The Unofficial Shopify Podcast

Unlocking the Power of Customer Lifetime Value

Episode Summary

w/ Daniel McCarthy

Episode Notes

Ecom Businesses often rely on Customer Lifetime Value (CLTV) to stay profitable, but the correct formula can be hard to calculate. Fear not! Mr. Daniel McCarthy is here with his expertise in statistical methodology and contemporary empirical marketing problems - which he published extensively on – ready to share why CLTV needs special attention from businesses and how they should go about doing it right. Don't miss this opportunity of a lifetime by learning more from an accomplished academic who's taken home numerous awards for research excellence.

Show Links


Never miss an episode

Help the show

What's Kurt up to?

Episode Transcription

The Unofficial Shopify Podcast
Daniel McCarthy

Kurt Elster: Hello, my friends. Yet again we’re here thinking about making data-driven decisions. Where do we draw the line with our data-driven decisions? You know, and I know I should get eight hours of sleep a night, and I know I need to get up at 6:00 AM. Is determining my bedtime off that… Is that a data-driven decision? I know what my budget is. Is that data? Where are our data-driven decisions? How deep do we go with data-driven decisions, right?

And I noticed in the first quarter of the year, that’s when people get really interested in going, “We gotta deep dive into our analytics. We’re gonna use the data. We’re gonna figure out our goals and agenda for the year.” I actually think that’s quite admirable. However, I think it’s also analytics, statistics, data-driven decisions, KPIs, all that stuff is one of those things where you can look at it superficially. You can know enough to be dangerous. And then you hit a point where you go, “Man, I don’t know enough. I know enough to know I don’t know what I’m doing.” And I’m starting to fear that’s where I’m in. We’ll call it the trough of disillusionment.

And so, I asked on Twitter. I was thinking about customer lifetime value. And I wanted to calculate it easily, correctly, consistently, and so I thought, “This is an easy question.” And I asked on Twitter and that was my first mistake. Ask on Twitter. But I had some folks who were like, “No, no. It’s not as simple as you think and you’re gonna screw it up.” And I said, “Wait a second. If that’s the case, are statisticians gatekeeping how to calculate CLTV?” And someone said, “Maybe, but there’s a guy who does know how to do it and you should absolutely have him on your show.”

I thought, “That sounds good.” And then immediately I had a good friend who you’ve heard on the show, Andy Bedell, sees the exchange, texts me and goes, “You gotta talk to this guy. He’s the man when it comes to CLTV.” I’m like, “Hold on. Is there an entire sub niche in eCommerce devoted to CLTV?” And the answer, maybe. We’re gonna find out. Because joining me today is Daniel McCarthy who’s an Assistant Professor of Marketing at Emory University School of Business. His research specialty there is the application of leading edge statistical methodology to contemporary empirical marketing problems. Oh my. That is quite the mouthful. We got the guy who’s gonna break this down for us.

But first, I’m your host, Kurt Elster.

Ezra Firestone Sound Board Clip: Tech Nasty!

Kurt Elster: And this is The Unofficial Shopify Podcast.

Sound Board:

Kurt Elster: Daniel McCarthy, welcome. Thank you so much for being here.

Daniel McCarthy: Thanks so much for having me and for the very, very, very warm welcome.

Kurt Elster: In your own words, why should we listen to you?

Daniel McCarthy: I think there are some people who would say, “You know, okay, you’re doing all this fancy math. Blah, blah, blah. Maybe I’m 30% off but as long as I’m directionally correct then it doesn’t really matter.” And I think there is some conception that as long as you’re kind of good enough, that the mistakes can’t be that bad, can they? And I would argue that yes, for one, you could be off by factors of three or more in terms of your estimates at how much your customers are worth. Customer lifetime value in some sense literally will drive the overall corporate valuation of your firm and it’s gonna do it through the unit economics, basically through your ability to kind of carve out a path to profitability if you’re not profitable yet, and to grow quickly, and to sustain revenue growth.

So, it is absolutely one of the key drivers of that valuation and I think it’s not well understood because finance hasn’t really been using it as much as they should, or looking at it in the right way, so I think with the right kind of conceptual framework for it it’s pretty easy to see that it is really important, and ultimately it can drive whether your firm is gonna do well or whether it’s gonna crash and burn. So, you know, it's a pretty important thing for someone running a business.

Kurt Elster: Is it more important in eCommerce, where you’re selling a physical good and that long-term cash flow and inventory forecasting is really tough? And so, having that additional input of like, “Okay, we know the value of a customer, we know acquisition costs,” I would have to imagine that that helps give you a more coherent picture.

