Actuaries Put the “X” in DAX, w/Philip Trick

Rob Collie (00:00:00): Hello, friends. Today's guest is Philip Trick. Since it's just me and you, dear listener, I'm going to confess that I really enjoy it when people talk nerdy to me. And oh, did Philip deliver. You see, in addition to being a principal consultant here at P3 Adaptive, Philip is also an economist and an actuary, so he rocks that advanced math. And one of the themes that we've covered many times on this show is that the math behind the overwhelming majority of valuable business analytics is relatively simple. Here's a person, Philip, who not only is trained in the mathematics required to be an economist and to be an actuary, he also self-selected, earlier in his life, to go down that path. It would be natural to assume, or to guess, that he would find the simple math of most business analytics to be boring. Well, he does not find it boring.

Rob Collie (00:01:02): Anyone listening, who works in business analytics knows this, that even though the math is usually very simple, the problems themselves, and the methodologies that we use, are anything but simple. There's a lot of detail and challenge and stimulation, and even creativity, that goes into cleaning data, building a data model, writing the formulas to produce valuable metrics that didn't exist before, and I'm happy to say that Philip's been eating that up in his role here at P3. But we use the opportunity to explore that boundary and to ask how much does Power BI and related technologies, how much does that leak back upstream into his actuarial and economist world?

Rob Collie (00:01:45): That, in turn, led to a discussion about the X functions and their relationship to the array formulas, capability of Excel, that was just chef's kiss. Love it. We learned some new vocabulary words in this show; vocabulary words that are so rare that I think they won't even count as a wordagami. We talked about some fun, little math and probability riddles, and even though the conversation definitely touched on some advanced topics, the conversation itself stayed at a very digestible, down to earth level. He's a really interesting person. He has a lot going on. In fact, he's one of the rare exceptions at P3, where P3 isn't enough to keep him professionally occupied, professionally utilized, so he wears a couple of other professional hats at the same time. All that and more when we get into it.

Announcer (00:02:37): Ladies and gentlemen, may I have your attention, please?

Announcer (00:02:41): This is the Raw Data by P3 Adaptive Podcast, with your host, Rob Collie. Find out what the experts at P3 Adaptive can do for your business. Just go to P3daptive.com. Raw Data by P3 Adaptive is data with the human element.

Rob Collie (00:03:03): Welcome to the show. Philip Trick, how are you today?

Philip Trick (00:03:06): I'm doing pretty well, Rob. Glad to be here.

Rob Collie (00:03:09): Let me start with a question that you might not anticipate. It's a multiple choice question. What is your middle initial, okay? And the only three options are A, J, or K. Are you looking for none of the above? Are you looking for the option D on this question?

Philip Trick (00:03:25): No, I'm looking for how I can fit into this A-J-K kind of concept. I'm going to have to go with A.

Rob Collie (00:03:32): You're going to go with A? All right.

Philip Trick (00:03:34): Yeah.

Rob Collie (00:03:34): Do you know why I'm insisting that it be one of these three letters?

Philip Trick (00:03:38): I haven't a clue.

Rob Collie (00:03:40): All right. Well, because if your middle initial was A, J, or K, your name would completely rhyme with the name of the author of Blade Runner, Philip K. Dick. Philip A. Trick is fine. J is also okay. K, of course, would be spot on, but if you're going to tell me that it's S or something, I don't know, man. I don't know what we're going to do.

Philip Trick (00:04:04): I love that. Blade Runner is one of the great movies, one of the original... I don't know if original, but one of the best sci-fi movies, especially if it's time.

Rob Collie (00:04:13): Agreed, agreed. It is a little slow. There was something about audiences back then. If you go back and you watch Star Wars, Episode IV... I remember showing it to my kids for the first time, and they'd been raised on '90s content and early 2000s content. I was blown away at the pacing of Star Wars, Episode IV. It is so slow. It takes forever to build into something interesting, in terms of attention holding. And the same thing's true with Blade Runner. It's a long and thoughtful movie. And then I think it culminates in perhaps one of the strongest 20-minute closing sequences of all time. Like it pays off.

Philip Trick (00:04:52): Right.

Rob Collie (00:04:53): But it can put you to sleep, if you're a little drowsy.

Philip Trick (00:04:57): It can, but what always got me about those movies, and those two in particular, is while they're slow, it's all about world building, right?

Rob Collie (00:05:04): Mm-hmm (affirmative). Yeah.

Philip Trick (00:05:04): In the Blade Runner, you're getting the whole setup, what is going on in the world and why is this particular character important? And then it just sets you up for the surprise later on.

Rob Collie (00:05:15): All right. So, in real life, it's not A, J or K, and that's okay. Let's try this one, do you at least have like an Android phone?

Philip Trick (00:05:23): I do have an Android phone.

Rob Collie (00:05:24): Fantastic. Do Android phones dream of electric sheep, is the original name of Blade Runner? Isn't that the original title? I think that's what it is.

Philip Trick (00:05:32): It's something like that.

Rob Collie (00:05:33): Yeah. All right. Well enough with that shtick. I've been thinking about that bit for the last three days. It went maybe about as well as I expected. We'll see.

Philip Trick (00:05:43): You got to try.

Rob Collie (00:05:44): You wear a couple of hats, don't you?

Philip Trick (00:05:45): I wear a bunch of hats.

Rob Collie (00:05:46): A bunch of hats. All right, why don't we run those down.

Philip Trick (00:05:49): All right. So hat number one is economist. So I've got a graduate degree in Economics from BU, from one of the great economic minds in the country, laurence Kotlikoff. If you're not familiar, he's the premier Social Security analyst in the country. And economics is just like an absolute passion of mine, and that's where my passion for data came from, digging through boatloads of data from the Bureau of Labor Statistics, from the CDC, from all those different sites. And at the time, using Stata and raw Excel to all right, let's make something out of this.

Rob Collie (00:06:24): Mm-hmm (affirmative).

Philip Trick (00:06:24): And then from that, I picked up the actuarial hat. That is the risk analysis side of it. And it's very, very closely related to that economics hat, but with an extra layer of math and statistics on top of it, just to be extra boring.

Rob Collie (00:06:41): Okay. That's interesting. I would actually think it would be the reverse, that actuarial work would be more specific, and therefore a little less complicated, than modeling the gigantic international economy, for instance.

Philip Trick (00:06:55): There's a little bit of that, and a little bit of the other direction. In our actuarial models, we have to take into account how the economy is going in addition to, how-

Rob Collie (00:07:06): Okay.

Philip Trick (00:07:06): ...risks are developing.

Rob Collie (00:07:07): Are you saying the outside world bleeds into your carefully crafted laboratory conditions? I-

Philip Trick (00:07:13): Really, it's more of an art a lot of the times. And we always are saying we'll give you a number, but it's not right.

Rob Collie (00:07:20): Yeah.

Philip Trick (00:07:21): It's going to be something around that number, we hope.

Rob Collie (00:07:23): When did you decide that economics was a thing for you? When did you decide that the actuary path... how early in life did you know that this was your path?

Philip Trick (00:07:32): I took one economics course in high school and it just hit me in the face. It's like, that is what I want to study.

Rob Collie (00:07:38): Okay.

Philip Trick (00:07:38): Just love the whole concept of trying to think about why people make decisions. Because that's what it is. Why do people make the decisions they do? The actuarial side... my mom is an accountant and she was an insurance accountant-

Rob Collie (00:07:52): Okay.

Philip Trick (00:07:52): ...for many years and kept telling me about the actuarial career path. I just blew her off because she's mom, right. Finally, my senior year of college, I'm like, all right, I really need to figure out what I'm going to do at this point. And I was taking econometrics, which is heavy linear regression theory, predicting-the-future stuff. And in the math department, there was a little flyer for the actuarial career that basically said, you like using data, you want to look into a crystal ball? All right.

Rob Collie (00:08:22): So you kind of knew what you were going to do in a way? To say that you were in on economics in high school makes you a bit of an outlier in that, profession wise, you're closer to where you thought you would be in high school than almost anyone else.

Philip Trick (00:08:39): Yeah. I mean it was lucky, I would have to say. I was good at math. Math in high school, I didn't care for, I didn't understand what the point of any of it was. But then economics, starting to apply it. At the high school level, you apply just very little bits of it. But I think the first thing that caught my mind was the concept of the Edgeworth box. You have two people and a limited number of supplies, you have water and you have apples. One person has all of the apples, one person has all of the water. How do you model where they trade?

Rob Collie (00:09:09): Okay.

Philip Trick (00:09:09): Just the concept of the math, to talk about the interaction of these people, caught my attention. And then I took some economics in college the first year as I'm trying to pick a major and it's just like, yeah, yeah, this is it. I have figured out my path, I love this stuff. And then I just kept digging deeper and loving it more. Game theory, econometrics, labor economics, health economics. Just every little bit was just another piece that fit with me. It's just dumb luck, really.

