Predict Customer Value in Real Time with AI

[00:00:00] Richard: Hey guys welcome to the Ecommerce Playbook Podcast. My name is Richard Gaffin. I'm the senior copywriter here at Common Thread Collective. And joining me today again, as always is Mr. Taylor Holiday, the CEO of CTC Taylor, how you doing?

[00:00:13] Taylor: Doing great. Yeah. I thought we were officially moving you to professor. Is that not happen yet?

[00:00:17] Richard: Well, yeah, not officially. You'll have to have a talk with Corey and see if he can make a professor my official title. So what we're talking about today here is something that's very exciting to both of us and to everybody at CTC in general. That is our relationship, unique relationship, I should say with Retina AI.

Which I'll try to describe it as best I can. it allows you to see in the present what the value of a client is going to be in the future with a fair amount of accuracy.

And if any of you follow what we've put out here at CTC, you know that cohort specific LTV modeling is a really important part of our philosophy and our methodology. And Retina gives us an even clearer picture into what it might look like. To see future results in the now.

I'll throw it to you then, Taylor. Give us some background. Why did you go down this rabbit hole in the first place of exploring or trying to find an AI solution to predictive LTV modeling?

[00:01:06] Taylor: So what you just described, this evolution of trying to make better decisions about the future is sort of fundamentally our job.

We are a growth agency, which means that we are making some assertion for our customers about what we can provide them in the future. And I care deeply about representing that. Accurately. Not to do it in a frivolous manner, but to be honest about what we believe is going to occur and to be able to validate that premise authentically and honestly.

I think about the evolution of forecasting , it sort of works like this generally in businesses. Level one is 2 million and you just write a number down in a spreadsheet. That's sort of like the first year, right?

Level two is a percentage against the previous period. We want to grow 20% year over year. And so we just take all of our numbers and we multiply them by 1.2 and that becomes the goal.

Level three is what I would say we're on right now which I would call cohort specific LTV forecast, extrapolated on the past. Meaning you look at cohort values of customers and how they've performed historically, and then you build a model off of that data about the future.

But in our model, the assertion is the future is like the past. Meaning customers increase 10% a month, one 8% a month, two 6% a month three. And so we model every new cohort of customers to be like the previous ones. Well the limitation with that, that I've always understood (I made a YouTube video about this once) about the idea that past performance is not always indicative of future outcome.

Well, this is very true with customer cohorts. Not all the customers you acquire tomorrow are gonna behave, act or be worth the same amount of money as the ones that you already have. And so our model right now that we use for all of our forecasting and our growth map, Is limited in this way. And I knew this and I was always looking for ways in which we could improve it.

And when Orchid Burleson, who's our COO, started here from Nestle. She introduced me to Emad as she began to understand what we were attempting to accomplish. So Emad's the founder of Retina AI, super smart dude from Facebook and PayPal and all these places where he been working on and understanding this premise.

And she said, this is the kind of solution that you're looking for to go to the next level. Which is to be able to forecast, not at the cohort level meaning a group of customers that have some shared attributes but at the individual user level. And to be able to build models not just on their purchase behavior on your data, but a richer profile of their purchase behavior based on credit data and other profile data about those users .To make predictions that are incredibly accurate about their value in the future.

That is sort of rooted and I went down this rabbit hole of learning about like buy till you die models and all these different things that it existed about predicting future customer value.

And that's what retina does really, really well.

[00:03:48] Richard: Yeah having looked into retina myself, it's definitely pretty, thorny and pretty complicated. Which kind of brings me to a question around, what is specifically interesting or exciting about our partnership Retina.AI

Why is the retina CTC one, two punch so potent?

[00:04:04] Taylor: That's right. Okay. So there are two main problems that I have been on this mission to solve more. One is this idea of not having to extrapolate the future based on historical data. I hate doing it. I know it's not as accurate as it could be.

The other is this disassociation between attribution and optimization. And so when I first met Emad one of the things I expressed was like, okay, even if this is true, how does this actually affect my ad spend? Because Facebook is still not optimizing based on this data. So even if I can report this cool CLV metric, how do I change it? How do I make it go up or down with my ad account?