Daniel McCarthy: Yes. Yes. I think you’re alluding to one of the very valuable use cases, which is FP&A, financial playing and analysis, that you can kind of see where the business is heading and what your inventory needs might be. Ultimately, what your revenue and profit’s gonna be next year, the year after, the year after that, and ultimately, again, it’s kind of this simple accounting identity that all the revenue and profit that your firm generates is coming from your customers. I think especially… So, you think, “Okay. Well, in some sense every business has customers that are placing orders. Maybe some have bigger customers and it’s lumpier, others not.” But I think what makes eCommerce a little bit unique is that you have really good visibility into what your customers are doing.

So, yeah, you mentioned Nike, and I think the reason why that is somewhat surprising as an acquirer for Zodiac is historically they sell most of their product through Foot Locker and other shoe stores, and so they don’t actually even really see end customer demand, and so they may acknowledge the importance of customer lifetime value and getting those people to repeat buy, but if they can’t see the underlying data then they can’t really do this sort of a thing. That’s not an issue to nearly the same extent in an eCommerce setting.

In eCommerce, you kind of naturally get a pretty good transaction log for free. There’s still issues as I’m sure you’re acutely aware, but yeah, there’s just less issues, and so it becomes a lot easier to be able to kind of do this exercise and get accurate answers for these quantities. I think the other thing is that oftentimes eCommerce businesses are pretty heavy users of paid media, and so they really need to keep track of what customer acquisition cost is and how it varies across their channels, and so some of the inputs are especially important for companies like that relative to say a Hershey’s or some big CPG brand where the nature of their marketing budget is gonna be a little bit different.

Kurt Elster: You brought up like Nike and big CPG brands. If someone’s listening, is there a point where this matters? Should I always be paying attention to this? Essentially when am I big enough to care I think is my question.

Daniel McCarthy: I think you’re always big enough to care, but you do need to be mindful of how some of these measures can evolve across your cohorts, that the customers that you acquire early on may be fundamentally different from the customers that you acquire later. If you’re young, this, again, typically young companies are not profitable yet, you know? Maybe there are some that are. Thankfully, actually after Zodiac I started a second company called Theta, and Theta does nothing but kind of customer-based corporate valuations, so the same sort of exercise that we did at Zodiac except more geared towards kind of the overall health of firms.

And in this setting, that’s kind of exactly what we would focus on. So, yes, it’s always important, but if you’re young it’s more for do I have a clear path to profitability and what is the cleanest trajectory to get there, whereas if you’re mature, in some sense you don’t have to worry as much about business pivots kind of fundamentally changing the economics of your cohorts. It can kind of become more of like a rinse and repeat machine, you know? Where you’re kind of just mechanically looking at those acquisition channels and making sure that you’re kind of allocating your budget optimally.

Kurt Elster: Well, just now you mentioned cohorts, and so I want to get into that too, where a customer is not necessarily a customer. We want to be able to segment and understand different groupings of customers. But all right, let’s start with the big, the elephant in the room. What’s the right model for CLTV? Where am I going wrong here? Surely, it can’t just be like, “I open up my spreadsheet and it’s however many purchases and I average this out,” right? Tell me the most basic way for me to screw this up.

Daniel McCarthy: Well, the first thing is, and not to kind of beat the cohorts again, but you gotta start with a cohort. And it seems like-

Kurt Elster: Oh, okay. So, I’m looking at it the wrong way around.

Daniel McCarthy: Yeah. Typically, you want to take a group of people and you want to see, okay, they were born at this time, and we’re gonna track them through their lifetime, and you can’t really do that unless you have some sort of a acquisition cohort. So, you know, ultimately we’re gonna try and get estimates for every single individual customer’s value, but typically the way that we’re gonna go about it is we’re going to take a whole bunch of people that were kind of acquired at around the same time, and then we’re gonna look at their activity over their life cycle. And so, you need to have kind of a clear birthdate for those people. And again, thankfully in a Shopify setting you’re gonna have that, typically. Whether it’s an email identifier or some sort of more refined identifier that also merges in ship-to and whatever else you might have that helps kind of do identity resolution. But once you’ve kind of identified, all right, these people, they were born at this time, then the next thing is we want to be able to track all of their purchases and then their revenue, and then ultimately the contribution profitability of that cohort over their life cycle.

Kurt Elster: And so, for this, the customer profile within our cohort, we’re looking for… When we say their birthdate, is that when they first discover us, they register on the site, or the first purchase occurs?

Daniel McCarthy: First purchase. Yeah. And this, I think… Yeah, to me I’m not dogmatic about it, but it could sound like tomato-tomato. I would define someone who hasn’t bought yet, like opened up their checkbook and actually put revenue in your pocket, I’d call them a prospect.

Kurt Elster: Okay.