Rob Collie (00:09:39): Well, everyone's got luck. It's just that most people's random path to where they are today, they sort of snap into their final form. Is it our final form? I'm not sure that it is. What we call our today form, we snap into it later in life. You had your accidental collisions earlier than most. It's not a binary thing. It's just a question of degree. So you like yourself some Nash equilibriums?

Philip Trick (00:10:03): Oh, Nash equilibriums, prisoner's dilemma.

Rob Collie (00:10:06): Yeah.

Philip Trick (00:10:06): Arrow's theorem.

Rob Collie (00:10:07): All that stuff. Okay. So, economist, actuary. What other hats you got?

Philip Trick (00:10:13): Probably the biggest joy in my life is I'm a dad. I've got a little eight year old girl. Light of my life. Loves playing Minecraft, is really bright, builds Legos just all day if she can.

Rob Collie (00:10:25): You're one of those parents. The ones that make the rest of us feel a little bad about ourselves, right? It's like, yeah, they're great and all, but man, they're also... my life is really hard now.

Philip Trick (00:10:36): I mean, there's days, right?

Rob Collie (00:10:39): So you're a dad. You're also wearing a P3 shirt.

Philip Trick (00:10:42): I am wearing a P3 shirt.

Rob Collie (00:10:44): So that's not a hat, that's a shirt.

Philip Trick (00:10:46): That's a shirt. So the data side. I'm also a data analyst, data collector. And that came about because I had to. Actuarial work requires boatloads of data, and trying to explain the kind of data I wanted to it guys like, all right, I need it in these cross sections of dates, these types of granularities, these dimensions. Or they send me a data set and it's out there. Talking to the IT team, I give them a request for the data, and it's got dates and dimensions and all these different pieces, but it goes over their heads.

Rob Collie (00:11:23): Yeah.

Philip Trick (00:11:24): And so I had to pick up a lot of ETL processes, pick up how to interact with SQL. And I'm kind of a lazy guy, I don't like to redo things over and over again. So my very first project, I learned VBA so I could-

Rob Collie (00:11:39): Oh yeah.

Philip Trick (00:11:39): ...query SQL databases without having to copy and paste out of them.

Rob Collie (00:11:45): Yeah. I wonder what the worldwide usage of VBA looks like, if you could chart it, at the introduction of Power Query. How VBA falls off? Even like a 2% or 3% decrease in VBA usage would be... that's a massive swing. But I bet, if such a thing could be charted, we would see that we're off-peak VBA now, thanks to Power Query.

Philip Trick (00:12:07): I'd like to think so. Funny story about that. I went to Microsoft's offices in 2016 as part of the ModelOff financial modeling competition.

Rob Collie (00:12:17): Oh, you were in that? Wow.

Philip Trick (00:12:19): I was. I was one of the finalists.

Rob Collie (00:12:21): You were one of the finalists in that? How did I not know this? Did I know it and forget? I don't know.

Philip Trick (00:12:26): So I was up at Microsoft's office. This is 2016, so note that Power Query has been out for six years at that point, or something?

Rob Collie (00:12:33): Three years.

Philip Trick (00:12:34): Three years. So I'm complaining to the Excel developers on the spot, like, you need a better query system. Have you used Power Query? What's Power Query?

Rob Collie (00:12:42): Exactly. Yeah.

Philip Trick (00:12:43): They introduced me to it right there on the spot and it invalidated, I don't know, 10,000 lines of VBA code that I'd written for, basically, doing all of those manipulations. And it was one of the best days of my life. The number of things that I've done with Power Query since then has just been incredible. And that's just Power Query alone.

Rob Collie (00:13:03): Yeah. And with regards to that P3 shirt, we have a little bit different deal with you than with our average principal consultant, don't we? We're not your only professional outlet or source of income. You do other things as well, so you're a time-sharer.

Philip Trick (00:13:18): That's right. So I work for P3 as a principal consultant, on a contract basis essentially. I pick up clients, [inaudible 00:13:26] clients and help on an as needed basis, at will. By day, you might say I'm the chief actuary for an actuarial consulting firm, where I'm primarily writing reports based off of the types of data that we're building at P3. So it's taking the skills that I've had to leverage, to accelerate my actuarial report writing and turning it into a dashboard. But also in a very different way, because on my actuarial side, I interact with, primarily, insurance and risk-based types of problems. So I've gotten to learn a lot at P3, working with stuff like sales data that I've never had to touch before. Working with IT-oriented data, and just a bunch of different types of non-finance data, that's been a neat statistics challenge.

Rob Collie (00:14:16): It's not like we have so many insurance clients that we can just feed you a steady stream of nothing but that, right. But one of the things that Kellan says is that six months in the principal consultant chair at P3 exposes you to so much business. Set aside the data challenges that you face, those are also kind of amazing. There's an amazing diversity, and so many different challenges that'll come up, even just in a six month period. The exposure to the business world, this cross-section of the business world. And I live this. As the first consultant at the company, I got to learn so much about how the world worked, in such a short, like a cram session. After living inside this Microsoft bubble for more than a decade, I needed a crash course in real world, and I got it in a hurry. So it's really cool to hear you echoing that same observation. When you were saying that just a second ago, about that exposure thing, there was something in your head. Were you thinking about the data problems or were you thinking about the businesses?

Philip Trick (00:15:16): I was thinking about the business problems.

Rob Collie (00:15:18): Yeah.

Philip Trick (00:15:18): At the end of the day, the data problems are all pretty much the same.

Rob Collie (00:15:23): Yeah.

Philip Trick (00:15:23): The data is messy, and you want it to be clean, you want it to be organized, you want it to be relatable, and you want to be able to say that it has meaning. But then it's the insight side of it that changes dramatically across the different business types. At P3, I've worked with an architecture firm, with a internet services firm, with a software as a service company dedicated towards governments. A bunch of different firms with all different types of questions, like how are our customer service representatives handling tickets? Which I did get to relate back to my insurance experience in terms of all right, well, a ticket is a claim, it's a problem. You open it, you handle it, you solve it. That's on a much shorter timeline, but you still want to analyze the kind of details that go into opening and closing these claim, or sorry, tickets.

Rob Collie (00:16:12): Tickets, yeah.

Philip Trick (00:16:15): Then the architectural firm is interested in how are their people using their resources? It's an internal reporting. There's no real finance involved, but it's like a cost accounting problem to solve, and allocating costs internally and understanding how it all breaks down.

Rob Collie (00:16:37): For an architectural firm, that's a very ill-defined, vague sort of challenge, which requires some really nuanced thinking and even a little creativity-

Philip Trick (00:16:47): Yeah.

Rob Collie (00:16:48): ...in order to come up with something that's actually useful. And I love it that problems like that are still things that people engage with. You could look at that problem and say, nah, just throw up your hands. As long as there's more money in the bank account at the end of the year than there was at the beginning of the year, all's good, right. BI by bank account. And I think a lot of firms in that situation would sit tight, on BI by bank account. So I love it when there's something nuanced like that. That's the, mwah, chefs kiss. That's the really, really cool stuff. I can see you lighten up a little bit talking about that allocation problem.

Philip Trick (00:17:22): Yeah. Finding these problems and identifying them is the fun. And then the other part that I really love is, there's all these aspects of risk that often get untouched, even in what we call modern BI. One of the problems that I'm working with Rob Davies on, is he's working with an insurance agency to produce quotes for insurers.

Rob Collie (00:17:46): Okay. Yeah.

Philip Trick (00:17:46): And I've been on the expert witness side of an agency that has been sued by an insurance company because of bad record keeping. And so we're building in new data, documentation points into this power app, that will give them the ability to document from production all the way to binding-

Rob Collie (00:18:06): Okay.

Philip Trick (00:18:07): ...in a way that, as an expert witness, I would say significantly reduces their risk. But it's not something that you'll see directly in the bottom line until you get sued.

Rob Collie (00:18:17): I see. Yeah. It's one of those stochastic events, right?

Philip Trick (00:18:21): Exactly.

Rob Collie (00:18:22): Getting sued or not sued. That was the word, the big word, from last podcast with Bill.

Philip Trick (00:18:27): Stochastic is one of my favorite words. That and heteroscedasticity.

Rob Collie (00:18:30): What? What the what? Say that again. That's not going to actually count as a wordagami, you know why? It's too rare. It's not in my database of words. And so it's going to get flagged as if it was like a brand name.

Philip Trick (00:18:45): Yeah.

Rob Collie (00:18:45): I drop out brand names from the wordagami model. So, okay, say it again. What was that?

Philip Trick (00:18:50): Heteroscedasticity.

