← DTC Podcast
DTC Podcast
Bonus: The Marginal ROAS Revolution and How Meta's Algorithm Actually Works with Constantine Yurevich of SegmentStream
00:48:52
Loading summary
Transcript72 lines
- [00:00]
A
There are many misconceptions about roas and there are also many different things you call roas. Platforms do not show marginal roas because they want to invest as much as possible. A lot of D2C brands, they invest so much into activities you should not be investing. When it comes to budget allocation, you should never look at Is it better.
- [00:22]
B
To just give Meta your budget and let them determine where it goes?
- [00:27]
A
For Most of the D2C brands, for example, this is one of the most important things how meta works. And Meta also has a model on creative level. Technology will always do whatever you tell it to do. A lot of people right now are afraid about whether AI is good or bad. The biggest issue in this market is your objective.
- [00:55]
B
Konstantin, welcome back to the DTC podcast. It's been about two and a half years. I remember our conversation well. You were talking about the big challenges of attribution. You had some specifically hot takes around attribution being dead and rather than trying to revive the corpse, let's use data to model better attribution. So I'm excited to catch up on the world of attribution and marketing measurement. How you doing first of all?
- [01:24]
A
Yeah, I'm doing great, thank you and pleasure to be here again.
- [01:27]
B
Nice. Well, your setup was decent last time, but you've got the rode mic now. You've got a great background. Your setup's even better now. So I can assume that your technology for marketing attribution improves as well. Even better. Yeah, exactly. So what hot takes do you have for us today in the marketing measurement and attribution world?
- [01:48]
A
So first of all, I would really love to share that and I really enjoy that more and more marketers actually awaken to the understanding that something needs to change. And when like few or three years ago people still had their they still believed that something can be improved in multitask distribution. Something can be changed, something can be tweaked, tracking can be improved. After that there was a huge hype about MMM and everyone switched to MMM and they tried to reincarnate this technology that was used like 60 years ago.
- [02:22]
B
Like the Madman era.
- [02:23]
A
Yeah, right. And mostly applied to different like huge corporations selling Coca Cola across the globe and they needed to have at least some kind of measurement. And after this problem with multi distribution, people tried to reincarnate this methodology and most of the companies failed to implement it or to get any reasonable results. And now more and more people actually start looking for some alternatives and actually it's a great time.
- [02:53]
B
Nice. Well, I have a note here. What what do you say, what do you find is misleading about the concept of roas? Roas as a performance. You know, we're a performance marketing agency. We throw around the term roas all day long. And what, what is deceptive about the. The broad term of return on ad spend, right?
- [03:12]
A
So, like, there are many misconceptions about roas. And there are also many different things you call roas, and in many cases it's true. But different people mean different things. For example, let's forget about attribution and let's imagine you invest in marketing, you invest 100k a month, and you get $1 million a month in revenue. So some people call ROAS, like, it's total revenue that you get divided by the total spend, right? So people call it roas. And essentially it is roas. Some people say that, okay, but it's just revenue we need to care about, like profit. So this way they have their margins, they remove cost factor. And for example, out of this 1 million, they have only like, I don't know, 500k in profits and they divide profits to ad spend. It's also roas. So what exactly did you mean in this roas? And then like, so this is more objective roas from your whole marketing investment. But then attribution comes into play. And now people look into Google Ads and they say, roas of my campaign is this. But this is not real ros. This is ROS that is based on specific attribution. Then you look in your Google Analytics 4 and you see different ROAS. Then you look into your Facebook ads and you see the third version of roas. And all these roas is like everywhere different. And then you run incrementality test and it shows different roas. And then you run mmm. So like, you can, you can have the same channel with five different roas. And also you need to understand and you need to be sure, are you talking about profits here or are you talking about revenue? And finally, you have such thing as marginal roas. So most people don't even know what it is. And many people confuse it with like, profits. So it's not about profits. So like I see recently some people invented a term pause. So it's like, I don't know profits on that spend, but it's still roas. So it depends on your base, how you calculate it. But marginal roas is like, it comes from economics. Actually, maybe I'd rather put an example. I don't know whether it will be possible to make some subtitles here to explain the simple math, but imagine you invest $1,000 a day in your Google search campaign, and in return you get $2,000 a day. So your ROAS is 2x, right? So you invest 1,000, get 2,000 in return. So your ROAS is 2. Then you decide, okay, 2 is amazing ROAS. Like, I can invest even more because my target roas is one, at least for most of them. Let's imagine it's based on profit. So I put another $1,000 in this Google Ads campaign, but this additional 1,000 brings me only $500 in return. So it means like overall