Earlier this week we ran a "how satisfied are you with your tracking?" survey.
Results are in:
On the whole, our participants gave their tracking/attribution solution a Net Promoter Score (NPS) of -14.
That is really bad.
(The Net Promoter Score (NPS) is the world’s leading metric for measuring customer loyalty & satisfaction. A NPS of below 0 indicates a company is in the bottom 25% of customer satisfaction.)
Long story short: If you're reading this, there's a good chance your tracking & attribution setup just isn't cutting it.
And that's why we're going to be going deep on tracking & attribution content & training at VidTao.
...Including going into detail on what we've been doing at our agency Inceptly to level up tracking & attribution in this post iOS 14 world.
Stay tuned for much, much more. But in the meantime:
Step 1: Go and follow our VidTao & Inceptly Co-Founder Brat Vukovich on Twitter, here: https://twitter.com/Brat55136783. He's sharing a ton of gold on tracking, attribution, and marketing in general.
Step 2: Read the post below from our Inceptly agency Data Scientist Dr. Dobroslav Slijepcevic. (btw Meta did a big summit on the attribution methodology Dobroslav talks about, literally today. Here is a great summary: https://getrecast.com/meta-mmm-summit/)
Let's get started...
A Short Introduction to Marketing Mix Modeling (MMM)
Hello to all, let me introduce myself to you. My name is Dobroslav, and I’m an astrophysicist making his way through the wonders and mysteries of the marketing world as a data analyst for Inceptly Agency.
For those who don’t know anything about MMM, I’d like to say a few words and try to make you familiar with it and to be able to understand its massive value for marketers generally.
Basically, MMM is a statistical analysis that helps us estimate our marketing channels' actual impact on our KPI (or whatever we define as KPI, revenue, sales, leads, or something else) and predict future performance based on past data. And, for things to be even better, you don’t need to know anything about your customers, and there’s no need to track them, thinking about the first click attribution or the last click attribution, the only thing you’re interested in is that the sale or any conversion is made, and it had a gross on your revenue!
Of course, you know what is your ad spend divided on different channels, and, basically, it’s all you need to make your first simple marketing mix model!
If you really think about it, marketing existed before the internet, and marketers had to develop tools that help them to calculate the ROAS of their marketing activities. It wasn’t so simple, there’s no way to directly measure how many customers watched TV ads or listened to Radio and decided to buy your product. In that case, you needed an econometrician (or simpler said - a mathematician) to help you to solve the problem.
Today, all marketing platforms have ways to track customers, but the number of possibilities for something to go wrong is not negligible, let us only mention IOS 14 which doesn’t allow you to constantly study your potential customer’s behavior on the internet and target them with your ads.
If you add the constant problem of “stealing” the credits for the conversions, when you never can be sure about it which marketing channel is the one that was crucial in the customer’s decision to buy (well-known problem, the customer sees a youtube video then sees an ad on Facebook, and, 2 days after, he or she purchased a product via a google search. And which channel was the most important and make a customer from a potential customer? Ok, data-driven attribution is here to help you, but, as I said before, MMM doesn’t need any tracking, which changes things from the core, and gives you a totally different perspective of the whole problem.
Of course, wise people will always look at things from multiple angles and decide after looking at all possible solutions. Because, as we all know, total truth does not exist, it is always somewhere in between. Only pure knowledge about multiple things leads to success.
What do we need to make our first model?
Welcome again, my friends. This is the second part of my short course about the basics of MMM. If we even want to talk about modeling, first, we need to define our KPI. There can be only one for a certain model. Generally, we choose the revenue, sales, leads, or something like that. All of our marketing activities have one goal: spread brand awareness and get income in our KPI numbers! It seems so simple. We are searching for new customers or we want to remind the old ones that we still exist, and, for that purpose, we are targeting them with ads in many possible ways, through google, youtube, Facebook, TikTok, mailing lists, or through the old fashion ways like TV, Radio, Billboards, phone calls, etc… Many of these marketing channels are paid, but some of them require only goodwill and a strong database.
So, the most important parts of the dataset for our first model are our KPI variable (for example, revenue), ad spends for different channels and non-paid marketing activities. They should be given as weekly data, or even monthly data since there is a lot of noise in daily data to try to predict anything, have in mind that we try to model real people’s behavior, and it is always a big challenge for all kinds of science. And marketing mix modeling is nothing more or less than a mixture of science and art!
Is this all? Well… The answer is no! The real power of marketing mix modeling lies in its ability to involve many external factors that undoubtedly affect our KPI. Just think about it, if you are the Ice Cream factory manager, independently of your advertising activities, you must rely on the temperature and season changes. An increase in temperature will for sure affect your sales, but for how much? What is the real impact of it on your sales? Believe it or not, marketing mix modeling can get you an answer!
Or another example, a few years ago, the world changed irreversibly - Covid 19 entered the stage. Do you think that there is one single company unaffected by it? Rather positive or negative? Personally, I doubt it! Does MMM have a tool to measure this, and turn the empty words into precise numbers? Yes!
And what about the Ukraine war? A good data scientist will find a way to incorporate it and all the relevant external factors into the model.