Rob Collie (00:18:52): Oh yeah. That's right. That's right.

Philip Trick (00:18:54): Super simple.

Rob Collie (00:18:55): What the hell does that mean?

Philip Trick (00:18:57): So have you ever fit a line to a series of points? Yeah. Do you know what assumptions you made?

Rob Collie (00:19:04): Lots, I'm sure. Like that all the other dots, if they were there, would look the same, for instance.

Philip Trick (00:19:12): So one of the biggest assumptions you've made in fitting that line is, you've assumed homoscedasticity across your points.

Rob Collie (00:19:20): Okay.

Philip Trick (00:19:21): And that's to say that, there's the same level of error within each point.

Rob Collie (00:19:26): Okay.

Philip Trick (00:19:26): Heteroscedasticity is the assumption that each point has a different variance, and that variance is somehow correlated itself. Excellent example of that is house prices. A $100,000 house listing will vary by, call it, 10%. So that's $10,000 plus or minus. A $1 million house listing will vary by 10%, plus or minus $100,000. So you have a differing variance between those two points. So if you plot a line, you need to adjust it. And your typical Excel tools and stuff won't do that by default.

Rob Collie (00:20:03): Oh, interesting. Okay. If you take the low end of both of those, 90 to 900, is the slope of those two lines going to be the same? From 90 to 900, and 110 to 1.1, the slope is going to be the same. It's still a 10x slope in both cases, right?

Philip Trick (00:20:20): It's a 10x slope, but you're going to be getting further and further away from your initial fit line.

Rob Collie (00:20:26): Interesting. I'm trying to visualize this in my head. If the slope is the same, that means those lines are parallel, if I drew them on the graphic.

Philip Trick (00:20:33): Oh, sorry, sorry. The slope won't be quite the same, because you're going from a listing point of 100,000 up to a listing point of a million.

Rob Collie (00:20:44): Yeah. My intuition's breaking down on this one just because I need to sit down and draw it. I was expecting the slope of those two lines to be different because of the wider fluctuation. But then I did, oh, it's 10x.

Philip Trick (00:20:55): Yeah. Yeah. The 10x is the middle line. The bottom line will be 9x and the upper line will be 11x.

Rob Collie (00:21:01): I see. Okay.

Philip Trick (00:21:01): Roughly.

Rob Collie (00:21:02): The thing I'm doing here is I'm not accounting for Y-intercept, mx+b. I'm just assuming that they're both zero. Yeah, it's definitely a different slope. What are we talking about? Yeah.

Philip Trick (00:21:13): Takes of thinking, it doesn't just come to people.

Rob Collie (00:21:15): All right. So hetero, that makes sense. Heteros is different.

Philip Trick (00:21:20): Heteros different.

Rob Collie (00:21:21): Not all the same. But the second part of it, scedasticity?

Philip Trick (00:21:23): Scedasticity.

Rob Collie (00:21:25): Scedasticity.

Philip Trick (00:21:27): I don't know what the root is, but it's essentially variance.

Rob Collie (00:21:31): I think this might have come from Mary Poppins, actually. I think this was a song that they sang in Mary Poppins. It sounds like something she would say, right?

Philip Trick (00:21:39): It does. A whole bunch of letters just kind of gobbled together. Supercalifragilisticexpialidocious. Scedasticity.

Rob Collie (00:21:49): Well that's right. Yeah. Supercalifragilisticheteroscedasticidity. Even though you think the slope is the same, it's not.

Philip Trick (00:21:54): A spoon full of sugar helps the variants go down.

Rob Collie (00:21:58): Yeah. Check the Y-intercept. All right. So the other branching-off point there is that, the world that you come from on the actuarial side is still incredibly foreign to me. I'm technical. I got a computer science degree, in theory it wasn't very helpful. I also got a math major. I navigated my way through that major without having to take any of the things that you would think a math major would have to take. I was pretty good at that, actually.

Philip Trick (00:22:27): No diffiQ, no linear algebra?

Rob Collie (00:22:29): Oh no, no, I had to take all those things. But those were required by the CS degree, believe it or not. And that's what left me so close to having the math major, I was like three more classes, whatever. Yeah, linear algebra, diffiQ, calculus in general. Yeah, I never used it. None of it. And in some sense, I could imagine, I know this isn't the case, in an academic sense I could imagine two things. I could imagine number one, that the problems you run into at P3 Adaptive with our clients, like mathematically, they might be boring mathematically by comparison.

Rob Collie (00:23:00): And the second thought would be... and all these Power BI tools, clearly Power Query always helps. You said that dirty data, you've got to get it organized, so Power Query is going to be a godsend in any environment. I would have this suspicion that maybe DAX isn't as useful in your actuarial world. That's the two points. One, are the math problems boring in the business world, outside of actuarial sciences? And is DAX useful in the actuarial sciences world?

Philip Trick (00:23:31): So one, yes and no. Yes, in the sense that, while I'm developing the data up to the points of those problems, and in some cases even developing measures and visualizations to basically plot these stepping stools for these more complex calculations. Going to that ticket's problem, as an example, the previous analysis had been looking at it very one-dimensionally. And so I put together a set of measures to look at it on a two-dimensional date basis, not just here's a 365-day window, let's take all the tickets.

Rob Collie (00:24:11): Yeah. There you go.

Philip Trick (00:24:12): It had claims tickets in this window and within the same window, and instead turning it into a bucketing mechanism that gave you statistical rigor around how these were defined. In this case, identifying when a ticket was a repeat ticket.

Rob Collie (00:24:30): Okay.

Philip Trick (00:24:30): So you close the ticket and it reopens.

Rob Collie (00:24:33): Yes.

Philip Trick (00:24:33): And so that was a very fun statistical challenge in terms of, all right, how far back do we want to go to figure out if it's a repeat ticket?

Rob Collie (00:24:41): Yeah.

Philip Trick (00:24:42): A year, like, all right, if I have my AC get checked every year, I'm a repeat customer. Is that repeat ticket a problem, or just standard business?

Rob Collie (00:24:53): Mm-hmm (affirmative).

Philip Trick (00:24:54): But if I'm calling back within 30 days or within 45 days, that's a different issue. That is, all right, did we actually resolve it or did we just close it and we're having to re-address it? So I got to do a lot of statistical formatting around that kind of a problem.

Rob Collie (00:25:11): Okay. Let me challenge you a little further there. That problem you just described, I've actually dealt with exactly that same problem for a client back in 2013. This was a managed services company that would come in and install systems in an office building. We looked for exactly that same thing. So after the install, did we have to go back within the first seven days or first 14 days or whatever? I thought that was beautiful. I loved looking at that problem. I loved working on it. There was so much nuance to it. And there was all kinds of things, like, was a support ticket, when it was filed, was it dealt with within the... what's the acronym for how quickly you have to solve something? SLA. Was it solved within the SLA?

Rob Collie (00:25:54): But in the end, I didn't think of that as a statistical problem, I was just defining a numerator that was considered a negative, like not a good thing for the business as that numerator went up. But then you call it statistical. Were you bringing something, the extra statistical plus plus, to the story?

Philip Trick (00:26:11): Yeah. So if you assume that this ticket today, you don't know if it's a repeat or not, because you don't have any history, and you say, let's go back one day. And all right, one day it's not a repeat ticket, two days it's not a repeat ticket, three days so on. And you find that, all right, if we go back 65 days, it is a repeat ticket. Let's say you have a thousand tickets like this. You can create a distribution that says, here's the percentage of it growing.

Rob Collie (00:26:37): I see.

Philip Trick (00:26:37): You can identify inflection points there that say, all right, at this point we have a pretty big drop-off, what does that mean? In this client's case, they are an internet services company, they provide modem updates, so at 365 days, there's a pretty significant inflection point where, all right, you would have calls, somewhere like more like 350 days, where you'd have pretty big drop-off. And then there was another pretty big drop-off after about 30 or 60 days. These seem to be just standard service calls. This kind of window is maybe it's a middling problem, some type of hard-to-diagnose problem, that's somewhat recurring, like imagine a loose wire in your backyard that heats up during summer and disconnects, versus the, all right, this percentage of problems are definitely repeats, it's within 35 days. And then using that to create benchmarks for the different customer service reps.

Rob Collie (00:27:35): Interesting. Okay. That's cool. I get it now. Defining what the look back window should be.

Philip Trick (00:27:41): Exactly.

Rob Collie (00:27:42): In the other situation, the other client I was talking about, it was just sort of defined. Also a number, even just internally, a number had been chosen. It's like, if we're back within the first two weeks, it's a problem. Maybe that number wasn't quite optimal. It was probably close. Like the shared birthday problem. That blows people's minds. What's the break-even for 50-50 likelihood that two people share a birthday in a group? Isn't it around 20 people or something? Do you know, off the top of your head?