At the heart of this is a very deep issue that's plaguing our industry right now. Which is that there are all these attribution tools that cause brands and businesses to disassociate, attribution and optimiz. And when you do that, you remove your ability to actually impact causally the numbers.

So what do I mean? Like if you're using a third party attribution solution and that's your definition of success and you're using it to make ad buying decisions, but Facebook doesn't have that data. They're not using it to inform who they deliver to, what the bid is, what the expected conversion rate is, all those things.

Then these metrics are often not correlated to each other at all. Like we'll run this data and we'll see this attribution metric and the Facebook metric don't have any relationship. So as a media buyer what am I supposed to do? Well Retina Understood this and they actually did the work to solve it. So in partnership with them, we have begun optimizing our ad accounts on data at the predictive individual customer level, CLV level, cuz it passes that data back to Facebook and informs the optimization. Ah, the marriage of attribution and optimization not just at an average order value level, not just at a ROAS level, but on a one year CLV to CAC level.

Now that is really,really powerful.

[00:05:54] Richard: For those who may not be familiar what is CLV? That's kind of the central metric and really what Retina is...

[00:05:59] Taylor: Yes.

[00:06:00] Richard: Can you unpack that a little bit?

[00:06:01] Taylor: Okay. Great question, Richard. Thank you for going after my acronym use. In our world, there's a very confusing thing we do in conflating financial metrics. So average order value in eCommerce refers to the total revenue captured from a customer. But value as a financial term, is a reference to a net outcome. So if I say the value of a customer, what I really mean is the amount of dollars left over after you've paid the COGS, after you've paid the CACS, after you've paid all the variable costs along the way.

So when we say customer lifetime value, we don't just mean customer lifetime revenue. In other words the amount of money. But inside of Retina's CLV (customer lifetime) ,the value is actually a consideration for the cost of goods in that order. And so when we say CLV to CAC, we literally mean the net outcome over the cost of acquisition, which shows you the actual dollar amount or the return on invested capital that you're making in one year on that customer with a consideration for COGS.

So it's more powerful than ROAS, which has no consideration for value. And it's more powerful than average order value, which only shows you the initial realization of value from a customer.

It gives you the one year total value cost considered metric from the customer. And it uses that number to inform optimization. Mind blowing, incredibly powerful.

[00:07:21] Richard: Yeah. So I mean in essence, it's like optimizing towards contribution margin or something like that.

[00:07:26] Taylor: Yeah, exactly. Now, contribution margin is another term that you'll get differing opinions on.

But yes, that's the general idea. And this is something if you go and you watch my hierarchy of metrics video (in CTC's YouTube channel), you'll see what Richard's referring to. This is sort of what I consider to be the scoreboard or the most important metric that if I was running growth at a brand or when we partner with brands, we try and push them towards measuring contribution margin. Which we would define as Net revenue (so after returns) minus cost of goods, minus ad spend, minus R fees, right? Like in other words, how many dollars flow through to cover your opex or flow through to cover your fixed costs and ultimately to operating income. Because that's the part, as a marketer, we control, We don't control their opex usually.

And so we look at contribution margin. And when we think about CLV to CAC, it's a view into that reality over a one year time window.

[00:08:12] Richard: Yeah. Makes sense. So I guess it's just kind of interesting that it's operating on two levels. It's operating essentially as an LTV stand in or whatever, or a better version of LTV will at the same time bringing the idea of contribution margin or actual operating income into the conversation as well.

[00:08:27] Taylor: Yeah. So you are getting married attribution and optimization here, affecting your ad account in a critically important time . Where data is wonky in that world and we know it. You are getting an instant view into the value of your customer, within 96% accuracy for most brands, it's about the number I see. (And we'll talk about that in a second I know you're gonna ask me something about that). So you're getting instant access to LTV which is a thing you usually have to wait a really long time to understand. Right. And then you're getting that relationship then to that value against the cost to acquire it.