Daniel McCarthy: And certainly it’s extremely important to track prospects and how they move through the acquisition funnel, but to me that’s kind of a conceptually distinct process. So, basically once the customer… The customer is born when they make their very first purchase. And then the key is let’s kind of track their purchases after their point, and then typically as long as the cohort’s not extraordinarily old and everyone has clearly kind of churned out of the cohort, there’s typically some predictive model where you’re going to then say, “All right, this is what they’ve done so far. This is kind of like the trajectory that they’ve been on. And now, based on that, I’m gonna make this prediction of what this customer will do into the future over some horizon.”

Kurt Elster: And so, my cohort is a group of people who make a purchase, because the people who raise their hand and say, “Oh, I’m considering purchasing,” they’ve clicked an ad, they’ve signed up for an email, they have taken some engagement action and we can see that, still prospects. And potentially those are prospects who cost me money if they haven’t made a purchase yet. They make that purchase, they have now opened their wallet. There’s a big difference between saying yeah, sure, I’d buy, and actually doing it. And so, okay, now they’re customers, and when we are looking at this data it sounds like you’re saying you need to look at it as a cohort.

So, it’s like all right, grab a group of people who all bought in the same month is the data I want to look at as opposed to like, “Yeah, we’ll just grab the last two years and lump that together.”

Daniel McCarthy: Yeah. Or, you know, last month, last quarter. It could depend a little bit on the firm. Typically, you want it to be as narrow as possible, but you want to have enough customers in the cohort to… The problem when you go down to individual people is it sounds nice, like I should be able to run a model on every customer individually, but for 40-80% of the customers, they’re gonna make that first purchase and never come back. And so, a lot of the signal that will come from the sort of predictive models that we would use, they’re looking at the patterns kind of across the cohort. They’re saying, “Well, you know, this many people have bought no times yet, or repeat bought zero times yet. This many people bought once. This many people bought twice.” And you can kind of see curves and trends that make it a lot easier.

So, the cohort tends to be a lot more well behaved and predictable than any individual customer within the cohort. And so, typically kind of the first stage, you can think of it as like a first stage model where we’re kind of modeling the cohort’s behavior as a whole, and then there’s a second stage where we’re kind of going within the cohort and saying, “Well, now that I know what the cohort’s gonna do, what are you gonna do?” And that becomes a much… It’s much harder to go off the skids when you do it that way.

Kurt Elster: I’m trying to reduce noise, so essentially what we’re doing here is trying to keep this as statistically significant as possible, and when we make it too small, potentially, our sample size gets too small and it really… Yeah, I can make the prediction, but it’s not useful. It’s not going to be accurate or useful.

Daniel McCarthy: Tremendous noise. Yeah. It’s extremely uncertain what any… We’ll often say I don’t really know what Bob is gonna do, but I know what the Bobs are going to do. And you know, I think there’s a real element of truth to that. I think the other aspect of it is if you think about what you’re gonna do with this information, it’s like, “Okay, everyone has this dream of one-to-one marketing, but ultimately you’re gonna run a campaign and you’re gonna target segments.” And so, immediately you’re back in these groups again, you know? It’s not even like you’ll be targeting every single person and sending them a different message. In theory that’s possible, but yeah, I think ultimately you’re gonna be working at the segment level anyways. And so, you kind of get the best of both worlds. You get much more predictability and you’re working at the level that you’re actually gonna run your campaign.

Kurt Elster: Yeah. Everything realistically is always gonna be one to many, even if I’m grouping it into many groups of many, like say I’ve got several of them, it’s still one to many. I’m not individually cold emailing customers based on this data. And you know, nor should you be. Where do I go from there? We’ve got the cohort. Now what?

Daniel McCarthy: Yeah, so you got the cohort, you’ve got that data on purchase incidents and on revenue. Typically, contribution margin data will sit in a different place. If you have that integrated into your transactional data, then great, but I’m not sure Shopify-

Kurt Elster: What’s contribution margin?

Daniel McCarthy: Yeah, exactly. Yeah, there’s this really meaty question.

Kurt Elster: Truly. I don’t even think I’ve heard the phrase before.

Sound Board: What?!

Daniel McCarthy: We haven’t even started with customer acquisition costs.

Kurt Elster: I know. I know. I already was so naïve when I tweeted like, “Just give me the easy formula.” Like, “Oh, crap.”

Daniel McCarthy: Yeah. A lot of people, they’ll stop at revenue and what I’ll often say is revenue doesn’t put food on the table, you know? You have to pay money to buy that product that people then buy from you, you know? And the key is how much profit are you making after you do all that stuff. And-

Kurt Elster: 100%. It’s such an easy mistake to make in any business, but particularly in eCommerce and retail where we have to resell a good.

Daniel McCarthy: Yeah. And I think the key then becomes okay, what expenses do we include? And maybe for some of our expenses, what proportion of the expense do we include? And what I mean by that is for the former, if you’re selling shirts, well, obviously direct labor and material that sources the shirts, all the fulfillment expense that… all the shipping that you have to pay, and then all the fulfillment expense that you may be subsidizing when customers make purchases if you give them free shipping.

