On this episode, you’re going to learn about Real-time LTV, which, according to our guest, is a North Star metric for eCommerce. It’s a great episode you don’t want to miss.

—

In this episode, we’re talking with Tobi Konitzer. Let me start first by saying that if you are into data science and if you’re into retention, this episode is a must-listen. Tobi is a PhD at Stanford University with a background in researching public opinion at the forefront of machine learning. He worked with Facebook prior to this, set up a startup prior to this, and now he’s using AI to determine what is called real-time lifetime value, that’s RLTV, which does not wait for historic purchase data to determine your LTV. With machine learning and certain data points, it’s determining LTV based on the first purchase per channel so you could optimize your channels.

In this episode, I grill him as to what our LTV is, how they come about, and the metrics. What was more insightful from this is other potential uses of RLTV in other eCommerce tooling or your eCommerce tech stack to make them more efficient at personalization, that’s where the real power is. This episode is geeking out in retention. Tobi knows his stuff. His company is Ocurate. It was a decent conversation with Tobi.

One thing he mentions is you don’t have to be a Facebook with regards to experimentation but you must embrace a culture of experimentation to get the most out of real-time LTV, RLTV. It’s an interesting concept and I’m glad I’m covering it on this podcast first before many other podcasts. I hope other podcasts haven’t covered it yet but it is an interesting concept. I’d encourage you to read if you’re well into retention. Our previous episode was a masterclass on retention marketing with Jess Chan. This is a follow-up on the new North Star metric, according to Tobi, called real-time LTV. Enjoy this episode. We’ll catch you on the other side.

—

Tobias, welcome to the 2X eCommerce Podcast.

Kunle, I’m very excited to be on.

Where are you calling in from?

I’m in the gray and rainy city of San Francisco, which is a typical summer for us over here.

I’ve been in San Francisco over the summer and the weather is a bit temperamental. Do you live in San Francisco full-time or are you passing by?

I do. Myself and the family live in San Francisco and like it. For all these negative viewpoints on San Francisco, they’re going to provide a controversial/conflicting opinion.

You’re a handful of Stanford University alumni I’ve had on this podcast. Do you want to give your journey thus far? What brought you here? You’re the Founder and CEO of Ocurate, which is a predictive AI-type platform, which we’re going to talk about. There’s lots to talk about there. What’s been your journey thus far?

I’ll say a nonlinear journey. Probably my background is best described as using data and machine learning to specifically predict the effect of interventions on human behaviors. I did a PhD Computational Social Science at Stanford. I was at Facebook Research for a little bit working on something similar. I then founded a company called Predictwise, a data company, and a targeting company in the political space building out targets of persuadable and elastic individuals that can be persuaded using big data machine learning.

Ultimately, without getting too much into that too deeply, I got tired of US politics. Maybe living in Europe, you’ll understand that the noise of the discourse is one that I think over the long-term is not that beneficial to your health. I will disclose, we are, as a company, on the progressive, here in the US, the liberal side of things. I was always interested in understanding pain points of B2C companies, internet consumer companies.

I started working a little bit with companies in that space after the 2020 election. That’s where the idea comes from that we’re going to talk about and the big vision behind this company, which is to develop a central KPI, an organizing principle, if you will, against which everything in the internet to consumer space is measurable, scalable, and comparable. That central KPI has become predictive real-time LTV, what we call RLTV for us, and that’s very much at the core of what Ocurate has been doing over the years.

That’s a pretty huge mission in the sense that, in commerce at least, most brands look at their LTV and their LTV to CAC ratio to offer guidance for their strategy. Whether it’s an annual strategy or whether it’s a quarterly strategy, you get your LTV and that will dictate how much you’re willing to spend from an acquisition standpoint. It is almost like your ammunition. If I hear you correctly, your thesis here and what you lead with your company at Ocurate is we should go a bit deeper by going for this RLTV, which is a predictive LTV. Do you want to break down the difference between a standard LTV, which most of our readers are used to, to this RLTV? How do they differ?

That’s an important question and there is probably a lot to unpack here. D2C brands have heard for over fifteen years from all kinds of wise people, investors, and private equity, “LTV is the North Star. You need to do everything LTV.” The current LTV metrics that are in circulation and even the predictive LTV metrics are not amenable to, for example, change the way that I, a consumer-facing company, interact with my customers. You alluded to that.