I invest $2,000, I get 2,500 in return, so I'm still profitable. Like, overall average roth is still 1.25. But marginal ROAS like ROAS based, like my incremental costs that I invested and incremental revenue that I've received is like, in a sense negative. It's 0.5. I'm losing money right now. For every $1 I invest, I get only 50 cents in return. So now we shift from now we have not only roas with all these different types of roas. Now we have also average roas and marginal roas. And most of the marketers and most of the ad platforms show only average roas. And it means like, in this particular situation, imagine you have two campaigns. One campaign that I've described right now, like, where you invest $2,000 and you get 2,500 in return and it gives you 1.25 average ROAS. And you have another Facebook campaign where you have 1.1 ROAS and you make a decision based on your last week data or last month data, which campaign to invest more money into. And you see, okay, so Google's Roth looks higher, so I will invest in Google. But in reality, Google's marginal roth is already 0.5, so you're already losing money. While your Facebook campaign might still scale linearly. So even though you get there. So the average rolls for Facebook campaign might be 1.1, but marginal royals is also 1.1 because it scales linearly. There are no diminishing returns. And many people, like confuse average rolls because this is what platforms show you with marginal rolls. And platforms do not show marginal rolls because they want you invest as much as possible because otherwise, if you would understand that right now you're burning money essentially, like you would stop investing even like having 2x RAWs, even though your target Ross is one, because you're already wasting money. You've already reached diminishing returns. So, so there are so many things about Ross. We can talk like the whole episode about this.
- [08:18]
B
It's such a moving target. Everything in e commerce is this moving target changes so dynamically. Attribution reminds me of the parable of the five blind men who come across an elephant and they're each feeling a different part of the elephant and they think that the elephant has, you know, oh, it's got a giant nose. Oh, it's got, you know, everyone has all these different things. And so marginal roas is the solution is, you know, is one of the solutions looking at it this way. But like, how should brands think about this in order to make sure they're not burning money improperly?
- [08:51]
A
The first fast win is to understand that when it comes to budget allocation, you should never look at average roas. Average rows, doesn't matter. So it might be that you have an amazing campaign that performs amazingly, like up to $10,000 per month. But once you start investing like it performs amazingly, you get like 5x rolls and you, and you believe, like, I can push like all my budget there. But eventually you start investing more and at this spend point, it starts, returns start diminishing greatly and you invest additional 5,000, additional 10,000. And now instead of this amazing 5x average ROAS, you have like 2x. But essentially your marginal rolls might be negative. So the first step is to forget about average rolls and never optimize your campaigns based on average rolls. But here is a trick. Like imagine you have a campaign and for the whole year, for the whole. Many companies do like this. For the whole year, you just have static daily spend. You just invest like $1,000 per day and you never, ever change budget for this campaign. And because of this, there is no way to measure marginal Roth because marginal roas can be measured only in dynamics. So if you have static budget, the only thing that you know is average ros because, like, you don't have any budget shifts. You don't know any changes. You don't know how additional spending influences additional revenue or how decrease in spending influences revenue. You don't know this. That's why you need to do this. We call it controlled budget shifts. And our platform makes this automatically. So we make this like plus, minus 15% shift depending on average rolls we have. So that I love this word. We need to destabilize a system. We need to destabilize analytics to start understanding these incremental shifts, to build derivatives and to build this very, very accurate diminishing returns curve. And now once we have this diminishing Returns curve. We can predict at any spend point what is the optimal spend so that your marginal roas is ideal and after that you just reallocate your budget based on marginal roas instead of average roas. And sometimes we had really crazy recommendations. So some of our clients initially when we released this feature, they asked like, why are you recommending me to allocate budget from our Google campaign which has 1.5 ROAS, to this Facebook campaign which has like 1.1 ROAS. So this definitely something is wrong here. And then we show them actually your Google campaign already has diminishing returns and your like real, like marginal roas is already 0.1.
- [11:38]
B
And it makes perfect sense in this example because you're talking about a demand capture platform versus a demand generation platform. And so you're, so is it, is it common that you'll find that on demand capture platforms you have diminishing roas? Maybe sooner because you're not. And then when you're, when you're advising people to spend into demand generation platforms like Meta or even you know, really top of funnel, things like television or things like that, that it actually impacts the lower funnel. Maybe if you're, maybe if you are spending more on the demand generation platforms, you can then push your marginal roas out further on your demand capture platforms like Google.