And then there’s things like returns. Returns are a little tricky in the sense that at the time of purchase you won’t know if a return will be made yet, you know? But typically you can infer what the trajectory of your return rate will be, so I’d think of it as like a return curve, like the proportion of the initial sale that I have to give back in returns across my orders is a function of how long it’s been since the purchase was made. And typically, that’s gonna asymptote at some level. Maybe you offer free returns but it’s for a month, so returns can happen after that, but first it’s gonna be a lot less expensive, and people probably won’t be returning past that point. And so, maybe after a month 15% of your product will be returned.

You’re gonna want to make sure that you’re accounting for that at the time that you buy it, because you know, sure, you made 50 bucks in profit off the shirt, but if you know that… I’m gonna give a crazy example. If you know that 50% of that product is gonna be returned to you and you’re gonna have to pay a whole bunch of money to kind of like facilitate the return, then you didn’t make the $50.

Kurt Elster: Well, you’re 100% right, it’s such… It’s an important consideration, especially now. Our Q4 ’22 data for Black Friday says returns hit year-over-year way up over the previous year. Potentially returns were at a record level, which… Interesting. And so, certainly that’s a cost center for a business, and in a calculation like this, I can see where not including that… You can drastically end up overstating CLTV.

Daniel McCarthy: Yeah. And where it can be particularly confusing is if you don’t have some sort of prediction for returns, then what can happen is it looks like your cohorts are getting better over time. But the reason why is because all those old cohorts, all the returns happened already. They got dinged for it. You know, but the young cohorts, they haven’t returned their stuff yet, and so you’re like, “Oh, man. It’s just so much better.” Then you look at it and you’re like, “Wait a minute, because they haven’t returned stuff, but they will.” So, you really want to kind of make sure that you’re properly accounting for that.

The rough rule of thumb, anything that you have to spend the money or you know that you will spend the money as a function of the orders that come in, that that is a variable cost, and contribution profit is just revenue minus all of your variable costs. So, direct labor and materials, obviously returns, fulfillment, payment processing. Typically, there’s also other kind of effectively variable expenses, like you know there’s gonna be some proportion of people that are gonna pick up the phone, they’re gonna call you, and so customer service, you’re gonna want to bake some portion of that in.

And you know, I think there’s kind of the inevitable next question of how far do you go. I tend to be kind of a hard ass with my variable margin, but if you want it to be marketing material, typically you’ll just kind of stop at the immediate direct variable costs. If you really want to use this to run your business, I would further recommend that you incorporate all of your effectively variable overhead expenses. And so, this would be things like accountants, legal expenses, and things like that. You think, “That’s definitely not variable. I would never include something like that.” But yeah, if you look at a company like Microsoft, they’re still spending 10% of their revenue on SG&A, and ultimately there’s some portion of your overhead that’s gonna continue to grow. You know that you’ll still need more lawyers, you’re still gonna need to pay your executives more.

Kurt Elster: Yeah, so you end up… If I use CLTV, or you know, are we going with CLTV or CLV? I’m just used to saying CLTV.

Daniel McCarthy: I like both.

Kurt Elster: Okay. CLTV, the way I’ve traditionally seen it done, it’s just looking at top line revenue. And the danger in that I’m hearing is because it’s revenue, not profit, if we have variable expenses that are going to increase as our order volume increases, then potentially our own success eats us alive. And just it becomes a particular problem in retail and eCommerce with a physical good because of those input costs to make the sale.

And so, all right, we want our cohort, and then we want our CLTV to be contribution profit. All right, I’m with you so far. At what point do I get a number out of this thing?

Daniel McCarthy: We need one more acronym.

Kurt Elster: Okay.

Daniel McCarthy: CAC.

Kurt Elster: Customer acquisition cost.

Daniel McCarthy: And this is the other reason why you really need to pay attention to… You need to kind of do that contribution profit calculation, is because what the heck? You gonna compare revenue to CAC? One’s a top line measure. The other is an expense. And it’s like well, if we just focus on revenue alone, I totally get that. I want to get lifetime revenue and track that across my cohorts. And that’s a useful measure. But as soon as you start that deducting expenses, it’s like why deduct one expense but not the others?

So, yeah, you need to deduct CAC, and that’s just a whole other ball of wax, so I’ll spend-

Kurt Elster: Yeah, that could be its own episode.

Daniel McCarthy: I always joke, and I teach this class called customer lifetime valuation, and I spend one lecture and have a full homework on CAC, and how you define it, and what to include, what not to include, how people cheat, and I’ll often joke at the beginning of class that I could spend a whole semester just on CAC.