Philip Trick (00:28:10): I don't, but there are a lot of fun problems like that. Like one that I saw, a really awesome visualization on recently, was the train schedule problem. A train shows up every 20 minutes and you show up randomly. What is your average wait? Well, that's easy, it's 10 minutes, because the train is showing up every 20 minutes. But now, once you make the train average and not a fixed schedule, you show up randomly, the train shows up randomly, the train shows up on average every 20 minutes, what's your average wait? It's going to be 20 minutes. And it's because of the integral of the two uniform distributions mixing. It's a really neat, simple solution. And it's always going to be that average of the train.

Rob Collie (00:28:54): Wow. Okay. So my gut instinct says that's not right.

Philip Trick (00:29:01): The gut instinct is really rough.

Rob Collie (00:29:03): So the break even is 23 people. If you have a group of 23 people, you've crossed the 50% likelihood that two of them share the same birthday. Bizarre, right? Wait a second, it's 365 days, it seems like the chance should be like one in 30. The only way I ever made sense of the birthday problem is to reverse it and say, imagine 22 of the people are already on the map, their birthdays are spoken for-

Philip Trick (00:29:26): Mm-hmm (affirmative).

Rob Collie (00:29:26): ...and none of them overlaps, so you've got 22 distinct dates. And now you've got to throw a 23rd dart at this map and miss the other 22. Okay. Chances are you're going to miss it, but every one of the darts is like that.

Philip Trick (00:29:37): Right.

Rob Collie (00:29:37): You have to throw 23 darts, and miss 22 others, every single time. And that's the only way that I've ever been able to make sense of the birthday problem. But this other one, this train schedule problem, never heard it before. I know what I'm doing while I'm falling asleep tonight. I'll not be listening to podcast, I will be trying to make intuitive sense of the train schedule problem.

Philip Trick (00:29:57): To make it kind of easier, the smallest level of average wait that you'll have is 10 minutes, which is the perfect distribution. The longest average wait you'll have is, let's say that it is over a 24-hour day and every one of your 12 trains shows at the end of the day, now your maximum average wait is actually 120 minutes. And so you're going to be integrating between those two points. And the average number of possibilities will come out to the average show up of the train.

Rob Collie (00:30:29): This reminds me a little bit of one of the more interesting classes that I took in college. My computer science degree was, in theory it was about operating systems. The professor was very much a theory person. And so it was all about queuing and Markov chains. I loved that stuff. Like a lot of nerds, myself included, at one point or another in their lives go, Ooh, I bet it'd be fun to be a traffic engineer and control stoplights and traffic pattern designs and things like that.

Philip Trick (00:31:02): I could do this better.

Rob Collie (00:31:03): Yeah. And it's the same sort of thing. There's a flow of jobs, cars, coming into a system, in what order do you process those jobs? Are they all equal priority? So one of the cool things in all of that was, there's many cool things about it, but one of them was that you can have a scheduling algorithm that is fair, or you can have a scheduling algorithm that produces the lowest average wait time. And the lowest average wait time, that algorithm kind of ignores fairness. You're a long job. You had a job that's going to take like 30 seconds of CPU time. And there's another job that came in after you that's only going to take like a split-second. Process the split-second job, send it on its way, make that person happy. So the 32nd job, in theory, could sit there forever.

Rob Collie (00:31:52): Right.

Philip Trick (00:31:53): It's like the FastPass line at the roller coaster park, you just sit there in line while all the people walk past you all day long. And that's not fair.

Philip Trick (00:32:02): Right, it's not fair. But you get lower average wait times. And I had a very slight slice of that in my Comp Sci 102 course, with that exact queuing thing. You line people outside of the professor's office for office hours and one person's going to take an hour, every other person's going to take five minutes. If you take that hour-person, and the other five people wait an hour, your average wait time is an hour. But if you process each of those five-minute people and then do the hour-person last, even if they were there first, your average wait time is five minutes or something.

Rob Collie (00:32:33): Mm-hmm (affirmative). So I wish I remember the exact example. I wanted to have this professor come on the show, but unfortunately, it was really sad, he passed away relatively young. But he told the story about how he modeled the School of Engineering's mainframe systems. They're not really mainframes, so like Sun Microsystems, Unik Systems. Big server, big shared resource.

Rob Collie (00:32:55): He modeled it and came to some sort of completely counterintuitive conclusion that was along the lines of, if we shut one of these computers down, like a CPU or something like that, it was like turning off a resource in the system. Without changing the incoming loads at all, the whole system would run faster. The system admin who ran that whole thing said, no, it won't. So they actually had like a case of beer or bet, and on a weekend, which was the only safe time that they could really do such a thing, they went in there and they turned off that resource. And sure enough things ran faster.

Philip Trick (00:33:32): That's crazy.

Rob Collie (00:33:33): Yeah. I actually had an experience with this related to Power Pivot in the early days. We stood up our own server farms when I was working at this other outfit, my first job outside of Microsoft. We had these servers stood up in our Rackspace environment, and we were running our own private cloud of Power Pivot for SharePoint. This is before Power BI service exists, and the cloud service, all that kind of stuff. We didn't have those things back in my day, we had some models and some reports.

Rob Collie (00:34:00): They were slow. You'd click a slicer and you'd wait. We kept telling ourselves though, this is okay, because right now we're only running on the single-CPU version of this server, so let's go get the four-CPU server. And we were just thinking it's going to have impact on the response time. We're like, okay, if the response time were a quarter of this, we'd be fine. The day comes. Great fanfare. The four-CPU server rolls in, we hook it all up. Which was hard. You have to install all of that shit. It was not fun. And then we go and we click the slicer, and it's slower than before. And we're like, wait, wait, wait, wait, wait, that must have been some sort of startup of a [inaudible 00:34:44], something wasn't warm.

Rob Collie (00:34:46): So we do it again and still significantly worse. Rather than four times better, it was significantly worse. And it turned out, the euphemism for this was called negative scaling. Let that roll off the tongue, negative scaling. More CPUs, the noise between them communicating and all that kind of stuff, for certain types of loads, actually made things slower. So in that same server, the engineer at Rackspace went in and physically pulled three of the CPUs out. And now it was faster again. It wasn't faster than what it had been. We didn't get any improvement, but at least it took away the negative scaling. It's like pulling three CPUs, everything went faster. I guess we don't really do multi-CPU servers quite so much anymore, as a multi-core thing. Across cores in a CPU, we didn't have that problem. But across CPUs, you did.

Philip Trick (00:35:38): That's fun. I mean the multi-core, multi-CPU threading in Excel was a really big deal. I guess that was Excel 2007. Our big models were five-minute calculation times, and we developed them very specifically to be clean and linear to process quickly. But once that multi-core came out, and Excel could create these calculation trees, we revisited the model entirely. And we were able to take that five minutes down to 10 seconds just by making it wide, instead of tall.

Rob Collie (00:36:12): I sat in Dwayne Campbell's office, Dwayne being the architect who was in charge of that effort, and listened to him explain, one of the smartest people around, he just explained very simply, very matter-of-factly, how he was going to go about the design for how to make Excel multi-threaded. Because 2007 was the last release of Excel I worked on and I was there from the beginning. And I remember him saying like, look, we get these recount dependency chains, and this is the way we sort, where we analyze them and sort them for processing by the CPU.

Rob Collie (00:36:45): Spoiler alert, it's a single-column sort because we're only using one core. And he said, so now we're going to look at the number of cores available to us, and we're going to make four columns if there's four, and see how much we can do in parallel. Because Excel has, for other obvious reasons, it has to have perfect knowledge of recount dependency chains. It was actually relatively trivial. I'm sure that the implementation wasn't, but it was just the simplest algorithm ever that he was explaining. And he wasn't bragging about it. It was just like, this is the way we do it. I was like, oh, this feels like my CS degree, this feels a little bit like it. That was a good five minutes.

Philip Trick (00:37:23): That happened. We took our formulas and we classified them and created basically a diagram of how the different sections went together. Initially being one to the next, and I, well, we don't need these for this one, skip, skip turning into a tree. And then the scaling as we got more cores for that was just incredible. It gave us better understanding of what we were doing anyways. I was overhauling a previous actuary's work and it was just awesome.

Rob Collie (00:37:49): That was one of our features we put in the Wall Street bucket, at the time. You weren't working on Wall Street, so you weren't playing fair, you were using our feature anyway. We could have called it the Actuarial and Insurance bucket, but we didn't. That's really neat. I never got into designing spreadsheets to take advantage of that. How do you go about designing, approaching spreadsheets differently once it went to multi-threaded? How do you know? It's not like Excel has a feature that tells you, Hey, if you change this formula here, here, and here, you're going to get better performance. It doesn't do that for you. So what does that analysis even look like?

