Hello, I'm Matthias. I'm with Windsor.Ai. I'm joined by Enes, who
is also from Windsor.ai. We had our customers asking us
what is the difference between a MTA, a multi-touch attribution
model and a marketing mix model. This question has been
popping up many times. And today we will try and
provide a simple answer. So let's start with MTA. Now MTA is the way you measure
the different touch points. Your customers have. So let's start
with touch points: As a user, before doing a purchase or a form fill on a
website, I have multiple touch points. I can only have one, but
let's assume I have a few. So let's say. Buying a sneaker. The first touch point
I have is a display ad. I'm being targeted. The second touch point I
have is being retargeted. Then I click on a display ad again. And the third touch point is
I then search for the brand. And click on a brand search. So it's a branded paid search. And the last touchpoint
is an organic search. So now attribution is the way
you give credit to each of the touchpoints which are present. So with the first touchpoint, I would
give all the credit to the first touchpoint, which was the display ad. Very simple. With the last touchpoint attribution. We give all the credit
to the last touchpoint. And now I would like to expand
a little bit on that now. Last touchpoint was used for a good
10 plus years during the days of GA UA because whenever you looked at GA UA
Google Analytics: Universal Analytics report, you will be presented by last
touch numbers and it's sub ideal. Because you're only giving credit on
the last touch, meaning that you're leaving out everything which happened
before, upper funnel channels, which are. Display or paid social, which are often
not the closers were completely left out. Of course, it's really to the
advantage of channels like Google organic or Google paid search, because
that is already very intent driven. So you're very likely to convert
when you search for Nike running shoe, or in my case, I like ASICs
when you look for ASICs running shoes, because you're ready to buy. So that is the world until
2023, and now we go into the data driven attribution GA4. When it started, it came with the last
touch attribution and now Google started switching into data driven attribution. Now we work with a known data driven
model, which is called a Markov model. And a data driven model should
help us to make it better decision on how we should reallocate the
credits for each touchpoint. So instead of looking at it in a very
simplistic fashion and give credit to the last touchpoint or the first touchpoint
only, we want to get as close as possible to the real customer journey. So we look at what the historical
probability is that a channel impacts an outcome, meaning in the context
of performance marketing or digital analytics, we look at each touch
point and try to distribute the credits based on the likelihood that
the conversion is going to happen. So let's say we have. 200 converters and we
have 1000 non converters. We then model the data and have the model
tell us that in case we run a simulation and let's say we remove Facebook ads
from all the historical conversion journeys in that simulation, what would
the overall impact be to conversions? And then based on that, we give
credit to each of the touchpoints based on that simulation. It's a much cleaner way. It's more complicated, but it's
a cleaner way of giving credit to each touchpoint instead of a human
deciding: now we give credit only to the last touchpoint, or now we give
credit only to the first touchpoint. So it helps to improve decision
making .Now the drawbacks are that MTA is getting harder and harder
and harder because you have a lot of the wall gardens, not allowing you
to, for example, impression track. I mean, Facebook ads and Google ads. So in 2023, as we record this, you
can only do an MTA based on clicks. So that means you lose all the
impression data because Facebook decided in 2017 that they will
stop ad serving impressions. And therefore you can
only work with click data. it would not make sense to use
impression data on one channel and only click data on another channel because
then , you will skew the whole model. So basically everyone in 2023
is using a click based approach. So that's one thing. Then the second thing is that
now cookies are harder and we have cookie consent and GDPR. In order to do an MTA, you need to have
users allowing your cookie banners. So everyone who clicks reject
cookies, they will not be able to be tracked with an MTA. We work with companies across
the globe, but there are certain markets where the probability that
the user clicks no is much higher. And for example, in places like Germany,
we have seen, depending on the audience that you can have up to 40% off the
transactions without a trackable journey. So it's then questionable whether one
can rely on such a data set to make decisions and how to reallocate budgets. Now, in light of this and the whole
advance of AI, MMM is having a return. MMM Is the way of correlating spend
to revenue instead of looking what touchpoints caused the revenue. I will hand over to
Enes as the expert here. Thanks. Media Mix Modeling, or MMM, is a
method used for tracking and measuring the overall performance of your model. You can do this by playing with
different parameters and seeing how they affect the overall advertising result. For example, suppose that I run a
car dealership and I have various ads across different platforms. I also have the performance
of each of these platforms. I have metrics such as sales, how
much I spent on each platform, the country city, et cetera, which is
great, but I want something more. I want to analyze my data further. I want to set a goal. For example, I want to set a goal of
selling 200 cars in the next six months. By using MMMs, I can test the model
with various parameters, and I can say if I increase my spending in London by
10%, how many more cars will I sell in general, or if I increase my spending
on Facebook, does that give me better results, or should I stick with Google? So it gives us the chance as
managers, to see if my changes are affecting the overall performance
of my marketing strategies. Some of the effects that MMMs can measure
are seasonality or other external factors. In this case, let's say weather. Seasonality. So suppose that I have a ice cream
shop, and I will see that throughout time I will sell more ice cream
during summers than during winters. I will see that my data always
peaks during summers, but during the winters, then I lose a lot of sales. MMMs can take into
consideration seasonality. They can tell us, Hey, during this
season, you are getting more sales. So we are not skewing and we
are not causing issues with our final model prediction. Another thing that MMMs are
good at is external factors. For example, we can incorporate
how different parameters can affect our final result. In this case, let's say I sell umbrellas
and I will see that during rainy days I am selling much more than during sunny days. In this case, I am taking into
consideration an external factor, something that in other models will
not be taken into consideration. So now that we know what MTA is and MMM is. The real question is when to use which
and how to use these different strategies. We know that MTA is good in cases
when you have granular data. It's good when you actually
have impressions, likes, clicks. So you can actually analyze
the journey of each user. However, as Matthias mentioned earlier,
the disadvantage here is that not all the users are being tracked anymore. And other disadvantages that you can
track only ads, which are online. Let's say that you are having some
ads on the spheres of marketing, let's say TV, radio, and so on. You cannot track them . So the idea
is that if you have data which you can use to analyze your users, such as
likes impressions, you can definitely use MTAs but use it with caution. If you want to also include into
your analysis a different factor, that is MMMs so in that case, you
can also use them because they are good to set company goals. Let's say if you want to have a prediction
for the future to estimate how each ad is affecting the overall sales performance. The advantage here is that MMMs give
you something good without much of data. You don't need all these small different
parameters to estimate the final model. However, the issue is that you don't
know how your user is behaving. For example, you don't know which ad
is actually causing more sales, which platform is causing more sales, and so on. So, use MTAs mostly if you have
user based data and granular data. Use MMMs. When you want to make
predictions for the future. Now last, I just wanted to mention that
we have a different or another marketing strategy called unified marketing
measurement, which actually incorporates both of these methods, MTAs and MMMs. The information that you lose
from one platform, you can actually gain from the other one. And this is great, because you
can know more about your users. However, the disadvantage is that , your
analysis becomes more complex, because you need to take into consideration all
these different pattern mirrors coming from different methods and strategies. So, that's all we had for
this small little demo. I hope that we answered
most of your questions. Back
to Matthias. Thanks, Enes. Thanks very much for watching. If you do like that please
subscribe and leave us a comment if there is any question, we'd
be happy to answer it for you. Thanks very much.