- [12:16]
A
Yeah, actually like why I've said like the first step is to understand that you should be investing based on marginal roas and not based on roas. Because at some point if you do this small thing, it sounds small, of course it requires some things to be implemented in the marketing measurement. But what we've unleashed, that actually attribution doesn't matter. So it doesn't matter whether you use last click attribution, first click attribution, multi touch attribution, and sometimes even our visitor scoring attribution. There are some campaigns where returns diminish so quickly that it doesn't even matter which attribution you use. So for example, like retargeting campaigns. Yes, last click. Of course last click is very biased towards retargeting, for example on Facebook. But once you start investing more and more and more and more money into retargeting, even though it shows amazing rows based on last click attribution, at some point additional money you spent into retargeting won't bring any additional revenue because you just exhaust your audience. And you show this creatives and this ad so many times that people just keep clicking, but they do not buy anymore. Like adding additional $1,000 won't impact. So you might still have really amazing average Ross for your retargeting based on last click. But your marginal ros can be already negative. The same thing we see for brand campaigns. So many companies worry about the incrementality of their brand campaigns. And of course we know about the cannibalization and people might be searching for your brand but they click on your ads. But this is exactly what we see with brand campaigns as well. So you keep investing more and more. And even though we know that last click is very biased towards brand but at some point brand returns diminish so greatly that even brand campaigns marginal rows based on last click is very very bad. And in this case like attribution doesn't matter and you understand like okay, I need to go upper funnel. And some clients ask when do we need to go upper funnel. So should we make like a balanced mix like you've asked, should I invest in demand gen because it scales linearly. But for example, imagine we have a company which CFO is very conservative. Like they say, I believe only last click, I believe only cookie attribution. Okay, let's use cookie attribution, but let's use marginal rows. And at some point even if you invest only in paid search, brand search retargeting, you will exhaust your audience and at some point your marginal rows will become so bad that even like other mid funnel and upper funnel campaigns on Facebook will look good even based on last fake attribution. So this actually very interesting things that we've explored a lot.
- [15:01]
B
That's super cool. And when you mentioned branding campaigns, I know a lot of the platforms with all of their AI automations, they like to include as much brand as possible to make their numbers look good as well. So there it's almost like the AI developments are sort of incentivizing you to, you know, include as much brand spending into their algorithms so that they're so that their numbers on platform look good. So that's another thing to be wary of.
- [15:26]
A
Yeah, of course, of course you should separate brand from non brand. So I believe most of the marketers already know this and they already do the same way as they separate prospecting from retargeting. So I believe this is already something we have in the marketing even for even any junior marketer know that it should be evaluated separately. Of course, like Google made an attempt to mix everything together, but market is no better. They've met a lot of resistance and a lot of critiques from marketers. So I believe they they already ruling out and actually Adding some features to exclude all this brand keywords, et cetera.
- [16:05]
B
I think you know one of the main themes on the podcast this year is figuring out how to unlock more strategic top of funnel spend that creates you know, halo effects across the lower part of your funnel. What can you give me some examples of brands that have been able to do that essentially unlock budget to bigger and better demand generation operations when they avoid spending too much saturating the bottom part of the funnel.
- [16:33]
A
What I see in the market that most of the decisions in terms of like whether invest in lower funnel upper funnel, super upper funnel like brand awareness, so called brand awareness that out of.
- [16:45]
B
Home campaigns or something.
- [16:47]
A
Yeah most of the brands they scale too fast. So I've seen DTC brands that actually invest like 200k a month and they already diversify across TikTok, Snapchat out of home TV. They explore different things but like what I've observed in the best scaling brands that with Facebook only you can scale up to 1 million per month easily. I'm not talking about but just with Facebook alone. And the idea is why like all this diverse of course I understand the importance of diversification. Like we don't want to depend on one channel. This might be not safe for whatever reasons but the idea is like more you diversify, harder it becomes for you to measure because you won't be able to run geo holdout tests. You won't be able to understand the fact on the small channel like I've seen a brand that invests like 10k a month on TikTok, 10k a month on Snapchat, 10k a month on DV360, 10k a month on paid search, 10 months on performance max and they cannot measure anything. There is just no way and when we connected our visitor scoring attribution to their DV360 campaign and to display campaigns usually this is where we see and they didn't have any conversions there. They didn't have any last click conversions from DV360 and display campaigns and they connected our visitor scoring attribution to see whether they're missing something. And we've measured these campaigns and we also didn't see any effect from there. So we say actually these campaigns are useless. And they say but we believe in brand awareness. We believe that people going to see our display campaigns and going to come back later from another device and browser and will eventually buy. This is a very naive thinking. Probably you have premium YouTube but I would imagine if you, if you wouldn't like imagine You've been watching a lot of videos today. How many brands you've never heard about that you don't care about, you've remembered to go directly afterwards. So when you're a well known brand, when you're like a Coca Cola that launched a new taste now for Christmas or for St. Patrick's Day, and now you have this green new can and you make this, this amazing ad on YouTube with Halo Effect. And after that, now people hear about this and they go to the shop and they see this can and they buy it or they go and search for it. I don't know. Usually it works for brands that already have some recognition. If this is a brand that you see for the first time ever and you see this display ads or YouTube and actually it's irrelevant and you don't care and you just want to watch your favorite podcast and just skip in five, ten minutes you're going to forget about it. So a lot of what I see, a lot of D2C brands, especially below like 1,2 million per month, they invest so much into activities they should not be investing.