Philip Trick (00:38:23): So, for me, it was revisiting it with a little bit of my limited programming knowledge. I knew that a clear dependency chain, without cross references, could calculate on its own. We had a bunch of these smart formulas that were like, all right, take this and take this and take this and take this, and then give me a thing in one cell. That was merging all of these cells into this one place, and in a [inaudible 00:38:52] dependency chain, that's all right, we're stuck with linear. So it ended up being, use more cells to do the same thing, but instead of calculating all five things in this one cell, that's a feed on the next cell, calculate each one in [inaudible 00:39:04] place.

Rob Collie (00:39:04): Right. Right.

Philip Trick (00:39:06): And then when I want the summary, create an extra summary. So it's just like, all right, there's my summary, over here off to the side, and then keep calculating off of the individual pieces. It creates that wider chain. Actually the best tool for doing that was the trace dependencies, arrows in Excel. I'm trying to find an error, now I was trying to optimize my formulas, I wanted one arrow back and I didn't want splits all over the place.

Rob Collie (00:39:31): Mm-hmm (affirmative). That makes sense to me. Excel, its algorithm for breaking things up across cores, it never subdivides one cell of the spreadsheet across cores. If you're tying lots together, lots of inputs from other cells together in a single cell, you make it so that Excel can't split that workup. Well, that's a problem we don't get to think about with DAX.

Philip Trick (00:39:52): That's a problem that I hardly think about anymore at all.

Rob Collie (00:39:55): Yeah.

Philip Trick (00:39:55): So your second question earlier was about, DAX doesn't seem to be all that useful.

Rob Collie (00:40:00): In the actuarial sciences, yeah.

Philip Trick (00:40:02): In the actuarial sciences, it absolutely is. Because... my favorite functions are the X functions. I love the X functions.

Rob Collie (00:40:09): I bet.

Philip Trick (00:40:10): Because I can do things. Like in Excel, I have to create this giant matrix, X across the top, Y across the side. And I'm filling it in, and I'm doing sums across diagonals with these crazy SUMIFs. In DAX, that is a SUMX across a CROSSJOIN.

Rob Collie (00:40:27): Mm-hmm (affirmative).

Philip Trick (00:40:28): And it's just that simple.

Rob Collie (00:40:29): Yep.

Philip Trick (00:40:30): Not only is it just that simple, but now instead of building a pivot table that does this, and then having filters and having to reference these pivot table features, I can throw them into visualizations. I can build paginated reports off of them, and print out the 150 different variations on this for my report, and create 10 times the product that I was before, in half the amount of time.

Rob Collie (00:40:54): Okay. So let me you tell you a little secret. I expected you to say something along these lines. Not the X function, I get that, we're going to dive into that. I was expecting you to say, yeah, it's super, super, super useful. And if you didn't, then I was going to say, oh, come on, you haven't thought about it enough. So this has been something I've been saying for the better part of 10 years, which is go find places where Excel array formulas are being used. It could almost be an industry of its own. Just go find places that array forms are being used, and turn them into X function, calculations. That could be a whole company business model, right?

Philip Trick (00:41:29): It probably could be.

Rob Collie (00:41:31): It would be a very difficult thing to advertise. It's like, Hey, but-

Philip Trick (00:41:35): You have array functions. I have a solution.

Rob Collie (00:41:38): Yeah. If it was an efficient market, you could just connect with those places and do this. Oh my gosh.

Philip Trick (00:41:43): Oh yeah.

Rob Collie (00:41:44): Here's an example. So a member of my family who is taking a medication, and they take it twice a day. And there's not a lot of guidance on how much to take, almost like this titration process, like how do you feel? Do you feel okay? Take too little, you feel crappy in one way, take too much, and you feel crappy in another way. Okay. So it turns out that the half-life of this medication is important. You can't just say, Hey, I take two doses a day. The question is, what blood level in your system corresponds to feeling a certain way. So the only way to really model this is to model this series.

Rob Collie (00:42:25): In theory, let's say you've been taking this for 50 days, twice a day. There's a 100 doses, and the very first dose you took is mostly gone due to exponential decay, but it's still there a little bit. And so all 50 of these, the rate of decay is the same, but you have all of these individual decay curves that then superimposed to create one, like, here's what's in your blood right now. It's crazy how it builds up and slows down and everything. And so you couldn't just model how you felt versus the dosage that you took recently. The real thing you needed to get to was, what's the blood level that feels right. And then back into what dosage stabilizes as close as possible to that. In Excel, that is a very difficult problem.

Philip Trick (00:43:10): Oh, it's a nightmare.

Rob Collie (00:43:11): Yeah, because it's a rectangle that's getting bigger on both dimensions at all times.

Philip Trick (00:43:16): Right.

Rob Collie (00:43:16): There's more doses being added, so those you're getting more rows. And you're getting further and further into this life, more days, more hours, going to the right from each dose.

Philip Trick (00:43:30): Yeah. Right. In the actuarial sciences, we intelligently call that a data triangle.

Rob Collie (00:43:34): Ooh, okay. Yeah. All right. That makes sense.

Philip Trick (00:43:36): You know, your oldest dose has 10 years of life on it-

Rob Collie (00:43:40): Yeah.

Philip Trick (00:43:40): ...your oldest exposure period. Your newest has one. In essence, you have three different trend periods. You have your calendar trend period, you have your dosage trend period. And then you have your age trend period.

Rob Collie (00:43:53): Mm-hmm (affirmative). Yeah. And so I have a Power BI model now, that is named Infinite Decay. Infinite meaning I don't need to be constantly adjusting the size of my ranges, and my references and all that kind of stuff. And it's just so damn elegant.

Philip Trick (00:44:10): Isn't it?

Rob Collie (00:44:12): It is so cool. And we actually solved it, right. We found the blood level range that worked. And I cannot imagine we would've ever cracked that riddle without the X functions. Thank you. X functions.

Philip Trick (00:44:24): The X functions are amazing.

Rob Collie (00:44:26): They put the X in DAX.

Philip Trick (00:44:30): They do. I've got one similar to that, but much darker life expectancy.

Rob Collie (00:44:34): Oh, here we go.

Philip Trick (00:44:34): Plotting out the expected life of people. It even takes advantage of Power Query, going to the CDC and getting the most up to date version of the table. I can plug in somebody's death and project exactly how long they should have lived.

Rob Collie (00:44:48): Is this kind of like the birthday problem? Like how many times does he avoid getting hit by a truck?

Philip Trick (00:44:53): Kind of. Yeah.

Rob Collie (00:44:54): That sounds kind of grim. He's going to cross the street 5,000 times in his life.

Philip Trick (00:45:02): Right. And you know, each time there's some probability that he doesn't make it.

Rob Collie (00:45:06): Non-zero. That's right.

Philip Trick (00:45:07): Yeah. Some non-zero probability. And each year your salary goes up, so it's not like you can just, all right, he survives for 30 years multiplied by his average salary. It's a changing probability distribution.

Rob Collie (00:45:20): Yeah. I like that salary goes up every year. That sounds good. It turns out if you play the start-up game at all, it's not always true. There's a very, very sharp inflection point down, at one point, in my career. We're doing okay now, but-

Philip Trick (00:45:34): I feel you. It takes some time to get off that bump.

Rob Collie (00:45:36): Yeah. I think there was some heteroscedasticity in my... did I even say it right?

Philip Trick (00:45:41): You did.

Rob Collie (00:45:42): I still don't really know what it means. Okay. So the X functions.

Philip Trick (00:45:44): The X functions.

Rob Collie (00:45:46): Yes, absolutely gorgeous.

Philip Trick (00:45:51): Magical.

Rob Collie (00:45:51): Yeah. We could just sing love songs to the X functions. And oh my God, PRODUCTX.

Philip Trick (00:45:58): PRODUCTX, SUMX, AVERAGEX, all incredible. I just found out recently that you can write a whole measure inside the expression section of the Xs, and have just gone wild. It's just gotten out of hand. I might need to stop myself.

Rob Collie (00:46:14): Yeah. You might need an intervention. Now, the reason I like PRODUCTX so much is because we didn't have it at the beginning, and doing these logarithmic transforms to turn it into a SUMX was really awful. Somewhere, there's a bunch of pages of notes where I was doing algebra, sophisticated algebra, there was factoring going on, polynomial factoring was going on, so that I could turn this thing. Oh. And then PRODUCTX came along and I could just throw all those notes out.

Philip Trick (00:46:42): Oh, thank God. PRODUCTX is second behind SUMX. Probably my most frequently, actuarily used ones.

Rob Collie (00:46:48): I bet.