LTV, even predictive LTV, the way that it is done now, is much more of a guidance. I would say it’s a diagnostic. The way that most LTV machine learning works is you wait for purchase data coming in over 3 or 4 months and you can then project out what is the revenue of, let’s say, Bob, who has been with your company in the last year and has bought regularly, what is the revenue that Bob will bring in over the next year? That’s a metric that is good for VCs if you want to sell the company. It’s good for your overall strategy. Should you put more money into, for example, acquisition or retention? These things can be answered.

Of course, it will never change how you interact with your individual customers on a day to day. If you need that much purchase data to project out what the future revenue is, at that point, these guys are good customers, you probably don’t want to do anything differently. If you’ve been a client of a shoe company for a year and have regularly bought shoes, you’re a good customer as is.

If that’s the first time I can come up with your LTV, it’s not that interesting. We call that the applicability gap. The mission behind our LTV, behind predictive real-time LTV, is to close that applicability gap, to offer LTV as a dynamic predicted metric that you can and we have built certain applications on top of that change the way that you interact with your customers early in the customer journey as an internet consumer company.

How did you do it?

How does the technology behind it work or what are the applications?

How does it happen? How do you predict it in real-time?

No easy answer to that because that has kept me up for months. We always were convinced that this task, if we want to push LTV into a KPI that you can make decisions on, that task depends on the right data calibrated for the right machine learning and the right machine learning calibrated for the right data. I’m not going to describe the whole journey as to how we came about, this was more or less a dialectic and empirical process.

I’m open. I want to. I’m inquisitive. I want to know. Go for it.

If we want to predict LTV, for example, before the first purchase occurs, that’s what we’re talking about here. If we want to predict a zigzag LTV that changes every day while the purchase cycle takes months, we need external data that is not purchases. The first idea came from my last job, it was pretty much a stupid idea in retrospect. We had built out this massive database of demographics starting with voter records in the US. We said, “If we overlay that data with customer data, we solve what we internally call this cold start problem.” Particularly, how can you predict LTV without any signal? There is nothing.

Unfortunately, of course, this database that we built out is a static database. Your demographics are going to change ever so slowly. Every year you become a year older, at least that’s true for me, it’s true for most people. There is no dynamicism in there that you could exploit to get to that zigzag always changing revenue potential. Ultimately, the answer came serendipitously, which I hate to admit but that’s the truth. Companies that we’ve been working with started to send us more and more of what’s called event stream data.

An easy starting point to that is, how many emails did you open? Telemetry data, behavioral data, and things that inherently are dynamic. We started to see a relationship between these data points and LTV. Ultimately, what we have done is we’ve built out pretty sophisticated data collection technology, we call that thing Ocuboost. That thing collects every behavioral data point on the site that is related to LTV in a personal identity-preserving way.

I’m going to be a little bit more specific here. Let’s say you are an anonymous customer who comes to a haircare site and you look at 3 or 5-star reviews, you spend 30 seconds on each, and then you go and you look at a specific 1-star review and you spend five minutes there. It’s that kind of data we collect that then goes into our big machine-learning engine. Maybe, in closing, I’m going to describe the intuition behind machine learning and I hope that starts to make sense.

Let’s assume you’re coming to this haircare site, we’re going to collect this behavioral data from you, what reviews you look at, what products you look at, what pages you load, what product display pages you look at, and what educational content you consume for how long. The machine-learning says, “Kunle’s data or ID5712 stars data,” anonymized, the digital fingerprints that individual leaves behind, does it match the digital fingerprints in terms of purchasing data of existing customers or not?

In other words, do you behave like a customer who spends a lot all the time based on your behavioral data on-site or do you behave like someone who buys once and drops dead? With this technology on the machine learning side, we can do two things, we can predict LTV before any purchase occurs with high accuracy, and a big component here is validating the accuracy, and then update these predictions based on any engagement that you have with a brand not relying on purchase data.

From my perspective, with the information you have from predicting data about what user ID, PIP, 101 is worth, how is that going to change how I market? With LTV or my CAC to LTV ratio, I know that if my LTV is $500 over an eighteen-month period with a customer and my margins are so much, I am willing to spend $200 on any acquisition platform to acquire them. That dictates my budget, it dictates the ads I create. It gives me an insight to prepare and execute.

If this data is coming in real-time, is that not just data overwhelm? What do you do in real-time, particularly for a first-time customer? How does that feed into decision-making? The problem with analytics out there is we’re overwhelmed with data and we don’t know what to do with data. How are best-in-class brands using this data effectively to make more money or attract more customers who align with them from a revenue and even value standpoint?