And so you're getting what is really (ROY) Return On Invested Capital. This view into the one missing piece in that would just be like a consideration for the cost of the capital, but pretty dang close to really understanding when you spend a dollar, how much are you making?

And the great thing about Retina is they work with a lot of large businesses. So let's say you're a brand, you know, that has a longer brand of CLV. They can feed back a three year CLV to CAC. That we're trying to push with them to actually get a 60 day LTV to CAC and remember we care a lot passed back. But their models are sort of structurally different in a way that makes that more challenging.

But still ,right now we've gotten them to shorten it. And what's so cool is because it's a custom conversion event, we can actually in Ads Manager and in stateless see, CLV to CAC in our dashboard. And we're not just going, oh no, I hope it works. We're now actually able to know that Facebook's optimizing against that outcome too.

We're incredibly optimistic about the potential here.

[00:09:48] Richard: Yeah. The idea of being able to optimize, for the right like the right individual is pretty unbelievable. Thinking back to my times and growth teams as well and how frustrating it is to be using ROAS as main performance metric let's say. And it often doesn't result in the end, in the final outcome that you want, that's being such a disappointment because you've been acquiring the wrong person, or it's been jacked up for the wrong reasons.

But it seems like CLV is a little bit close, as close as you can get to a BS proof performance metric, perhaps. Would that be fair?

[00:10:17] Taylor: The other thing that it does is because it's not at this like group level, like a lot of times people will look at things that like LTV by channel, It's a really flawed way to think about this versus the LTB of people.

Individuals, right? Like humans. And so who at the individual level, Is your most valuable customer? And then you know what's really cool, Richard, you'll appreciate this as a creative, is now we can begin to understand, what angle to what type of person generates the best CLV. We try and sort of, get into this by looking at first product purchased as the indication of value and we can assert something about the kind of people that buy that product. So like Born Primitive, the brand we're gonna talk about, we'll often draw distinctions between male and female customers.

Well, we don't actually have their gender identity as a data point, but we use the products they buy as proxy for it. So we'll say things like purchasers of women's leggings are about substantially higher CLV than purchasers of men's workout shorts.

And we'll do comparisons that way. But again, the shared attribute there is actually just somebody bought a product. We don't actually know who they are or why. But now we'll be able to go a level further in being able to say like, this set of individual users were worth more. Now here's some of their common attributes in each other and maybe it's product purchase but maybe it's geography. Maybe it's interest, maybe it's other things or maybe it's the creative angle that we're bringing them in on that creates the best premise or promise to create the best experience.

Like there's just so many new ways to now view your business and begin to solve problems here. That is really exciting.

[00:11:44] Richard: Yeah. Well, I was gonna say from a creative perspective, it is, it's incredibly exciting to have a tool that allows you to break out of the kind of prison of the, persona kind of making these sort of blanket assessments that I think this type of woman age, whatever to whatever is purchasing our product, therefore we should make their messaging like match that.

But it sounds like part of what CLV does, going back to the theme of the elimination of BS, is sort of forced you to think about what the product does for happens to be buying the product. Let's say a knee brace as an example. There are young people and old people that might want that product, and at the end of the day the fact that one of them watches Fox and the other one of them is binging Emily in Paris on Netflix or whatever, doesn't really matter that much.

What matters is they have a specific pain that is important for them to solve. So there's almost a humanized equality to it.

[00:12:31] Taylor: It does, when I think so often in like cohort is this word, there's a humanity to the value of customers. And then what it also does is, on the other side of retina also enables us to like, really highlight highest value customers and lapsed high value customers. And how much money might be left on the table if we were able to reactivate those folks .And to think about then what is a target CAC on a reactivation of lapsed high value customers? Ooh. Now that's a very specific, interesting premise. That again the specificity required to solve the modern growth problem is actually a lot harder, right?

We have to be more specific. We have to have a deeper understanding. We have to have clarity of value. It's just hard right now. And everything is ambiguous in a way that's really confusing. So I think we should probably get into a little bit of the specific results. Okay. Taylor, this sounds great.