- [19:50]
B
It's kind of like marketing textbook thinking. Like I should be everywhere, I should be omnichannel. D2C podcast told me to be omnichannel, so I should be everywhere rather than letting the data tell the picture.
- [20:01]
A
Yeah, it's very, for example for our company, so we are a software as a service company. So like we know exactly who are our target accounts. We can list like I don't know, 100,000 companies like in D2C space and software as a service lead generation that might need our product, that have decent spend and we can build this brand awareness promoting our content like recording podcasts, etc. Etc. Etc. When you're a D2C brand and essentially everyone can be your customer and you have a small budget, you should be very careful into how you of course like recording a podcast might be a good idea if you know your target audience very well and like if you have your niche and there is specific podcast in this niche. But if you try to run like TV campaign, you might be just like showing your product and your ads to like millions and millions of users who don't care about this right now and they will forget about your brand tomorrow just because you don't have brand recognition.
- [21:06]
B
And you don't have enough saturation and omnichannel omnipresence, like you may not be in every retail store that they'll see you on the shelf and all of these other things that Coke benefits from because they're everywhere. So that one little trigger can maybe send you to the store versus someone who's spending, you know, 50k a month isn't going to get anywhere near that coverage. I like, I like your idea. I've heard this before that, you know, you can spend a million dollars a month on Facebook easily, essentially. I'm curious, within a platform like Meta, do you see advertisers doing well from breaking out different parts of the funnel and thinking about awareness campaigns and thinking about, you know, retargeting campaigns? Or at this point is it better to just give Meta your budget and let them determine, you know, where it goes?
- [21:53]
A
In most cases, yes. So in my opinion, it might be still a good idea to separate like retargeting and prospecting campaigns just to focus Meta on reaching more new customers. But definitely I see a lot of inexperienced marketers who don't understand how meta's algorithms work. So a lot of people still create these audiences they target based on interests they build, lookalike audiences, etc. While right now Meta probably is the only platform where this is absolutely not necessary. So in Meta you can go like 100% broad. And if you properly train algorithms of meta with proper signals, meta will be able to find anyone in this world. And I was even surprised I have a dog and I was looking for a new food for dog and I was chatting with my friend in WhatsApp and like in, in 30 minutes I've seen like hundreds of ads from different shops selling like organic food for dogs, like raw food, etc. Because they process like they are inside your phone, they are inside your WhatsApp. They know about you. Like you wouldn't be able to ever target your so perceived like audience, you know, like target audience better than Meta because they use like this enormous AI and machine learning. But some people don't understand how to create these campaigns. And of course, like they just launch a new campaign and they just add some event, like purchase event all the time. They use our synthetic conversions and they believe that this campaign will start performing well. But actually it will not perform well because Meta has a lot of layers of machine learning. So I don't know whether it makes sense to go deep there. So it's like, yeah, what can you.
- [23:51]
B
Say about really understanding the meta algorithm.
