedIn.

[00:37:26] There's like a lot of resources.

[00:37:28] There's your bootcamp and whatnot to help break into data.

[00:37:30] My big thing is like, if you're truly interested in like doing a job in

[00:37:34] data, then you know, pursue it.

[00:37:36] Don't get discouraged.

[00:37:37] Like don't let discourage or be intimidated by something like,

[00:37:40] you know, just go do it.

[00:37:41] I love that.

[00:37:42] And I think it's really important coming from that message coming from you guys.

[00:37:46] I'm considering we just talked for 40 minutes about like PhD level math type of

[00:37:51] equations, but the point that I think you're saying is like, even though that's,

[00:37:54] you know, what we studied, it's, it's what we kind of do.

[00:37:57] It's like, you actually get to the job.

[00:37:59] Once you get past the interview, everything's kind of easier than the

[00:38:02] actual interview, like a lot of the time, at least from a technical perspective.

[00:38:05] That's not always true, but it's true a lot.

[00:38:07] And I like what you said, like, don't be discouraged because we all used to suck

[00:38:11] at everything that we used to do and you just get better with it at times.

[00:38:14] So also, and to add on to that is that like, I think it's worth it at the end

[00:38:19] because you hear all these people like complain about jobs and things like that.

[00:38:22] And like looking back at it, I realized like it's a pretty cool job.

[00:38:26] Like in general, I think like data jobs tend to pay higher than a lot of other

[00:38:29] jobs, like you, I feel like the work I've balanced is like fire, like more

[00:38:33] businessy jobs, you're on challenging problems.

[00:38:35] I think you get to use cool technology.

[00:38:37] So, and yeah, so my thing is like, if you're interested in data, pursue it.

[00:38:39] And also I think at the end, in most cases, like it's worth it.

[00:38:43] It's worth the squeeze.

[00:38:44] Yeah.

[00:38:45] I like it.

[00:38:46] It is a pretty cush job.

[00:38:47] You get paid a lot.

[00:38:48] You don't have to, the deadlines are like pretty manageable.

[00:38:51] You get to like work remote a lot of the time.

[00:38:53] So.

[00:38:53] Okay.

[00:38:54] So I agree in the sense, like that's like been my experience, what you just

[00:38:56] described, but then like actually other people unfortunate in data, like they

[00:39:00] might've had different experiences, but I mean, still I'm a big proponent

[00:39:04] of the type of work.

[00:39:05] Yeah, of course there's, there's bad apples in every industry.

[00:39:08] I think that's for sure.

[00:39:09] Okay.

[00:39:10] Well, thank you so much, Adam and Steven for coming on the podcast.

[00:39:13] If you guys want to connect with these people or follow them on LinkedIn,

[00:39:16] we'll have their LinkedIn URLs in the show notes down below.

[00:39:20] You guys can follow them.

[00:39:20] They're posting great stuff about data careers as well as things

[00:39:24] like operations research.

[00:39:25] So if you listen to this and you're like, yes, operation research

[00:39:29] is the right path for me.

[00:39:31] Follow these people.

[00:39:32] Steven, Adam.

[00:39:33] Thank you guys so much.

[00:39:34] Thank you.

[00:39:35] Yeah.

[00:39:35] Thank you.

[00:39:35] It was a fun conversation.