What is actually happening here?

[00:13:16] Richard: Let's hear it. Yeah. What's actually happening here, Taylor?

[00:13:19] Taylor: So when they came out with this optimization thing, we partnered with a couple of brands. They're running this alongside SmartyPants Vitamins and we brought Born Primitive to the table and Facebook came to the table.

We worked very directly with them. Facebook is putting money towards helping us to make sure that these kinds of tests (cuz they have skin in this game) so again, another opportunity for you all to think about what does Facebook care about? What kinds of programs are they, getting behind?

This is something they care a lot about. And we designed a two cell test for Born primitive. Same ad, creative, same audience. Just the only difference in the test was the optimization. So in one case, we're optimizing for purchases, lowest conversions, standard campaign set up. And the other one we're optimizing for one year CLV, right?

Like, so again, is the value optimization that's affecting the outcome? And we compared them all across every metric. So we compared them not just at CLV, but we compared them on ROAS we compared them on CPA, we compared them on AOV. All the different metrics and the obvious thing happend. Okay, which is that there would be an increase in CLV of the customers that were being optimized for on that premise.

Okay, cool. That is like reinforcing the initial hypothesis that the value of the customers that we were optimizing for CLV would have a higher CLV and they did,. They had a 94% higher first year CLV per customer in that test cell. Okay, but where it really became interesting and what forced me to send this tweet out the other day that we may have just stumbled into something really powerful is that something even better happened.

We actually lowered the cost of acquiring those customers by 30%. So what does that mean? Not only were we finding higher value customer, but in finding those people they were actually more likely to purchase and cheaper to acquire. So you're lowering CPA and increasing CLV. So what that does to CLV to CAC is you have a significantly higher outcome . That is where you go, oh my gosh.

Like cause in your head you'd think like, well, maybe we have to pay more to get these high value people. But the reality is that in the initial test , they were actually cheaper to acquire too. And that's where if you can replicate that, and now we're in this sort of like next phase of continuing to roll this out more broadly.

Even if that didn't happen, even if CPA was flat. If you increase CLV by 94% that's a game changer. But if you have the opportunity to do both, it just speaks to the power of giving Facebook the right signal. And this is why like I'm such in a war against this attribution thing is because the signal you pass back to Facebook about what is working. It's the whole game and without it, the rest of it is nonsense.

[00:15:52] Richard: We'll have cut here. Got a little excited till set a coffee fit. I mean, it is exciting. What can I say? But to your point then, like makes a certain amount of sense. Even if it didn't beforehand, that the type of customer that would be most valuable to you in the long run is also the type of customer who would be excited about your product in the short term as well, right?

That's right. And that was like a hypothesis we didn't necessarily have. But it's played out for us that way which is fascinating.

[00:16:15] Taylor: That's exactly right. And so what I would say is that in this moment, I think we all need to acknowledge the maybe desperation is too hard of word. But the need that we have for a novel solution to the present problem. I feel a lot of days like we are trying to improve the horseshoe when we need a, automobile.

Like we need a fundamental alteration to the way that we're doing the thing that we're doing in this new world. And this to me represents such a stair step level up from just trying to insert this proxy attribution metric but no data being passed to Facebook and then just like running around in circles feels crazy. Like I can't explain enough how insane the behavior is.

Watching it play out in real time with customers inside of businesses making decisions, it's killing them. But this represents I think functional level up in a way that could become like a core part of the Facebook feature. The obvious endpoint here is that Retina becomes a Facebook product at some point. Like they get absorbed into the optimization if this continues to work in the way that we hope and expect it does.

And so, we have the ability now to bring this in partnership with our service offering to you in a way that combining CTC and Retina is unique for three really specific reasons, that I think we are worth addressing. So those are number one, is that just very simply we have a price of the product based on our relationship with them, that when combined with our service, is a price that you can't get without us and that's just Emad's relationship.

We believe that our combined efforts represent a mutually beneficial solution for our partners in a way that you're gonna get the best out of both of us together, more so than separate. And so the price that you get working with us for Retina is better than anything you can get from them directly. And vice versa, like the price of our service when you add that on has benefit as well. So one is just that the very practically the access point is cheaper.