Philip Trick (00:46:48): Because we often talk about things as multiplicative development.

Rob Collie (00:46:52): Yeah. Yeah.

Philip Trick (00:46:53): And so, we have a factor how much did things change between time one and two, time two and three. Multiply it together to get your time to infinite.

Rob Collie (00:47:01): Yeah. Yeah. Okay. Well, some other things we wanted talk about. Let me just cue up a softball. Do you ever try to get people, like youngsters, interested in actuarial sciences? Do you ever go out there? That seems like it'd be really difficult. Just a softball across the plate.

Philip Trick (00:47:18): So I do a lot of actuarial outreach.

Rob Collie (00:47:20): Actuarial outreach.

Philip Trick (00:47:22): I volunteer for the Casualty Actuarial Society, which is the actuarial group that I'm a member of. I'm on their webinar team and I've worked with Angel to promote Power BI within, but I've also gone to high schools and talked to college students to promote pursuing this actuarial career path. Because it's a STEM path, it's a hugely in-demand position, it's constantly rated among the top five or 10 careers you can have. But it is very mathematically challenging and it's not very well known. It is very boring, like nobody wants to talk about spreadsheets all day.

Rob Collie (00:47:58): Yeah.

Philip Trick (00:47:58): I'm competing against stock brokers who have Wolf of Wall Street to reference. There's Along Came Polly with Ben Stiller and Jennifer Aniston, but that doesn't feel like something that you want to... like, Hey, super risk-averse and terrified of everything out there because you know that anything can kill you at any time.

Philip Trick (00:48:16): But Fight Club was one of those fun references. It's got kind of a dark edge to it. I use a particular clip, from where Edward Norton is talking about this car that has blown up-

Rob Collie (00:48:27): Uh-huh (affirmative).

Philip Trick (00:48:28): ...it's killed the family inside it. It's super dark, and he's talking to this lady on a plane about, we look to see what the chances of this happening are. What the average out-of-court settlement is, we figure out how much that's going to cost us, and then if it's more than the cost of the recall, well, we don't do one.

Rob Collie (00:48:46): I remember this scene, basically word for word. Take the number of vehicles on the road of this model, A, multiply it by the average rate of failure, B, and then times the average cost of an out-of-court settlement, C. A times B times C equals X. If X is less than the price of a recall, we don't do one.

Philip Trick (00:49:10): That is amazing. And so I've taken that clip and I've overlaid the actuarial terms that we use for these things. That's what we call a frequency-severity model. It's one of the most commonly used models in estimating costs. As an expert witness, I've been involved in cases where we do do that type of analysis to say, all right, what was the probability of this happening? What is the expected cost of this in terms of life, limb, property? Try to put a value on that, and then turn it into a suit.

Rob Collie (00:49:42): To close the loop at the end, she goes, which car company do you work for again? He goes, a major one.

Philip Trick (00:49:47): A major one.

Rob Collie (00:49:52): I guess it is really dark and grim, however, there is something to it, I suppose, which is, they don't have infinite resources. They have to prioritize the problems they're going to solve, right?

Philip Trick (00:50:02): Right.

Rob Collie (00:50:03): You'd want it to be prioritized based on the pure number of fatalities, but the amount of negligence that the jury is willing to find, or the amount of leverage that the victim has in proving that it was your fault, drives the cost of a settlement up. By doing it in dollars, you are prioritizing based on your perceived negligence. It's a weird metric to optimize around, probably not the one we would choose in a utopia.

Philip Trick (00:50:32): True. It's a challenging problem. On the actuarial side, you often have people on both sides presenting, here's what we think the value is, here's what you think the value is and let's negotiate over it.

Rob Collie (00:50:42): Yeah. My spreadsheet can beat up your spreadsheet in court.

Philip Trick (00:50:46): That is a fun conversation.

Rob Collie (00:50:48): Yeah. I bet. Like, yo look at this spreadsheet, he doesn't even know XLOOKUP. Your honor, I object to opposition council's terrible use of the LOOKUP function.

Philip Trick (00:51:02): This guy's using VLOOKUP, I use INDEX MATCH. Clearly superior.

Rob Collie (00:51:05): Yeah. We need a third expert witness that comes in and testifies to the quality of the spreadsheets. That would be a really funny movie watched by 10 people. We could do it, right?

Philip Trick (00:51:18): Absolutely.

Rob Collie (00:51:19): Jumping around. When you went to ModelOff, was that in New York?

Philip Trick (00:51:22): The one that I went to was in London.

Rob Collie (00:51:24): Wow. So you went across the pond for this. So did you take your own keyboard with you?

Philip Trick (00:51:29): I took my own laptop.

Rob Collie (00:51:31): Your own laptop. Okay. And that's allowed, not scanning for cheats?

Philip Trick (00:51:35): Yes. So the contest was very open in terms of, you could use whatever you wanted. But there are interesting problems, stuff like solving wordsearch puzzles, solving for these airline optimization problems. Things that you probably don't have things necessarily pre-built for.

Rob Collie (00:51:54): Sure.

Philip Trick (00:51:54): Or even if you do, they're going to be generic enough that it's your toolkit. And the kind of competitors... there was one guy that did it all in C#, he didn't even use Excel.

Rob Collie (00:52:03): That was considered okay. It's ModelOff, it's not SpreadsheetOff.

Philip Trick (00:52:07): Right. The expectation was that it would be done in Excel, and at the finals there was actually a dashboard presentation part of the competition, where you had to make it look nice and feel presentable and pretty. So it had to be in Excel at that point, but otherwise it was just, can you answer the question-

Rob Collie (00:52:23): Interesting.

Philip Trick (00:52:24): ...quickly and accurately?

Rob Collie (00:52:25): So you take your own laptop with you. Talk to me about the keyboard on this laptop. Anything special about it? Does it have a number pad? Are all the keys intact or any of them removed?

Philip Trick (00:52:36): All the keys are intact except for F1.

Rob Collie (00:52:39): Okay.

Philip Trick (00:52:40): Obviously.

Rob Collie (00:52:42): Oh, duh.

Philip Trick (00:52:42): You hit F2 to edit that cell, you fat finger F1 and you wait for the help to open, just you can close it.

Rob Collie (00:52:48): Yep. So you pry off the F1.

Philip Trick (00:52:50): You pop off the F1, you put the sticker over it or something. My laptop is massive. It's not something you put on your lap. It hardly qualifies as a laptop, I think it's 19 inches.

Rob Collie (00:53:02): Okay.

Philip Trick (00:53:03): I had to get a bag special ordered to carry it.

Rob Collie (00:53:04): Okay.

Philip Trick (00:53:04): Full number pad.

Rob Collie (00:53:05): Full pad. Yep. Yep. Yeah.

Philip Trick (00:53:07): Full keyboard.

Rob Collie (00:53:08): Yeah. This is a beastly machine. Yeah.

Philip Trick (00:53:12): Yeah. Something that can run MATLAB and do simulations for me while I'm on the road.

Rob Collie (00:53:15): That's so awesome. There are other keys that other people remove from their keyboards for ModelOff as well. But of course they remove these for their business lives too. But in particular, when you're being timed, what are the other keys that are frequently removed? F1 is by far the most common.

Philip Trick (00:53:36): The most common one, like even on the golden trophy had the F1 popped off of it.

Rob Collie (00:53:40): Awesome. Awesome. That's so cool. But what are the others? Num Lock or something? I forget what the... I think there's like three keys that people like to remove.

Philip Trick (00:53:49): I would imagine that Scroll Lock and Num Lock would be kind of tops on that.

Rob Collie (00:53:53): Yeah.

Philip Trick (00:53:55): But I don't know. I've never popped off any other keys. Just one of the benefits of having the large keyboard is I'm pretty far away from the Num Lock and Scroll Lock keys.

Rob Collie (00:54:02): That's right. That's right. Yeah. What's the mean time to removal of the F1 key on a new keyboard in your possession? How long does it take?

Philip Trick (00:54:10): I don't know. I haven't removed one in about five years now.

Rob Collie (00:54:13): Really? That is amazing. CPUs haven't been getting faster, as fast as they used to.

Philip Trick (00:54:19): That's true. And I have the same old mechanical keyboard I've had for, I don't know, it's awfully dirty.

Rob Collie (00:54:24): I see. But the laptop hasn't been replaced in five years, either.

Philip Trick (00:54:27): The laptop is pretty old at this point.

Rob Collie (00:54:29): Yeah. Might be time for one of those gaming rigs, like you need some liquid cooling, some overclocking.

Philip Trick (00:54:35): I just might.