- [23:54]
A
So how meta works? So Meta has many machine learning models. It's not like you launch a new campaign and like, absolutely, new machine learning model launches and it starts training based on like 50 signals you send per campaign. So Meta has layers. So it has like account level machine learning model that simply understands like learns based on all data in your account. Then it has like pixel level data. So you deploy pixel or conversions API on your specific website. So it trains on the data of this specific pixel. So even if you don't imagine you've launched like five, 10 different campaigns before and now you launch a brand new campaign, this brand new campaign will not start like brand new campaign with like new machine learning model. It will already reuse account level model, it will reuse Facebook pixel level model and more importantly it will reuse event level model. So that's why sometimes you optimize towards purchase. But then you see, oh, we don't have enough purchases, let's test some new event. Maybe we create some fancy event, combine this and that and you launch new campaign and it doesn't perform. It doesn't perform not because this event is bad, but because you don't have warmth warmed up model for this event. And that's why what I usually recommend is first if you have event that you've never used before, for example, many of our clients, especially lead generation businesses, but also like DTC businesses who optimize towards LTV for example, or only towards new customers versus existing customers, they create a new event and they just want to use this new event for optimization and they just run a new campaign. It doesn't perform. First you need to warm up your event level model and for this purpose you can create new event and you can run it only for your retargeting audience. Or in this particular case you might have combined campaign where you don't separate prospecting and retargeting. It is temporary. You just need to warm up your new event that you are running and to build initial model for it. But once your model is warmed up, you can shut this campaign down and then you can create a new campaign. Now it can be brought, now it can be excluding all your existing customers and it will ramp up really quickly and you won't need like enormous amount of events. So like meta has lots of these layers and also how you structure your campaigns, ad groups is very, very important when you understand how this machine learning works under the hood to do a.
- [26:16]
B
Website we can go super deep on all because I think it's a topic that the audience is always interested is just trying to understand the algorithm, right?
- [26:24]
A
And this is like why I've recently posted an article in my newsletter where I say I don't believe in pure software as a service in marketing science and marketing Technology and the idea behind this is today's ecosystem is so complex that non technical marketer, even if you give them software, most probably they don't know what to do this and how to apply it properly. So it's like you have a Boeing or Airbus and you've created this amazing aircraft and you just sell business class tickets and you can take a cockpit, you can fly. No, you cannot fly. Even though you've purchased a business class and you have a lot of budget and actually you want to reach your destination, you need professionals. For example us, we're in martech for more than 10 years. In many cases when we have calls with Facebook or Google support, we train them and we coach how their platform works because they have no idea. I've never, I've never seen like there were many calls with their support and it never happened that they were able to answer any question that we didn't know answer for.
- [27:38]
B
So that's a good sign. Let's, let's talk a little bit more about segment stream and how you, how you guys have evolved. Like what does it actually look like for clients using your platform on a like day to day basis?
- [27:49]
A
Yes. So at the moment segmentstream evolved just from being a visitor scoring attribution and we talked about this a lot.
- [27:56]
B
Synthetic conversions.
- [27:58]
A
It also used synthetic conversion so you can replace this last click signals for your Facebook or Google with more precise signals that can measure your upper funnel, especially when you don't have enough signals. But now we evolved into I would say a marketing intelligence platform where we can pick different modules and we can pick different instruments particular business needs. For example, DTC business might need proper attribution of upper funnel. Sometimes they might need synthetic conversions if they don't have enough or they have really long sales cycles and sell very expensive products. In many cases they need LTV scoring. So we've added this new module that you can use. Most of DTC brands need marginal roas and there are no ad platforms that can calculate this because for ad platforms this will be a disaster. The best recommendation of most of account managers in ad platforms is increase budget. And if they're going to introduce marginal roas into Google or Facebook, most of the recommendations will be decreased budget. That's why I believe this feature will never be introduced in this platform. So we see that for most of the DTC brands, for example, this is one of the most important things because profit matters. So if you have profitable marketing mix, you can scale more and you can sell more eventually. So we added this huge component of automatic budget allocation and calculation of marginal rows by these controlled budget shifts and building this very accurate diminishing returns curves. We also added lead scoring. So now we work a lot with lead generation businesses, especially with longer sales cycles. And for them of course, like if you have a lead, if you're like an insurance company, whatever, you have a lead. And sometimes, or real estate or travel.
- [29:54]
B
We'Re working with a flooring company that you get a sample shipped to you as part of the lead process.
- [30:00]
A
Right. And sometimes it takes like few months for the lead to convert and different leads convert at different rate and different leads have different order value. So you need to understand like how to quantify these leads. So not just to optimize to maximum volume of these leads. Sometimes really bad leads that do not convert and Facebook will try to buy them cheaper, but instead sometimes you want to buy more expensive leads which convert with much higher probability. So this is a lead scoring. So essentially our platform is like marketing intelligence platform for predictive marketing. So we predict like lead values, we predict ltv, we make visit scoring. So essentially it's a scoring technology, visit scoring, lead scoring, LTV scoring, predicting marginal Ross, predicting proper budget allocation for the next week. So it's like a whole suit combined with our expertise because as I've mentioned, we tried a lot and we've got a lot of frustration when just to save money, some of the clients decided just to use our software without our services and without our expertise. And in many cases they failed and they were not able to use even 10% of the technology. So right now we shifted the paradigm and right now if you want to use our platform, probably this is our unique approach. It goes together with our expertise, with our consulting, with our team. So essentially we become an extension of the team and we work together with all these aircraft so you can fly.