Two is the shared belief and relationship and methodology here ensures that we don't have to become fluent in understanding how to make use of the software. The reality is there are so many tools in this world that can do something cool for you, but you have to be an expert on the utilization of the tool.

Our teams have calls every week about the product development. We're in each other's product roadmaps. We are intimately having strategic calls between our strategies and there is about how to make the best of this data such that our implementation and Emad would tell you this directly if he asked him, is better than any agency that exist.

And so there's just so much shared ethos around something we've been talking about for so many years around CLV as a core feature, and then also around this tying together of attribution and optimization. Like that is our narrative in the marketplace. That's who we are and who we've been for a long time.

I've been standing in this center of the town square having rocks thrown at me over being the anti guy. And this is why, is because there's better answers.

And then the third thing is like, because we already have all of our customers data warehouse in stateless, our integration timelines with Retina are faster because of it.

So you get faster access to the data, you get the best possible pricing and you get the best strategic implementation when we work together in a way that I think can be transformed in for businesses. And we're starting in a small cohort of them and now we're beginning to broaden that pool out in a way we're really excited about.

[00:19:16] Richard: Yeah, I was mentioning before having, seen the platform itself of Retina. It looks like it would be like flying the space shuttle or something like that. It's very, very complex and so unless your agency partner, fundamentally understands the framework and the philosophy behind it and has that tight relationship with them they're not going to be able to understand it either, and that's fundamentally a problem.

So I think like we're uniquely positioned to understand the tool that we're given and then put it to use for you.

[00:19:42] Taylor: Right. And then being able to build a financial forecast off of a CLV to CAC number is like the next step. Right. Okay. So you have one, you have a CLV to CAC number. What does that mean about your business?

What does it mean about your capitalization? What does it mean about your cash flow? Well, that's all built into our service, like we do P&L level cohort specific forecasting. Right. And so now understanding how that might impact your business is a very clear reality for what we do.

[00:20:05] Richard: Yeah. Couple straight questions.

So one, we've tested this with Born primitive one of our clients.

[00:20:11] Taylor: Yep.

[00:20:11] Richard: They're an apparel brand. To what extent do you think those results are specific to that vertical or how do you think it would change depending on the change in vertical?

[00:20:18] Taylor: That's a great question.

So this is obviously going to matter more the wider, the discrepancy in value that there is in your customer base. Okay, Born primitive has a very wide discrepancy between their highest value customers and their lowest value. They have whales in their business.

[00:20:35] Richard: Mm-hmm.

[00:20:35] Taylor: I call it a casino business, right?

Like they have people that are worth putting on a private jet and getting 'em to the warehouse to buy product because they're worth so much money. That is obviously going to be when there's clarity of who is the most valuable. Then there's a subset of those people that exist that's going to be the most effective use.

That's why we chose them. The other categories are gonna be subscription businesses that's true. There are usually businesses where CLV is critical to understanding and understanding The discrepancy of those customers is huge. If you are selling furniture, probably not the product for you. This is probably not the solution to your primary problem.

Right? You're a first order value business. Understanding value discrepancy in customer can happen on a first order basis. You don't necessarily have to predict LTB to do that. So there still can be some discrepancy in that but ultimately this is going to work best for businesses where lifetime value is a critically important metric and part of their consideration and desire to go find the most valuable customers within a profile of customers that's wide ranging.

[00:21:28] Richard: So we're talking not only apparel but beauty on cosmetics for sure. What else would fall in that category? Food and beverage?

[00:21:35] Taylor: Yeah. Skincare supplements is another one. Any subscription based product is gonna fall into that category. So any consumable, like you said is gonna fall into that category. Apparel is an obvious one. Any wellness product really is gonna, do well here.

There's so much opportunity. There are less businesses these days that are you know, dismissive of LTV as an important part of their future growth journey than there are than there was maybe three years ago.