Kurt Elster: We traditionally think of CAC as like, “Well, what am I paying to Zuckerberg this month,” right? That’s how we think of it. How am I wrong there?

Daniel McCarthy: Yeah, to some degree for a business like Shopify, to the extent that Facebook is one of the biggest kind of marketing channels, then that wouldn’t be incorrect. But the devil’s in the details, so first you’re gonna have a whole bunch of different channels, and there’s a question of do you want to use this for kind of more reporting purposes, or for tactical customer acquisition purposes? And the reason that distinction’s important is because for the former it may be more helpful to think about your blended or average CAC, and for the latter it may be more important to focus on your marginal CAC, and the main difference between those two is that your average CAC or your blended CAC is… It’s just the total amount that you spent on Zuckerberg, or all those other channels-

Kurt Elster: Yeah. Google. Whatever your PPC ads are.

Daniel McCarthy: TikTok. Yeah.

Kurt Elster: It’s probably gonna be the majority of that.

Daniel McCarthy: Yeah. Google. And then you’re gonna divide that by the number of customers that you acquired. Now, the number of customers you acquired, those customers, they could have been acquired through Facebook, but they could also have been acquired through organic channels. They could have been acquired through some podcast that you did, or an event that you went to, and so each of those different channels will have different costs. And obviously organic… I always say organic with the quotes because truly there is no truly organically acquired customer typically. It is something that you spent that kind of got their attention and brought them in.

So, essentially that average CAC figure, it’s averaging across the CACs of all of those different channels that you had used. And that can be quite different from this other number that the marketing department is gonna be most interested in, which is all right, imagine that I spent another $100 on Facebook. How many customers will I bring in incrementally? And you think, “Well, can’t I just get that from that first number?” And the answer really is no. So, typically if you have some sort of an experimentation platform, or you run some suppression tests, that can help you get an estimate of that marginal CAC. What is the marginal cost of that next acquired customer?

And that’s gonna be extraordinarily important for your business, but it’s just kind of a different number.

Kurt Elster: Okay. So far, CAC sounds like the easiest one to figure out here, at least within the realm of a Shopify store owner.

So, it sounds like that’s all of our prerequisites to accurate measurement?

Daniel McCarthy: Well, we haven’t even touched the predictive model, but…

Kurt Elster: Right. This is looking into the past of what happened. We can also predict, all right, now what’s gonna occur going forward? Oh my… All right. What next? Keep going.

Daniel McCarthy: Yeah, so we’re going through the various stages of hell. Oh, man. Yeah. Getting the bookkeeping right, it seems like the bookkeeping should be the easiest part. Still, it’s been surprising to me how infrequently I’ll see people really grappling with all that.

The good thing is the predictive model in some sense is the easier part either if you’ve contracted out to the right vendor… Obviously, this is something that we do every day all the time. Or you know, you’ve kind of… You know what the right validations are and kind of empirically assess how accurate you are. And I know this goes back to the Twitter thread a little bit that in some sense customer lifetime value… Well, not in some sense. It is a difficult prediction problem. And I think a lot of people, they’ll kind of say, “All right. Well, you know, here’s my CLV formula. Plug and chug. Just get my inputs, get my R, get my M, and let it rip.”

And the problem is it’s not really acknowledging that it’s a prediction, and maybe that formula worked, but maybe it didn’t. And I think a lot of people, they’ll just kind of run the formula, they’ll take out that number and treat it as truth, and they won’t-

Kurt Elster: That’s my issue. When you see especially a predictive one, I look at it and I go, “That sounds… Does it pass the gut check? Sure. I have no idea if this thing is right or not.” And it certainly, if I’m looking at CLTV in Klaviyo, it’s not like this thing is telling me its statistical confidence.

Daniel McCarthy: Yeah. Yeah. Uncertainty is a whole other ball of wax that I’m not even-

Kurt Elster: All right, we save that for the fourth episode. It’s one of those things, and I alluded to it at the start of the episode because I kind of figured this is where it was gonna go, but it’s pulling a thread on a sweater, where it’s like once you scratch that surface of like, “Hey, maybe there’s something here I’m missing,” and then you talk to someone who knows it and it’s just like you’re just peeling back layer upon layer.

The thing I’ve been thinking throughout this is no one likes to be wrong, but there are scenarios where I’m happy to discover I was wrong, right? Where it’s like, “Oh, this is so much more complicated and interesting than you knew.” And that’s how I was wrong about it. All right. Please continue. Where were we?

Daniel McCarthy: Prediction.

Kurt Elster: Prediction. All right.