- [31:38]
B
The 747 and don't crash.
- [31:40]
A
Yeah, I have many examples how Boeing crashed when it comes to marketing science.
- [31:46]
B
Yeah, give me one.
- [31:48]
A
Yeah, I will give you an example. So we've had a self serve client and essentially they were in education space. So actually it's like higher education where you can apply from anywhere in the world. It's a lead generation business and of course it's a long application process and after 3 months, finally you pay for like 50k or 100k depends on the course you're picking. We've built an attribution for them and they started optimization based on marginal analytics. And instead of optimizing towards leads, instead of implementing lead scoring and optimizing towards lead value. They decided just to use target CPA and optimize toward the number of leads. And eventually our platform is fully automated so it understands what is the best campaign to allocate the budget, how much predicted amount of leads. And eventually they started following recommendations of the platform. And in like a few months they've got enormous amount of leads. Like they've never seen this amount of leads before. Most of these leads were coming from India, Saudi Arabia. While like amount of leads from us like decreased significantly, amount of leads from UK decreased significantly. And platform did exactly what they asked it to do. We want to maximize number of leads. Our goal is lead and we want the smallest price per lead. And this is how you use Boeing and you crash because this is the wrong objective. Technology will always do whatever you tell it to do. With AI, a lot of people right now are afraid about whether AI is good or bad for humanity. It depends what will be objective, like how you're going to train this model. That's why Elon Musk emphasizes a lot about how we train this model. What is the objective like, because otherwise.
- [33:40]
B
Like when you ask a genie for a wish, they always find, you know, you can be very specific about what you want, but often the genie will give you, you know, the implication of what you're asking for rather than maybe the specific thing you're not. We don't take into account the entire, the entire picture.
- [33:54]
A
Right. If you're going to train your AI like we want better ecology, we want to solve different problems with pollution. The best way is to get rid of humanity.
- [34:03]
B
Yeah, that's right. We need to maximize widgets, so get rid of the humans that, that get in the way. So we're coming up on time here, but this is such an interesting conversation. Just walk me through like a brand who's spending say 50 to 250,000 right now, which I think is a lot of the, maybe the sweet spot of the listeners in the audience. What, what, what's the best way that they should think about, you know, getting started with marginal attribution aside from giving you a call and subscribing to your newsletter.
- [34:30]
A
Yeah. So first of all, if you have like in DTC space, I would say it's not a huge spending, so it's better not to diversify at all. So if this is like a low margin product, probably paid search might be very expensive and clicks might be very expensive. While on the other hand, if you have huge like, big like huge average order value, paid search might be a good Channel to start. But the next best channel, for now, at least from my experience and from what I see with our clients, is of course Facebook. And with this spend point, never ever diversify and never ever have many campaigns because like for smaller budget, in many cases Facebook will perform worse compared to larger budget. So if you invest like 50k on Facebook, 50k on TikTok, 50k on Snapchat, 50k on YouTube, this is how machine learning works. To build a robust machine learning model and for Facebook to build this proper segment, proper lookalike audience based on the signals you send, you need to send as many signals as possible. So if you just invest 50k and you have just like, I don't know, 500 conversions, it might be not enough for Facebook to find a good audience. That's why first you need to invest as much as possible into one channel, into one campaign, and also to build marginal returns. For example, it's also not enough to have like 5 or 10 or 15 conversions per campaign because of fluctuations week to week. So it's another like huge mistake that I see everywhere. When marketers, they look at the performance of the campaign for the last month or for the last week and they say okay, performance for the last week, like this campaign just brought 10 conversions, it's not that good, let's decrease budget. And next week they see that the same campaign, even with a smaller budget brought 20 conversions. Like what is going on? We decreased budget, number of conversions increased. But this is just idea of small numbers. Numbers fluctuate all the time. So that's why it's also not a good idea to look at historical data. You need to build, we call it derivatives, you need to understand how to properly build this curves and to be able to predict your revenue next week. And for this you need enough signals. And if you diversify, if you create many different campaigns with five or ten purchases per campaign, you won't be able to do any optimization. So in my opinion, diversifying too early can really hurt your business. Even though everywhere we hear about it now, what happens?