Rob Collie (00:54:36): That's something we did find by the way, in the end. To sort of close the loop on that old story with the four CPUs, we eventually gave up on the whole concept of server architecture servers. We had these things custom made, they were PC motherboards with overclocked CPUs. We had them overclocked to the highest speed that we could do, while still keeping them air cooled because the co-location server room didn't allow us to bring liquid cooled in there. They didn't want to deal with potential catastrophic leaks of a coolant system. So we cranked it as high up as we could. And we had to get them custom made in Kirkland, Washington, near Microsoft-

Philip Trick (00:55:21): Yeah.

Rob Collie (00:55:21): ...and then shipped to the co-located server center. And that overclocking basically returned a completely linear speed increase, 40% overclock was 40% speed increase in the calculations running. Because similarly to the multi-threading thing that we were talking about earlier with Excel... and I know that DAX has been optimized to death since the last time I was actually really caring about performance in it. But same thing with DAX, right? A lot of DAX goes single-threaded, and so no matter how many cores you throw at it, it's not going to go any faster when you're stuck in the formula engine. Basically we needed each core to be as fast as possible. Sharing it across cores, eh, not as useful. And Power Query, again, I don't know much about the performance characteristics, but Power Query, I think is very linear in a lot of ways as well.

Philip Trick (00:56:09): From what I understand about Power Query, it is 100% linear.

Rob Collie (00:56:14): It's a difficult problem to parallelize. I'm sure there's things that they can do, it's always incremental improvements. There's these other cases we haven't optimized yet or could never imagine optimizing. How responsive are these... first of all, I think it's hilarious, in a way, that there is an actuarial outreach. It almost sounds like some do-gooder nonprofit, right? There's people starving, we need to feed them. You wouldn't think of the actuarial community as something that required a sympathetic treatment, but like you said, you could imagine some of the lampoonish slogans; not quite as boring as accounting, but twice as hard.

Philip Trick (00:56:55): Yeah. Our jokes almost all come at the expense of accounting.

Rob Collie (00:56:59): That's the only one you can pick on.

Philip Trick (00:57:00): Right.

Rob Collie (00:57:03): I do find the actuarial stuff to be quite a bit more interesting. You've got to get past the surface of it to find out that it actually is very interesting. So how responsive are these students to the Fight Club clip? Most of them, no, none of them were even born when that movie came out. What's the response?

Philip Trick (00:57:19): My general response is silence.

Rob Collie (00:57:24): Yeah. That's when you know, you have them. Yeah.

Philip Trick (00:57:24): Exactly. You're in a stats class typically, you're talking to a room of 20 or 30 students there on Career Day, and they all look up at you with blank eyes, like, all right, anybody go into like finance? And you get a couple of hands. What do you want to do in finance? Oh, you want to do stock brokerage? Well we do that level of math, but in a different way. While I don't ever really directly see the results of my outreach, the number of actuarial programs at colleges, since I started... when I went to school, there were three colleges with actuarial programs. And since then, it's kind of exploded, there are now dozens upon dozens of colleges with actuarial dedicated degrees.

Philip Trick (00:58:03): So it's definitely growing. And now our profession is facing kind of a new problem from it, in that, before you didn't ever start your exam process until you were already in the industry. But now we've got kids coming out of college with 3, 4, 5 exams and zero experience. So it's been a paradigm shift in terms of all right, well, now we're hiring people that know a lot of the math, but know absolutely no ways to apply it.

Rob Collie (00:58:31): And in a similar vein, the whole work-from-home revolution, it's one thing for people who used to work in an office, who cut their teeth working in an office, but it's another thing, I think altogether, to graduate from college and go straight into a work-from-home environment. What kind of discipline... I'm trying to imagine myself in '96 going to work at Microsoft, but remotely. What a flame out that would've been for me. I needed that regimen of people expecting me to be there, the feeling of guilt or shame or nervousness of not showing up until 10:30 or 11:00 or whatever. I needed that pressure. That shaped me and put me in a place where I could work from home. I don't think that I could have gone straight to it.

Rob Collie (00:59:14): So yeah, the changing dynamics of... I know this is completely different, but it's just another example of, how do you grow and adapt to the nature of work changing, and candidates changing, and all of that? It's pretty interesting. I mean, P3 Adaptive's strategy for that particular problem, as well, we only hire people with business experience anyway. It's not that we have anything against recent college graduates, it's just that experience component is a must.

Philip Trick (00:59:38): Yeah, absolutely. It's pretty critical when you're going in to consult on business matters-

Rob Collie (00:59:43): Yeah.

Philip Trick (00:59:43): ...and being able to interact, just having basic client interaction capabilities.

Rob Collie (00:59:48): So in the fantasy football world, Adam Harstad, who has been on our show twice... there's sort of a long held conception that an NFL player, their career follows a curve. There's a breaking-in, learning the league, so they start low in terms of output. They figure things out, they rise in terms of their effectiveness and they reach a peak, but it's an aging thing. Not just aging, but also like mileage on their bodies, because they're just getting absolutely wrecked.

Philip Trick (01:00:18): Hammered.

Rob Collie (01:00:18): And then they enter a decline, and eventually they decline so much that they're no longer above replacement level for the league. And so then they're out of the league. Okay. So Adam didn't believe that. Or at least he questioned it, and he instead started modeling NFL player careers using morbidity models. And it turns out that, that approach models what happens much, much, much more effectively than the previous hypothesis of this arc. They generally don't fall off that much, but then suddenly all at once they're gone.

Philip Trick (01:00:52): That's interesting.

Rob Collie (01:00:54): Yeah. He has a lot of articles on that. And Adam is a deeply statistical and thoughtful thinker, which is why we've had him. He doesn't work in business consulting at all, he's never touched DAX, and yet we've had him on the show twice just to talk about these sorts of things. And just like you, by the way, he slings some serious vocabulary. Whether you're into football or not, his approach to things is one that is just fun to read.

Philip Trick (01:01:20): So the morbidity model makes a lot of sense. You're born into your NFL career. It doesn't necessarily preclude that same-age curve that we think of, but thinking of it from a different perspective, right?

Rob Collie (01:01:31): Yeah.

Philip Trick (01:01:31): Instead of maybe growing so much and then declining, it's more a matter of, at what point are you too injured to continue? Because football, I feel, is a little bit... I always think of baseball statistics, [inaudible 01:01:43] replacement, stuff like that, which is much smoother-

Rob Collie (01:01:46): Much.

Philip Trick (01:01:46): ...but baseball's also not... you're you're not getting hit by a 300-pound guy running at you.

Rob Collie (01:01:50): That's right.

Philip Trick (01:01:50): So football, thinking kind of mortality-wise, it's either you're in or you're not.

Rob Collie (01:01:57): Yeah.

Philip Trick (01:01:57): And you can look at players like-

Rob Collie (01:01:58): Brady.

Philip Trick (01:01:59): Yeah. He played it till like, what 45?

Rob Collie (01:02:02): He's still playing.

Philip Trick (01:02:03): He's still playing.

Rob Collie (01:02:05): He's still playing. He's-

Philip Trick (01:02:07): I don't follow football very much

Rob Collie (01:02:07): ...he's with The Bucks now. And he's still playing. Now, he retired for a month in the off-season, but now he is back. This is a guy who is almost my age. I put heel supports in my running shoes today to try to protect my achilles tendon from tendonitis, running relatively slowly on a treadmill. This guy's a year or two younger than me, and he's still out there.

Philip Trick (01:02:36): Performing at almost the same level he always has.

Rob Collie (01:02:39): Yeah.

Philip Trick (01:02:40): So, that kind of makes sense. It's a different type of cliff. It's not a, you get a little bit worse every year. I mean, maybe it is for some players, but it's either you're in and capable of playing at your level.

Rob Collie (01:02:53): And this is getting probably a little too much into the football domain specifically, but running backs in particular are a very fascinating example. It's kind of a problem for their labor market too, because basically almost no running back is ever worth paying their second contract. They sacrifice their bodies so intensively, in such a short period of time that, when someone gets up around 28 or 29 years old, we're talking about them like they're old. Oh, this person isn't 30 yet, and we're talking about them, like they're getting a little long in the tooth. Like it's-

Philip Trick (01:03:29): Yeah. Like, all right, it's time for him to retire.

Rob Collie (01:03:31): Yeah, kind of unbelievable. So yeah, the way that the NFL wage scale works is that, you get your rookie contract, and your rookie contract, when you're brand new in the league, is somewhat like wage-controlled. There's no negotiating a higher number on your rookie contract. It's determined by where you were drafted. There's a formula. But then the second contract is when you actually have some market pricing. But that first contract can last four years.

Rob Collie (01:03:57): They've been talking about things like, should running backs have a different rookie wage scale? That'll of course backfire, because then teams will just take them later.

Philip Trick (01:04:05): Right.