[00:21:57] Richard: Yeah. Here's another question then. We've talked a lot about how incredible the predictive power of this tool is. How are we corroborating the results that Retina gives us? How do we know that they're just looking into a crystal ball on the other end and making something up.

[00:22:08] Taylor: Yeah. So this is a great question. It's really important to understand about data, and this is another thing. We all as marketers have an obligation to answer this exact question, Richard, which is, why do I believe that the number I'm using is valid? And I, appreciate the question so much. again, another one of the first ones we asked is like, great, you showed me a CLV number you're predicting the future. What if you're wrong? What if I'm making decisions about something that's fundamentally wrong?

Well, the way you answer this in data is actually to do what's called back testing. Where, you create a model and you apply it to a cohort. Let's say that you apply it to a group of customers in January of 2021.

And you use the model to predict the LTV of a group where the LTV is actually realized where you have the answers in the back of the book, so to speak. So you use the model to predict their LTV, then you compare it to the actual. And so you say, how close was my model in predicting the actual LTV that came to life?

And then you refine, and then you refine, and then you refine, and you back test, and you refine, you back, test and refine. And that's how you build a model. That's how you ultimately determine that there's any merit to the idea at all, right? That's how you get to a product. And so this is what every time a customer starts, with Retina, they run those back tests and they show you the level of accuracy that they have in their back testing.

So you get a sense of how much can I trust this model? And they're very honest about in the instances where it's like, hey, we're only about 60% accurate here. But in a lot of cases, including the one that we're using with Born Primitive, the model is like 96% accurate in predicting LTV in a one year window.

That's incredibly powerful, this is better than your predictive expectation. It's better than my extrapolation of the past. And this is what AI is here to offer us as marketers going forward. It's what machine learning is here to offer us going forward, is the ability to back test the premise in a way that has an obligation to accuracy.

When I see media buyers running about making decisions based on this ad did this or this many impressions, it's like what you're removing from that is your own responsibility to the rigor of testing your own decisions against accuracy and this is what computers do really well.

And so Retina's committed to that. They run the back testing to give you the historical accuracy of their ability to predict your business's CLV.

So what do you do you just listened to this podcast, right?

You reach out to us, and I know this is like a very direct sales message, Richard, and I think podcasts it's a dangerous territory to wade into. But here's the thing. We have an opportunity for to access these in free trials for a period to get view into the data or there's opportunity to apply this in funnels as a test and project.

For you or there's a way to bite off the whole thing and get started. I know how hard things are right now and I know that many of us and I've felt this part of the reason I'm so committed to this is I am sitting inside of an organization where we're managing Facebook all all day going, If I don't come up with a reason to believe that things are going to change. They're not , just persisting in the old behavior is not going to lead to a change in this environment.

And right now we need change. We need something different. And so I would just simply ask yourself, what reason do I have to believe that I'm going to fundamentally get a different outcome in my Facebook out account than the one I've gotten for the last six? And if you're dissatisfied with what you've gotten in the last six months and you need a reason to believe in something better, then I suggest you pursue something like this.

[00:25:20] Richard: And for, certain brands like we've been talking about, it sounds to me like optimizing for CLV is kind of the single most important thing that you could do for your brand right now. And this is the best tool available to.

[00:25:32] Taylor: Hey, there's probably other things out there pursue them, check them out. If you find something better that's aligning attribution and optimization, giving you predictive LTV at the individual user level and helping you to make decisions in a way that's instantly accessible, by all means, go get it.

[00:25:45] Richard: Yeah .Thanks again for joining us on the eCommerce Playbook podcast. Remember to like, subscribe and leave us a review. It really helps us out as we embark on this new iteration of the pod. If you're interested in CTCs partnership with Retina AI and how it could benefit your business specifically, check out the link in the episode description.

There's more information there. You can read more in depth about our case study with more primitive and other clients and get a better sense of how it might impact you specifically . If you are interested in starting a conversation about working together, please feel free to drop us a line@commonthreadco.com.

We would love to chat with you. Have a good one, folks. And as we like to say here at CTC, happy scaling.