- [37:05]
B
How do you start to tell when you are seeing diminishing returns like Ridge maybe or one of these huge companies that is spending a million plus? What are the signs you're looking for on a platform like Meta when you are starting to saturate and you need to look to other funnels? And my next question would be, what do you see brands doing successfully when they do poke their head out of the meta ecosystem? What other platforms are working?
- [37:27]
A
Yes, this is a very tricky question because Imagine you've had a campaign which performed really great like 5x raws for a long time up to 200k budget. It performed really well. But now for some reason the audience was exhausted or for some reason like this campaign cannot perform and cannot scale anymore and it starts showing these ads to the same audience and your returns might be diminishing quite rapidly. You already might be having like 0, 0.9, 0.5 marginal ROAS while in average Ross, you won't be able to detect this because it was so good for such a huge budget. It was 5x. Now you invest additional money, now it's like 4.8, now it's 4.7. So like average ROAS changes not that dramatically compared to marginal roas. And when you just look into the platform, in many cases you won't be able to see this because all these platforms, they don't have this metric. So you need to either track this manually and you need to track all your budget sheets. Okay, this week we increased budget by $1,000. How did it impact our revenue? Or month by month? Or use platforms like Segment Stream where this can be done automatically. But unfortunately like it's very hard just to look at the numbers inside the platform and like quantify like extract this marginal rows from average rows because all you see is just marginal Ross. And in many cases is just based on some attribution and I don't know an easy way to extract this besides like very careful tracking of all budget shifts and measuring this increment in your revenue.
- [39:07]
B
If you are seeing this, if you're using your, your platform and you see okay, my marginal roas is decreasing. Like what other platforms are you seeing brands have success with?
- [39:17]
A
In my opinion, like it's not about the platform. So like that's why I don't like, like average numbers. Like some people say on average people started investing here and now they see great success. It's like saying on average like 10,000 years ago people lived till 30 years old and now we live till 80. But no one says that mortality was mostly like before you get five or three years old. So it was mostly child mortality. But Once you reach 10 years old, your possibility you were possible to reach like 80 and even 100 years old. So like that's why this like I don't like these averages. Like everything depends for D2C brands. I see a lot of success on Meta, especially when you use it properly, when you feed it with correct signals, when you understand marginal roas, when you separate new customers and returning customers. When you do proper LTV predictions and feed this data back to meta platforms when you use synthetic conversions for Opera. So like if you use a tech properly and you go all in in one channel, master this channel and make it work in many cases I see much better results compared to let's run everything everywhere and hope that something going to like you're going to get these average results everywhere.
- [40:41]
B
Meta loves this answer.
- [40:43]
A
Yeah, it might be not Meta for some businesses, it might be Google for some businesses. If your audience is really young audience it might be TikTok or Snapchat. So it depends, it depends completely on what you are selling to who you are selling and where you have expertise.
- [40:59]
B
How far are the other algorithms behind where the meta algorithm is, would you guess? I don't maybe. That's hard to say.
- [41:06]
A
Very far.
- [41:07]
B
They're very far. Yeah. Meta is just worlds ahead. Google's the one that would be closest.
- [41:12]
A
Absolutely not. So with Google it's easier because with paid search it's more like intent based.
- [41:19]
B
That's right, Demand capture.
- [41:21]
A
Write a keyword, buy this particular dress or this brand and this model and it shows you ads. So it's really good in this while when it comes to finding demand finding people who might need your product right now even though they haven't requested it explicitly. So I don't know any other platform that has so much data that meta has and we tested like many of our clients tested demand gen on Google Display. It's very hard to get similar results. LinkedIn is really good in terms of targeting but it's mostly for B2B business.
- [41:56]
B
B2B. Yeah, yeah, that's interesting.
- [41:59]
A
But it's very hard to compare anything to matter at this moment.
- [42:03]
B
I have to ask you about creative. Like I know you're mainly on the data side but what is your insight around like people are creating, you know, depending on their scale, 50 new creatives a week. How do you view creative in this, in this process?