Daniel McCarthy: Yeah, so you got that data, you gotta run it through the model. Obviously, we have the models that we think work well, and I think at Zodiac, I think by the time of the acquisition we’d run our models on 250 different companies. And at Theta, we’ve run it on probably an additional 150. So, you know, we’ve got a pretty strong prior just based on like, “Well, this worked. This didn’t,” across all those different prior companies. Whatever model you use, I think the yardstick, the report card that you kind of whip out to evaluate is my model getting an A or is it getting a C should be the same. And it’s just some basic things, like let’s remove the final year of our data. Let’s pretend we didn’t see it. Let’s train our model in everything before that point, predict what’s gonna happen in that final year, and then compare how our predictions stack up to what we actually observed.

Kurt Elster: Oh. Very clever.

Daniel McCarthy: Yeah. A holdout validation is what they call it. But specifically, it’s kind of a cross time holdout validation. Some people, they’ll do the opposite. They’ll do an across customer validation where they’ll say, “I’m gonna pretend like I didn’t see the data for 20% of my customers.” And I’m gonna predict that from the other 80%. And that’s a much easier yardstick. So, we typically would not recommend that, because it's so easy to look good on. So yeah, leave out that last year, predict it, see how well the predictions hold up, and then the question becomes what are our tests? How do we define good? And there’s a few… I’d say that the key things that we’ll focus on are are we tracking the evolution of the stream of purchases and revenue over time well? And then the second is are we differentiating the good customers from the bad customers reasonably well? So, this is kind of like what the distribution of purchases would have been for the holdout period, and this is what I would have predicted them to be.

And that allows us to say not only am I kind of getting it right at the aggregate level for the entire cohort, but also I can kind of pick out the good ones from the stinkers. And that’s typically kind of where we stop. There’s a whole bunch of other things that we’ll evaluate, as well, but broadly speaking they kind of fall into one of those two flavors. Kind of temporal validation and kind of cross-customer validation.

Kurt Elster: So, I love this concept, so it’s essentially like if I know in 2022 we did… Our model says our CLTV for customers who purchased in January is $100, and if I then use my model with data to try and predict 2022 by only using data up to 2021, December 31st, then I see how far off it is. What’s the benchmark for good or bad? How far off before you’re like, “This is just completely useless.”

Daniel McCarthy: Yeah. Usually, and that’s the great thing about these kind of visual diagnostics, so we’ll look at this chart and it’s like time, orders placed, and so typically orders placed from a cohort, customers, they’re like a melting ice cube. It’s just kind of how quick is the melt. And-

Kurt Elster: That’s a good visual analogy.

Daniel McCarthy: Yeah. Basically, you’ll kind of look at this curve of orders from the cohort and it kind of always looks like that. It just sharply falls at the beginning, flattens out, and then the hope is that it goes relatively flat, maybe even slightly lips up at the end if you’re a good business. And so, typically when you look at kind of what you would have expected that trajectory to be, you’ll just know. Am I getting that or not? Or is it clear that I’m inflecting down too much, like I’m killing the cohort more than it looks like they’ve been killed? Or am I way overly optimistic, like the actual curve is like this, but I’m kind of coming in like this?

So, for that one, you kind of know. It just kind of pops from the page when you look at the right figures.

Kurt Elster: It’s kind of it’s just like how statistically rigid we are, and then at the very end you’re like, “All right, you’re gonna validate it by eyeballing that bad boy.” I mean, you’re looking at data. It’s just visualized in a chart. You’re like watching that graph. But yeah, kind of funny. If I’m a Shopify store owner, how the heck do I do this? Are there tools that will provide this? It sounds like these models are very proprietary.

Daniel McCarthy: Actually, they’re not. There’s a version of the model that you could run in Excel. And actually, I teach that model in my CLV class. It’s a pretty good model. Depending on your level of technical aptitude, that could be a good place to start. I’ve got a number of other resources I’d be more than happy to share.

Kurt Elster: Run me through some of these resources. And of course, I include all this stuff in the show notes. Tap or swipe up on the episode art to get to the show notes.

Daniel McCarthy: The Excel model, so I teach that in my course, and even that is not very well gate kept, so there’s a spreadsheet that we use and I even went to the extent of creating YouTube videos. Take a publicly accessible transaction log. The Excel model is there, available on Dropbox, and then there’s YouTube videos to kind of handhold you all the way through the running of the model. And in fact, there’s another YouTube video that I created where we can take that fitted model and then use that to get CLV for a cohort. And so, contribution margin, CAC, discount rate. We didn’t talk about discount rate yet, but that part’s pretty easy.

Kurt Elster: Oh man.

Daniel McCarthy: And then kind of work your way all the way through to CLV for the cohort. So, that… I’d say the good thing about that is that you don’t need to know any fancy language. You just need to kind of follow the rules. And you kind of need to just do what you’re being asked to do. And so, once you’ve kind of internalized that, and you’ve run through, and replicated what was done in those videos, then you just delete that transaction log, put in your transaction log for your cohort, and then do it again.