We can call that the applicability gap. It’s super important that we have an opinion on that. It’s super important to avoid the trap of saying, “Here is an API in real-time LTV, and now go have fun.” We learned that the hard way a little bit. We’ve built two core applications on top of RLTV that our customers use with a proven track record of changing the way that they’re engaging with their prospects and customers to generate revenue.

Here’s what they are, I’m maybe going to go with one, and then I’m going to stop there and we’re going to talk about the second one maybe later. The first one is one thing that you alluded to, which is being able to evaluate acquisition strategies on this North Star metric of LTV to CAC. What is the monetary value of the customer in the first twelve months divided by how much money did I spend to acquire that customer? The big difference is we can do it in real time.

It’s important to address what the status quo approach is that most direct-to-consumer companies have to rely on. We can make this very concrete. Let’s say you acquire two customers from Google, all you can do is say, “Those are two conversions. For the same money on Facebook, I got one conversion. I’m going to go with the two conversions and I’m going to spend more money there.” That’s not what you want to know. I’m going to try to get a little mathematical without getting too mathematical. Kunle, if you allow me to understand why optimizing on conversion is short-sighted, we have to do a little bit of background math. Are you okay with that?

Let’s do it.

Let’s keep this in mind, we’re going to close the parentheses here at the end, this Y conversion optimization, which is all you can do is a bad idea. We’re going to have to go back to the principles of revenue. The principle of revenue distribution in a direct-to-consumer company is governed by the Pareto principle. That thing states that 80% of the effect is generated by 20% of the cause. In this case, it means that 80% of the revenue comes from 20% of your customers. If your reader base takes one thing away from this conversation, it should be that one.

That means that your high-value customers generate 16X in sales compared to your regular customers. That’s a big difference and it turns out it’s a much bigger difference than conversion. We can do a little bit of math. The difference between getting the right conversion and the wrong conversion and getting any conversion versus no conversion is 4X. Make this a little bit more concrete. In our previous example, you’re now optimizing over the campaign that got you 2 conversions versus 1.

If the one conversion would have converted a high-value customer, you’re missing out by a factor of 8X. That is what RLTV is. It allows you to evaluate acquisition strategies not by conversion and not by the initial sales data but by the predictive revenue that each customer brings in the first twelve months. It gives you a much more complete picture.

Ultimately, this is going long, but it helps Google and Facebook to optimize their bidding in ways that were impossible before. We’re able to tell Facebook, “You’re looking at Bob. Bob’s LTV will be $60. Dear Google, your bidding strategy now can be much more efficient.” The overall story here is that either based on platform optimization but even if you do channel strategy optimization based on RLTV to CAC, you can increase your revenue based on the acquisition campaigns that you run compared to the status quo.

Now you’re speaking our language when you brought the Facebook and Google examples into play. Does that mean, with your platform, that you’re able to push that data in real-time back to advertising channels such as Facebook, particularly the Meta platform? They’re fed more information from a conversion standpoint and then you’re predicted LTV. If yes, how does that happen?

I want to make clear here that I’m heard. That is a smart question and shows that you have a lot of experience in the space and know what you’re talking about. I talked about two different things and you picked up on this. One is the world today that RLTV enables you, which is understanding the LTV to CAC of every channel in real-time, putting more money into these things that work on a better signal, based on a much better metric, and down spending things that don’t work.

I alluded to something that you picked up on correctly, which is the ability to push that LTV, that RLTV, back into the acquisition platforms and then change their internal optimization procedure. That’s a different thing. If we can do that, Google can adjust its bidding strategy. This is what happens under the hood. On your behalf, Google is placing a bid based on every prospect. It happens in the black box. If we can do that, we can give Google and Facebook the power to do this with efficiency, knowledge, and accuracy that never have been there before. We can do that. That’s quite revolutionary.

Ultimately, how does something like that work? Without going too much into the details, there is a way in which we can establish a feedback loop between the acquisition platforms individuals that come to your direct-to-consumer eCommerce site and their LTV and push that back in real-time. This process that I alluded to on these acquisition platforms adjusting their bidding with this data that hasn’t been there before is something that’s called value-based bidding. We have to do some R&D around that.

To me, there is high hope that this will change the status quo of acquisition. Facebook was my former employer. We know that this is where these platforms are going, that’s where they spend their money on. If something like that will work and we figured out the technology to do that, which we’ve already done, we’ll revolutionize the way that acquisition has been happening for the last twenty years. My suggestion before I get ahead of myself as a CEO of a company, which is my job, is we want to be careful. Our background here as an academia, we’re big on R&D and testing. If that test comes positive, I suggest we do a two-and-a-half-hour podcast about that.