- [42:19]
A
So frankly speaking, yeah, we're exactly on the data side. So we are not involved in launching any campaigns for our clients design. So all of this is done in house or by digital agencies for sure. But of course I know how meta works and Matter also has a model on creative level. So again all your creatives like Meta has specific models that are linked to your creatives and of course like if you have really well performing creatives combined with event that you've been using in another campaign, when you launch an absolutely new campaign with new strategy, of course you should be reusing creatives that are performing well because you already have pre trained model for this creative. The same way you can test your creative first in campaigns where you have a lot of budget. So like some companies love to create this separate test campaign and to test new creatives. And the problem is yes, you have account level machine learning model, you have pixel level machine learning model and even event level machine learning model. But you don't have campaign level machine learning model and you don't have enough budget for this test campaign to properly build new creative model. And that's why it's better just to add this creative into campaign that is already running. There is much more chance to build a good model for this campaign and if this creative doesn't work and cannot outperform your existing, that's fine, it's not a problem. And that's why usually these test campaigns are very slow to train and usually it's very hard to get results that are better than your top performing. And many clients force budget like they know you should spend budget on this test campaign and here you can fail and you can lose. Maybe it was a right bet and you invest it like and still like this campaign will be underperforming your existing campaign simply because the model is still not trained. So yeah, probably it's like if you.
- [44:15]
B
Try to pick stocks versus an etf, you're going to lose, you're not going to beat the ETF most of the time.
- [44:20]
A
Right. It's risky. But again, I don't see any risk. You can just add this creative to your existing campaign and see whether it flies or dies.
- [44:29]
B
Yeah. Or crashes. I think this is good news. I think a lot of times people are talking about, okay, in the next two to three years, years media buyers will be, you know, replaced by AI. But with what you're saying, these platforms are not incentivized to be thinking about, you know, be thinking about marginal roas in the way that you're talking about it. Do you have any predictions for the media buying profession and the media buyers in the next two to three years?
- [44:53]
A
Yeah. So I believe the biggest issue in this market is again your objective, like Meta has its own objective. Digital marketing agency has its own objective. In many cases, marketing teams have their own objectives. Imagine CMO who last year invested, I don't know, half a million in TV and now needs to create a report for CFO at the end of the year. And their objective is not to be fired. Right. And in this case they will find some measurement technology that gonna justify that gonna justify their investment, they don't care about the truth, they don't care about understanding whether it was really incremental. They don't care that they didn't grow year over year. What they need is to find someone, some consultant or some technology that's gonna justify, for example, MMM is a good one, that this TV was a good investment. So even within the brand team you have people with different objectives and in my opinion like it all depends on the objective. And yes, machine learning will definitely improve and automate a lot of processes but you should be always cautious. Like with Google, with Meta they won't optimize a lot but they want to take premium and if you're acting smart you can fine tune and you can use third party technology which is not affiliated with Google and Meta even though like platforms like Google, Meta, TikTok, they try to go into third parties and also build partnerships and also build like some affiliation to prove that this works better than this. But in my opinion this is about objective and whether you are truth seeking and whether you're on brand side and you want to squeeze as much profits as possible from your budget or you are on ad platform site or you're an agency that just that gets commission of the budget and you also wanted to scale, scale, increase budget. So it all depends.
- [46:54]
B
Super cool. I'm going to introduce you to. I don't know if you've ever talked with Pilot House, but I would love to. I'm going to introduce you to our head of meta, Taylor and Dave, because I think they could really geek out with you on a lot of this stuff. But if you want to get smarter about your media buying, you've got to go to segmentstream.com and talk to an expert. Any other advice for people to basically get started with segment stream?
- [47:18]
A
Yeah, maybe a good point is like my marketing right now at this stage is education. So even there are so many complex things and right now even if you're still in this exploration and you're trying etc. It's always good idea to connect with me on LinkedIn and subscribe to my newsletter and maybe Eric, you will be able to add this in description. I don't know for sure. I understood that I want to post authentic content. It's not, it's not our company blog, it's my personal blog where I just share everything I learn hoping that at some point I will be able to import all of this into AI that will @ some point replace me in my company and then it's your exit.
- [48:03]
B
Strategy then you can go.
- [48:04]
A
This is my exit strategy.
- [48:05]
B
You can go to join Konstantin on the White Lotus and to have a.
- [48:09]
A
AI powered customer success manager.
- [48:11]
B
I love it. Well, I always love our conversations. I hope we can do it again soon. So much fun having you on the podcast. Yeah, add all your links in there. Go subscribe to Konstantin's newsletter as well. Check out his awesome posts on LinkedIn and talk to an expert on segmentstream.com super fun. Thanks again for geeking out with me on this.
- [48:29]
A
Thank you very much. Eric.
- [48:36]
B
Thanks so much for listening to today's episode. If you're not a subscriber to our newsletter, you can do that right now at directtoconsumeralloneword co. I'm Eric Dick and this has been the DTC podcast. We'll see you next time.