Obviously, you need to adjust some of the numbers, but once you’ve done that, then you would have the CLV for your cohorts. So, I think that is probably the path of least resistance. Obviously, there’s also the academic work for those who are real masochists and want to stare at lots of Greek letters, but full transparency there, the papers are all out there. They’re available for free for those who kind of know the right place to look.

Kurt Elster: But even if you’re saying like, “Okay, it’s too complex for me to apply this myself,” there is still so much value in conceptually understanding this, and even if you’re like, “All right, I vaguely get it,” understanding the pitfalls that have been outlined here.

Daniel McCarthy: And also, even if you don’t know how to run the models yourself, you know how to say this is a good one, this is not, or again, if you’re speaking with someone else who’s providing you with their model, this is what I need to see to know that I trust your model. And so, even if it’s not me… And the great thing is even if you’re completely non-technical, and we’ve had this happen a couple times, where you have some vendor, and they’re pitching you on some model. Don’t give them the last year. And so, they can’t cheat-

Kurt Elster: Oh, brilliant.

Daniel McCarthy: Yeah. They’ll do their best, and maybe they’re doing a hold out validation of their own. It’s like on held out, on the data that already has the hold out period. But then there’s literally nothing they can do. You say, “All right, just give me your predictions. Give me this. Give me that. And the other thing. And I’m going to see how well your predictions were or what actually transpired over that period of time.”

Kurt Elster: That’s so good.

Daniel McCarthy: That’s so good. Yeah.

Kurt Elster: I like that a lot.

Daniel McCarthy: Yep.

Kurt Elster: Now I’m hoping a vendor will hit me up, like, “Oh, hey, we can figure out your CLTV.” Go ahead. Let’s see how you do.

Daniel McCarthy: Yeah. It really makes you as a vendor… That makes you a little uncomfortable, you know? Because it’s as if you’re being evaluated and graded literally. But yeah, then that’s as clean as you can get. There’s no way to cheat that.

Kurt Elster: Okay, so I don’t know, I’m feeling good. I’m feeling dangerous. I know enough to be dangerous now, which I appreciate, and I want to go down this rabbit hole more, but I want to know how do you apply this as a business valuation model? Because traditionally you would just go, “Here’s revenue, times multiplier, there. We picked a number out of the air.” When the reality is if I’m selling and I got a 50% discount and I’m losing money hand over fist, it’s still gonna look like top line revenue multiplied times 3.5. This sounds like a way better way to get an accurate gauge of business value. Walk me through that.

Daniel McCarthy: Yeah. So, this can help inform valuation in a few ways. The first way is kind of traditional. Take whatever traditional valuation method you prefer. A discounted cash flow valuation, or some multiple of EBITDA, and so let’s say you want to use that method. Then, essentially what this can help you do is kind of inform that model. It’ll give you a more accurate projection of what future revenue and EBITDA is gonna be. And the way it’s gonna do that is it’s gonna first project out this is how many customers we’ve acquired so far, this is what I would project over the next handful of years, and then next level for the model is this is how many repeat orders I expect to get from those customers after acquisition.

And so, you’ve got all those existing customers, you kind of run that model that we talked about to get what we expect them to do in the future, and then typically you’ll kind of make some assumption that the customers you have not acquired yet are in some ways kind of comparable to recently acquired customers. You know, it’s a good place to start.

And then you have average order value and that gets you to revenue, and so if you knew that that model predicted really well on that holdout data period, that’d give you some confidence that at least it’s a reasonable baseline expectation for what revenue’s gonna be. And then obviously after you factor in your expenses, that can get you to whatever measure of profitability you might be most interested in. That’s way number one.

Way number two is to just stick with unit economics and instead of… It’s like keep doing the traditional revenue thing that you might have done, but you’re gonna really want to know what is my LTV, what is my CAC, and how has that been evolving across my cohorts and how has that been evolving within each of my acquisition channels? So, Facebook, Instagram, Google, et cetera, they’re all gonna have different customer acquisition costs. And so, it could be very helpful to kind of look at how many customers have I acquired over time and how has the composition been changing, you know? Before, maybe I acquired a lot of customers through referral and organic channels. Maybe now much more are coming through the Facebook ads and that can help us understand, do I have good economics overall, on an overall blended basis? And are the economics still good within each of those different acquisition channels?

And if I’ve kind of calculated all the numbers correctly, it’s just super important information to know because you’ll basically know yeah, my business is healthy. I may not know where I end up, but to the extent that people are using those revenue multipliers, I know that this should put me at the higher end of those revenue multiples because I actually have good economics. And when I spend my money on marketing I’m gonna bring in more customers and they’re gonna bring in a whole bunch of value that exceeds how much I spent to acquire them in the first place.

Kurt Elster: Absolutely.