Rob Collie (01:04:07): What are you going to do? How are you going to get more running back talent? Right now, if you're coming up in high school and you're playing football, which for many reasons I wouldn't be, you wouldn't choose to be a running back, you'd want to go to some other position. If you thought you had a shot at playing professionally, you'd want to play almost any other position,

Philip Trick (01:04:25): Longevity, personal health, all the repercussions of pounding your knees and just ramming yourself into a line of people all bigger than you.

Rob Collie (01:04:36): Seeing how we've been in a dark morbid zone multiple times, this won't be anything new. The players in football who suffer the most after football, with CTE with their brains, are actually not usually the running backs. It's actually the people who will play along the line, because every single play they collide. Football fans welcome the sound of helmets colliding. Like it's sick. They've found that it's not concussions that are the problem, it's all of the micro-collisions.

Philip Trick (01:05:09): Yeah.

Rob Collie (01:05:09): None of them are micro.

Philip Trick (01:05:10): Relatively micro.

Rob Collie (01:05:11): Yeah. Sub-concussive collisions. But you have 60 of them a game, in a year of your life, you could suffer 1000 collisions. Any one of which would lay me out. These aren't... that's why kind of funny saying micro, right? It's sub-concussive.

Philip Trick (01:05:34): Four downs, they get up and they do it back to back. And then they take a five minute break and come out and do it again.

Rob Collie (01:05:41): There's some modeling to be done there as well.

Philip Trick (01:05:43): In that area, there's actually been some fantastic data progress made, in predicting more severe injuries from relatively simple information. I went to a sports analytics conference here in Dallas, I guess it was about three or four years ago, and this group of doctors and this statistician found that, with a simple pressure plate... you just stand on it and it measures how your feet are balancing, and if you would jump, the way that you would jump and land would change, based upon how you felt about just different parts of your body. And in conjunction with what type of sport you were playing and where you were at, they could identify if you were favoring your elbow, even just the tiniest bit, by the way you were jumping and landing. And they're using it in colleges right now, in some colleges, to try to limit these injuries before they happen. Before your ACL tears, when your ACL is just a little loose, like you feel okay, you still run all right, but it's that catastrophic moment that is the big problem. And using data to do this.

Rob Collie (01:06:48): I've read a lot about this. And even in a way I've even brought DAX to bear on this problem for an NFL franchise. Now I did it for free. It wasn't a paid deal because I was doing it to help a couple people who were being paid by the NFL franchise. I just couldn't resist. They were using GPS and accelerometer data from practices, and so I got to implement a custom calendar. We usually were talking about the 4-4-5 custom calendar or something like that, four weeks, four weeks, five weeks. I got to implement a Time Intelligence custom calendar based on the schedule of the NFL season, including number of days between practices, and did you have a Thursday game, which changes the whole schedule and everything.

Rob Collie (01:07:34): In theory, you're supposed to be able to limit these soft tissue injuries. That you're right, they're... again, it's that stochastic thing. It's not like an ACL is degrading by a linear percentage.

Philip Trick (01:07:46): Right.

Rob Collie (01:07:47): It's either there, or it's not. When I was not quite 40, I ruptured my patella tendon at a trampoline park. Borderline middle-aged man ruined many birthday parties. That's like the [inaudible 01:08:02] article headline, Middle-aged Man Injures Self, Ruins Five Year Old's Birthday Party. Because they took everybody out of there while I just laid there. You want to know what it feels like to lay on a trampoline while everyone else is gone, except for the people who are shoving a clipboard in your face saying, we need you to sign this-

Philip Trick (01:08:24): Release of liability.

Rob Collie (01:08:25): As if I actually have to. I was jumping around, my wife and I. This was my wife's birthday, she wanted to go to the trampoline park for her birthday. So I took her and all was well, we'd done all kinds of things, hadn't gotten injured. I'd been doing this one sequence of hops that it was kind of fun. I said, Hey, you know what, we're going to leave. We were leaving, and I said, Hey, let me do it one more time, and you record it for the kids to see what we were doing. And I swear, that's what happened. So on the very last one, I turned around to go back in to do this, I bounce up and, in the air, I feel the knee just kind of unbuckle. And so we have it on video.

Rob Collie (01:09:04): It took me a long time thinking about it, but when I was about 10 years old, I was playing soccer one time, and I kicked the ball at the same time that an opposing player kicked it the opposite direction. We both kicked it very hard and it hurt my knee, and I was unable to run for a month after that. But we never took me to be looked at or anything like that. It was just, rub some dirt on it kid, that kind of thing. I think all these years later, I had some sort of like micro-tear, or some scar tissue or something still left over in that patella tendon, and that was the weak point that gave way. And now if you put me on a pressure plate, you will see that I absolutely favor that leg, because that leg is got considerably less muscle mass. It's a life-changing, permanent life-changing event. Anyway, these things happen.

Philip Trick (01:09:53): They do. And that technology, we're talking about it from a sport's perspective, but I've done, probably the most common insurance I've worked with is Workers Comp insurance.

Rob Collie (01:10:02): Mm-hmm (affirmative).

Philip Trick (01:10:02): It's long-term injuries. And this type of analysis, right now it's focused on sports, but imagine being able to diagnose back injuries before they become bad, and being able to rotate your workers in a better way and improve their quality life while improving the bottom line.

Rob Collie (01:10:21): Yeah. I like that kind of stuff. In the end, you can never perfectly predict these things. You would never have predicted my knee injury, unless you had known about the leftover scar tissue or whatever, the micro-tears or whatever were still left there.

Rob Collie (01:10:34): You can use it to enact broader policies that do provide a statistical lift across the board. There's no perfect story ever for each individual and each individual set of circumstances, but on the whole, they can't do this particular job all the time and not wear out. Like running backs, for instance, they should probably be humane. Let the running backs play another position, like three plays a game they get to go and do something different. All right, is there anything else that we didn't talk about that you think we should?

Philip Trick (01:11:04): Maybe what brought me to P3. So first of all, I found y'all like Reddit. You had a job posting on Reddit.

Rob Collie (01:11:09): Really?

Philip Trick (01:11:10): In my infinite boredom, I scroll the Excel boards and the Power BI boards handing out free advice. I've been called a Power Query evangelist.

Rob Collie (01:11:20): Ahhh.

Philip Trick (01:11:21): In the past.

Rob Collie (01:11:21): Okay.

Philip Trick (01:11:21): But I found y'all on Reddit. And at that point I had been trying to push Power BI into the insurance sphere without hardly any success. I had a whole webinar series on how to take bad data. And this other actuary had, here's a really bad, ugly data set, and here are a whole bunch of ways to clean it up, here are the problems that we encounter. And he sent it out to about 150 actuaries as a survey to see what they could identify, and not a single one of them used Power BI to query it. And so I sent him back a Power BI file just with 10 tree maps, like, all right, click on this one, there's that problem. Click on this one, there's that problem. He's like, wow, this is incredible, why don't we know about this?

Rob Collie (01:12:03): Like how many years did you spend to make this? Well not many, 20 minutes [inaudible 01:12:09]. That's my favorite reaction is, why haven't I been told about this.

Philip Trick (01:12:13): Right. And so, it's catching steam in the actuarial sphere, but it's still... there's a really big jump from the Excel formulas we write to DAX. It's a very different perspective. I remember the first DAX formula I wrote, just SUM of some amount, and then plotting it and having it do all of those IFs. And I was like, okay, where's the IF? I just did a SUMIFs, but I don't see the IFs.

Rob Collie (01:12:38): Yeah. Yeah.

Philip Trick (01:12:39): The whole jumping from... our formulas in Excel aren't easy to start with, and now translating to DAX, it's taken me a number of years to finally get to where I'm getting my stuff in DAX the way that I want it to be. But that outreach into the actuarial sphere itself has been fantastic. And P3 was like, Hey, I've been learning all this Power BI stuff and not getting to use it. This is awesome.

Rob Collie (01:13:02): No, that's really cool. I'm in the Excel Reddit, I'm in the Accounting Reddit. I know nothing about accounting, but I hang out there and just read. The humor in the Accounting Reddit is top notch.

Philip Trick (01:13:13): It is that perfect level of dry. It's great.

Rob Collie (01:13:17): Are you familiar with the user called Sneezes?

Philip Trick (01:13:19): Yes, I am familiar with-

Rob Collie (01:13:21): He's the one that tries to get people pumped up on Mondays. It's so funny. I love Reddit. It's my favorite social network by far. All right. Well thank you very much. It's a pleasure.

Philip Trick (01:13:33): Lots of fun. And hopefully something good comes out of this.

Rob Collie (01:13:35): That's right.

Speaker 3 (01:13:36): Thanks for listening to The Raw Data by P3 Adaptive podcast. Let the experts at P3 Adaptive help your business. Just go to P3adaptive.com. Have a data day.