They are value-based lookalike audiences on Meta advertising. From what I’m reading here, there’s an SDK. From past experience, we used to essentially upload CLTV data but that was exported from platforms such as Shopify. It was static and it was not real-time CLTV.

Kunle, let me make a guess here because we’ve tried this many times as well. It didn’t work.

Sometimes it works marginally. It was not like a 50% double-type thing. It was marginal. It all depends on whether you’ve maxed out that audience or not. It was relative. It didn’t always work. It worked sometimes. What I’m trying to get to is do you think that there’s going to be mass adoption of real-time LTV? It’s not my first time hearing about real-time LTV, I’ve just not paid sufficient focus on it. It’s my first time, in public, speaking about the topic.

Do you see the focus in general moving to real-time LTV? Do you think platforms like Shopify Plus or even third-party tools, and Ocurate is well into this, will be feeding this data? Would acquisition channels also be utilizing this data fed to them from their customers, advertisers, websites, or assets? Where do you see this playing out in the future? How are your best-in-class customers using this data effectively? It’s all well and good knowing about this in theory but the question is applicability and using it for competitive advantage.

Two questions that we should maybe dissect a little bit and take apart. I’m going to make some notes now too because my short-term memory is not what’s used to be.

It’s a long-winded question, I’m notorious for that.

I’m a good note-taker and I like long questions, there we go. I’m going to start with the first one, which is how do we see the future. You describe it very well. I’m convinced that this will be the future. I’m convinced that a metric like RLTV will become the KPI that is an anonymized KPI that exists everywhere in the ecosystem from Shopify to being pushed to ad platforms to being the organizing principle within the B2C companies as well. It’s going to have a lot of downstream effects.

By the way, I hope that we’re the ones who bring it about but that may not be the case. It may be in 2 years or it may be in 3 years. I’m convinced that it’ll go there. It’ll have a couple of downstream effects. The big downstream effect is you have all these smart tech companies, ad tech companies that are focused on what I will call a misguided derivative of LTV. Maybe it is conversion. Lots of platforms out there say, “We’re going to increase your conversion rate by 1x or 10%,” or something like that by adding 1x or 10% or something.

It’s appealing right now to D2C brands because that’s a language they’ve been speaking for a long time. As we outlined, it’s the wrong thing to focus on. Another example, and I should maybe specify this again. Because a conversion is not made equal to a conversion, you get a 16x conversion, it’s a big difference from getting a 1x conversion in terms of the monetary power of the customer.

Another example is product recommendation tools. Many product recommendation tools out there do something in the black box based on some correlative analysis. Ultimately, the penetration of RLTV as an organizing principle will mean that all these adjacent tools that don’t speak to each other will ultimately disappear. It’ll be for the better for D2C companies that are helplessly overwhelmed by all kinds of software solutions that address little it’s and bids of their customer journey, of their pain points, but never the whole damn thing. It’ll make that market better.

I do think that the foundational principles of D2C still apply, which is to build a good product that customers like. Nobody talks about that but that’s always the necessary condition to do anything. Frankly, in my journey, I’ve seen a lot of folks that haven’t done that when investors invest a lot of money in D2C companies. We’ve seen companies that had a higher cost of goods sold versus the sales price where the optimal number of customers was zero. That, plus the foundations of building a consumer-centric product will, at some point, displace everything and will be the future of everything. The question is when and the question is how. I hope we have a role in this.

I’m not a futurist but I see a use case here for RLTV based on the fact that it is the metric feeding SaaS-based solutions for D2C consumers rather than a metric that feeds into immediate decision-making by the eCommerce manager, eCommerce marketer, or what have you. If you use real-time LTV as fuel to power up essential modules like cross-selling or upselling to power up acquisition, to power up personalization in real-time, it’s existing at the data layer infrastructure and feeds all of this and it’s proprietary and unique. It’s an acceptable single source of truth for so many tools.

They’ll more efficiently deliver what they’re meant to deliver to D2C merchants. Those tools are more efficient in working in real-time to optimize the user experience as compared to me sitting in my dashboard as an eCommerce manager, I’m getting real-time LTV, what the heck am I going to do with it? If that’s piped into tools that I’ve put out there, automation tools, this is a nice segue into ChatGPT. If that’s fed into all these tools and apps then.