Daniel McCarthy: Yeah, those are the two that I would… I think those are the two ways that I would focus on it.

Kurt Elster: All right, so the thing I want to close on is I want to hear about your current venture. After Nike acquired Zodiac in 2018, you turned around, you’re like, “Look, I can’t stay away.” You co-found something called Theta Equity Partners. What the heck is that?

Daniel McCarthy: Yes. Theta Equity Partners is not a hedge fund or private equity firm, so typically these days we’ll just kind of go by Theta just to avoid the potential for confusion. But we basically just do everything now that Zodiac did, as well as customer-based corporate valuation, and so you can think of us as the doctor, and to the same extent that we go in and get an annual checkup, and see what our cholesterol is, and see what our Vitamin D levels are, and red blood cell count, and all the rest of it, we basically do that for companies. So, we’ll say, “This is your unit economic annual checkup.” Hopefully you should be doing it every year.

Now, inevitably you have some sort of a health scare, and that kind of prompts you to come back in. So, we understand if people don’t want to do it every single month or quarter, but I think that the analogy really holds that there’s just a lot of companies that are the walking dead, or they’re just unhealthy. They need to whip themselves into shape, but they don’t even know that they’re unhealthy because they’ve never gotten these numbers done, you know?

And so, we will help them through that whole process. So, we’ll do that for PE firms that are looking to see is the company I’m looking to buy healthy, and then we’ll do that for companies, as well, where we’re looking to help their executives typically understand, like is my business healthy? And if so, where are the pockets of value? Where are the segments in my business that are weak? And that can help them think about strategically should I reallocate my resources from the weak parts of my business to the stronger ones?

Kurt Elster: That does sound valuable. But I can also see where that… It might intimidate people. Do you find that clients are sometimes nervous about this process?

Daniel McCarthy: I think first, there’s always kind of the circle of trust that… Imagine that we do find that the business is not so healthy. Only they will know about it. And so, it’s not like there’s any sort of a risk if the numbers don’t look great that it’s gonna put them in trouble.

Kurt Elster: Yeah. Okay, you’re right. You’re not like, “All right, we’re gonna run this analysis and we’re gonna send it to you and then CC the Wall Street Journal.” It’s done in confidence.

Daniel McCarthy: Exactly. Yeah. So, I think most companies at least for internal purposes, it can’t hurt to know the numbers. And in the best case, it can actually be the difference between life and death.

Kurt Elster: You know, going back to the beginning I hear all the time people like, “Well, I want to make data-driven decisions. That’s what we should do is data-driven decisions.” What you’ve laid out, this is the way. If it’s truly what you want, it sounds like this is a really informed, methodical, reliable way to do this.

Daniel McCarthy: And quick. I know we just spent so much time talking through all these details, but you know, we’ll run through the numbers in a couple weeks. It’s not a long process. It’s really… It shouldn’t be intimidating in that way, either. It’s not gonna be this multi-month process. You know, typically it’s pretty quick.

Kurt Elster: Okay. Yeah. Good to know. All right, where can people learn more about you?

Daniel McCarthy: Yeah, so certainly there’s my website, I’ll share that. On Twitter and LinkedIn I’m quite active, and I do kind of post different things on different websites, so it could be helpful to look at both. On Twitter, d_mccar, M-C-C-A-R. LinkedIn, I don’t even know. Search is terrible on LinkedIn.

Kurt Elster: Isn’t it?

Daniel McCarthy: I don’t know if you ever found that.

Kurt Elster: Yeah. It’s like, “No, I’m 100% sure I’m connected to this person,” and they’re like, “No, you’re not.”

Daniel McCarthy: Yeah. Yeah.

Kurt Elster: Thanks, LinkedIn.

Daniel McCarthy: Yeah, so I think those would be great places to start. We’ve got great content on the Theta Blog, so we’ll do these deep dives into Warby Parker and a lot of other names that you’ve probably heard of, where we take no prisoners. We’ll say that this company… We said Wayfair’s equity is worth nothing. We said Warby Parker was undervalued. So, yeah, we’ll come in both directions. Bull and bear. But we’ll really lay it out, so we’ll go through a whole bunch of detail about how we came to those conclusions. So, I think that’s another third place that could be good to look. is the website.

Kurt Elster: I’m adding all of these into the show notes, but for sure I’m going through that Theta blog. I want to see. I want to see these valuations. I want to see what you think. Daniel McCarthy, this has been illuminating. I learned a lot. I learned where I was totally ignorant about some statistical things. And I loved it. Thank you so much for educating us today.

Daniel McCarthy: Thanks so much for having me. This was really great.

Sound Board:

Kurt Elster: Yeah, the audience just goes crazy at the end of every episode. It’s wild. It’s a standing ovation. It’s the best.

Daniel McCarthy: I know. They’re all out there.