But for now, to keep things simple, we will work only with a few variables, one for the time (week), one as a variable (KPI) we want to predict, and a few of them as the input variables (paid advertising channels and one for promotions). The simplified and edited version of the dataset looks like this ( we have 51 weeks of data, 52 rows in total, including the names of columns, and they are arranged in a proper way in the following table which can be called ANALYTICAL BASE TABLE):
But, if you ever want to build your model, know that the real data never comes in the form you need, about 70% of the time you need to build any model is about preparing and cleaning data. Anyway, it’s completely another story, and I would need two extra courses to talk about it, but you should know that.
And also, if you ever want to exercise all of this with some datasets, there are a bunch of them on the net and you can download them from specialized websites like Kaggle, but I personally recommend the code developed by Meta in R language called MMM simulator, which is written especially for simulating MMM datasets and it is completely free for use and very simple.
And now... Her royal highness - mathematics
Before we jump into action, we need to do a little math. One of the greatest mathematicians ever, Carl Freidrich Gauss used to say that “Mathematics is the queen of sciences” and who are we to argue with one of the smartest people who ever lived on our precious Earth? Don’t worry, if you are not familiar with math, I’m sure you will easily adopt simple concepts that will be discussed.
Every one of us has likely heard about linear functions. It is the simplest class of the functions which comes in the form: y=ax+b. Here, y is the dependent variable, x is the independent variable, and a and b are the constants. Constant a is known as a slope, which shows us the ratio between the change in y for a unit change in x, while constant b is known as the y-intercept, it shows us what is the value of the y variable when x=0.
If we know the values of a and b, we know how the variable y changes if we change the value of x. Graphically, this relationship between y and x can be shown as a straight line. Every point of the line represents one (x,y) pair which satisfies the equation y=ax+b for known values of a and b.
I assume that nobody ever counted the number of scientific laws that have a linear form… There’s no part of any science that doesn’t use this simplest function to describe the relationships between two variables. And, so do we, let us use the linear function to model the dependence between, for example, revenue and ad spend.
If you find yourself confused, again, have in mind that it is simple math, but not everybody likes it at all… Let me translate it into the marketing language:
- y is revenue or some other KPI. We want to predict it and explain how it changes in dependence on x, and x is the sum of our marketing activities (ad spending).
- In this context, b represents our baseline, the part of our revenue achieved without any marketing activities.
- Then, a is, in fact, the ROAS of our activities, return on ad spend, it shows us how much money we earned for every currency unit (for example, dollar) spent on advertising.
Let us say that we spent 13k dollars this week on advertising, and our ROAS is 1.60, which means that we earn $1.60 for every dollar spent on advertising. If our weekly baseline is $20,000, then our total revenue is:
Revenue = 1.60 × $13,000 + $20,000=$40,800
In marketing mix modeling, two of our total goals are to capture the ROAS of each marketing channel and to determine what is our baseline! Therefore, we go in the reverse order, we know what is our revenue and what is our ad spend, and we must determine ROAS and baseline. Sounds simple, isn’t it?
Last but not least important, let me introduce a linear regression to you. This is a statistical method that allows you to do exactly what we need - to provide you the constants a and b if we have a dataset of various (x,y) pairs! What is the catch - in the real world we can’t expect a perfect linear relationship between revenue and ad spending, there is always more or less noise in their values. Thus, the baseline and ROAS we find with linear regression aren’t some ground truth, they are only the best we can do to describe the real situation with the lowest error. What does it mean, you can see in the following figures:
None of the real (y - Revenue, x - Ad spend) points lie on a straight line, but every point lies near the line. But the revenue which the model predicts for certain amounts of ad spends lies exactly on the line. Hence, every revenue that the model predicts must have a deviation from the real value. Our job as statisticians is to evaluate this deviation and somehow determine if the model valuable to us or not. Or as statistician George Box said: “All models are wrong, but some are useful”.
Are models in the real world linear at all? Some of them, yes. But it is not the rule we can use in every situation. For a good marketing mix model, you’ll probably need to do more mathematical exhibitions, with a lot of coding. But if you understand linear regression, and know how to interpret its results, it is your first step in entering this magical world of statistics and modeling. If nothing else, you’ll be able to read and interpret the results of someone’s more complicated models and apply them in the real world.
Stay tuned for part 2 on this topic coming soon!
Dr. Dobroslav Slijepcevic
Data Scientist at Inceptly.com
*** VidTao Premium eCommerce YouTube ad training ***
At our agency Inceptly we have managed over $150 million USD on YouTube, have a team of 30+ including account managers, live action production team, creative managers & copywriters, editors/animators and many more.
It would be great if we could go into even more detail on what makes a successful eCommerce YouTube ad:
* From pre-production
* To production
* To post production
* To campaign setup, testing, launch, iteration and scale
And that's why this week, inside VidTao Premium, our Creative Director Aleksa Simic is going to do just that.
The training is going to be inside the VidTao Premium app and will be for VidTao Premium members only.
Go here: https://vidtao.com
What are your YouTube ad questions?
In the meantime though, what questions do YOU have about YouTube ads?
Let us know in the Comments section below, and we’ll make sure to cover your question in an upcoming post.
Have a nice day!
The VidTao Team
Sign up for VidTao.com: The Free Tool to Discover & Track Your Market's Best YouTube Ads.
Are you spending over $1k/day on paid traffic and want to scale with YouTube ads? Schedule your free YouTube ad brainstorming call here: inceptly.com/call