That’s an excellent point. What you’re describing is RLTV as the base layer of a completely vertically integrated stack of technology tools and I could not agree more.

Correct.

We don’t need to do all of it. Our vision is to, and you said it very well, make RLTV the fuel, that single source of truth. If there is a churn tool on top of that, that uses derivatives of that metric, that’s great. If there is an upsell tool or a product recommendation tool that uses this metric or builds derivatives off of that metric, that’s great. Yes, that probably provides a little bit more detailed vision of what I outlined. You’re right on this.

You also asked how are our customers using that data now. It’s a nice connection to this stipulation on the future because we have to start to build some of these vertical integrations ourselves. We decided to have two vertical areas of integration and one is evaluation of all acquisition strategies. In concrete terms, we offer you a dashboard in which you see the LTV to CAC of everything you try immediately.

That doesn’t exist yet but because we can predict LTV before the first purchase, immediately after the first purchase, we can essentially say to you, “Last week, you tested 50 variations of content with ChatGPT, and you ran on 10 channels. Here is the exact LTV to CAC, meaning the twelve months LTV total and average over the customer acquisition cost for all these things.” You scale up what’s working and you scale down what’s not working. We have some nice case studies out there. Good Clean Love shows that these guys increase the number of their high LTV customers by 19% doing that with then outsize effects on revenue, north of 15%.

Is this split test in on-site copy?

If you will, yes. It’s essentially whatever you run. We’re agnostic as to what you run. Even if you say, and this was the case study that we have, “I’m going to run ads on fifteen channels. That’s what I usually do.” We then put the dashboard on top and say, “Channel 1, 2, and 3, last week, had an LTV to CAC of 7 to 1. Put your money here.” Of course, next week, that may look different because channels change all the time. Acquisition platform algorithms change all the time. It’s allowing you that dynamicism in redistributing your money in a much more efficient and time-granular manner.

There’s also the challenge of creatives. If you perhaps plugged into your Facebook advert or winning creative, it might spike those metrics up from Facebook over time. That makes a lot of sense. For people who want to find out more about Ocurate, you’re on Ocurate.com. What’s a sweet spot for D2C brands that work with you? What D2C brands benefit from the data and the services you offer?

You can find a lot more at Ocurate.com. You can also drop me a note. I’m always interested in having conversations such as this conversation that I tremendously enjoy. With founders of D2C brands talking about predictive analytics, the future of AI, organizing principles, homogenization, and all these topics that we hit now, drop me a note at Tobi@Ocurate.com. This is how we can get that conversation started. I always love to have those.

As to what do you have to be as a D2C company to fully take advantage of what we have to offer? I’d say two things. The applications that we build out lend themselves to spending money on acquisition. If you don’t spend any money on acquisition, you probably don’t need to sweat evaluating acquisition strategies correctly. That’s one. The other use case that we’re after, which we haven’t talked about much and one that’s near and dear to my heart is offering a central measurement variable, a central criterion against which you can evaluate all your A-B tests that you run internally.

Running experiments and coming up with an experimental program is hard. We offer RLTV as a one-stop metric that allows you to run them faster, make them comparable, and understand the effect on revenue pretty much in real time. If you want to leverage that value proposition, Kunle, that’s the other big application that we build out, our other selective, and more vertically integrated approach in experimentation, you need to have a culture of experimentation. I don’t think you need to have an experimentation program like Facebook, for example, where we learned a lot of the lessons that we’re now applying but you should have a culture of experimentation in your company. It’s a great use case to do all the experimentation against a more efficient outcome, which is RLTV.

A VWO, which is a split testing platform, a CRO platform, how does it get the real-time LTV from yourselves? How do you work with a third party?

It works the other way. You run your experiments however you want to run them. We can log what’s called the individual-level treatment assignment, which is all we need to know. We need to know whether ID215 was a member of the control group or got the treatment. We can log that into the way that we collect data. All these experiments can be evaluated on the Ocurate platform.

The nice thing is we don’t have to worry about how we get the data from X to Y. We need to push the data into these platforms, which is notoriously hard. We can’t push data optimisely on these others. We can log it passively and you can simply evaluate these experiments on our platform much more quickly and in a much more comprehensive and exhaustive manner through RLTV.

Tobi, it’s been a pleasure having you on the 2X eCommerce Podcast. You’re quite active on LinkedIn.

Thank you, Kunle, for having me. I am newly fairly active on LinkedIn so we can also connect there. I tremendously enjoyed the conversation and thank you for the excellent questions.

Incredible. Cheers.