June 25, 2025 • 29 mins
In this episode of 'Digital Coffee Marketing Brew,' host Brett Deister sits down with Zeke Camusio, a seasoned entrepreneur and founder of Data Speaks, an AI-powered analytics platform. They discuss the importance of data-driven decision-making in advertising, challenges in marketing attribution, and the impact of third-party cookie deprecation. Zeke also shares detailed insights into different attribution methods, including media mix modeling and incrementality experiments, and how AI can enhance marketing analytics. The episode concludes with practical advice for small businesses and a look at the future of data privacy in marketing.
3 Fun Facts:
- Zeke Camusio drinks both coffee and tea—coffee in the morning and tea at night—but prefers strong Italian espresso, especially Lavazza!
- A whopping 61% of conversions are shared by more than one advertising platform, making attribution really tricky.
- Before the internet, marketers still ran ad experiments using incrementality testing and media mix modeling—old-school marketing science!
Key Themes:
- Challenges with marketing attribution accuracy
- Importance of data-driven decision making
- Reliability issues in platform-provided metrics
- The impact of third-party cookie deprecation
- Benefits and limitations of AI in attribution
- Cost-effective attribution strategies for small businesses
- Evolving privacy regulations’ influence on marketing
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Five out of six marketers believes it's absolutely essential
to make data driven decisions to get the best results
from your advertising. Only one in six actually
trusts the data that you use to make
decisions. And at the end of the day, so much of what
we do is doing things well, but so much of the
(00:22):
impact we cause is by choosing the right things.
But 61% of conversions are shared by more than one
platform and then 58% of conversions are non
incremental. So they would have happened anyway without the ad.
(00:49):
And welcome to a new
episode of Digital Coffee Marketing Brew. And I'm your host,
Brett Deister. If you please subscribe to this podcast and all your favorite podcasting apps,
leave a five star review really does help with the rankings and let me
know how I am doing. But this week
I have Zeke with me and he is a
(01:11):
serial entrepreneur, the founder of Data Speaks, an AI
powered analytics platform that helps companies
identify what drives her sales and invest in the right
marketing strategies. And he has a background in economics
and data science. Zeke has spent the last 20 years developing AI machine
learning and data analytics solution, embedding hundreds of companies
(01:33):
to make data driven decisions and accelerate growth.
So welcome to the show, Zeke. Thank you, Brad. Thank you for having me. All
right, the first question is all my guest is, are you a coffee or tea
drinker? I'm both. I have coffee in the morning
and then usually tea before going to, going to bed.
(01:53):
But I would say that I drink coffee way more than a drink
tea. Do you have like a specific favorite of coffee or is it just
whatever you can get your hands on? I love
Italian coffee, espresso roasts.
I like usually
medium roast, South American
(02:17):
coffee brewed in Italy if it makes sense. You
know, so there's a couple of brands, Lavazza that
I, you know, really, really like is smooth and, and
just strong but not too strong. Yeah. So you specifically like
espresso? Because when I hear Italian coffee, it's basically always espresso.
Yeah, yeah, it, it's. I,
(02:40):
you know, was born and raised in Argentina. We have a lot of, you know,
Italian influence and it's just stronger, you know. So if you go to
Starb, you know, any kind of coffee
shop, it usually feels like watered down to anybody
who's from either Italy or Argentina. So I like
my coffee a little stronger. So you more like the double shot or
(03:02):
like when it goes into your. It's more of a double shot than a single
shot because I think Starbucks does a single shot and you
usually like the double shot or more? I would say
yeah, and. Yeah, and
a little darker rose than most people have it.
Fair enough. But I gave a brief summary of your expertise. Can you give our
(03:23):
listeners a little bit more about what you do? Sure. So
historically, I would say I'm a serial entrepreneur. I've
had many businesses in the tech space.
Right now I'm the CEO and founder of Data Speaks. You
kind of touched on that. But we help marketing teams
(03:44):
understand what channels and campaigns drive their sales
so they can invest in the right marketing strategies.
Gotcha. Can you explain what marketing attribution is and why it's such
a crucial, yet challenging area for marketers? Yeah, of course.
So what happens is when you are
advertising a lot, you know, you have a presence on multiple channels,
(04:06):
you're running video ads on YouTube, you are
on search, Google, Bing, you're doing
social, Snapchat, Meta, Instagram,
Pinterest, and so on. So you're doing all these different
things. And then at the end of the day, your store sells,
say, you know, $200,000 a week of
(04:28):
widgets. And what you want to know is you want to understand
how much of that was influenced by each channel
and campaign, because otherwise it's,
you know, money going out, money coming in, but you're not really
understanding the ROI of each of these investments. So
what we found is that in the US 54%
(04:51):
of marketing budgets are spent on ads. So more than
half. And while five out of six
marketers believes it's absolutely essential to
make data driven decisions to get the best results
from your advertising, only one in six actually
trusts their the data that you use to make
(05:12):
decisions. And that's because they put all these pixels on the
website for Google Ads, Facebook ads, and
they make decisions based on those. And those can potentially
be a source of truth because each of those only
sees activity from that platform. So if the Facebook
ads, pixel sees activity from Facebook, but
(05:35):
it doesn't really know that you're also running ads on Google, that you
are sending emails, text messages. So you
really need an independent way to track all the activity and
how that then impacts your bottom line.
So what you're basically saying is it's fragmented data that you need to
centralize so you can understand exactly where everything is
(05:58):
going. It's kind of like podcasting, because podcasting
there is a lot of different third party data, but it's not all
centralized either. So I have to look at three or four different ones
just to figure out, like, what's going on. And the thing is, like Even,
even if you were so right in that it's
really a puzzle and each of those pieces helps you see the full
(06:20):
picture. But none of those can
help you figure out the actual revenue it drove. Because you
need to understand, essentially, it's called incrementality.
You want to understand the incremental impact of your ads. So if you go
and increase your spend
by say $10,000 a month for
(06:43):
YouTube, you start running more YouTube ads. You want to understand
how much revenue that's going to drive. And
that's really what good marketing teams are trying to do all the time.
Figuring out where is the best
place to invest my money, what return am I going to get from that?
And how much can I scale
(07:06):
each channel and do I have a way to actually measure that, test it
and see it through a real life experiment?
You know, and at the end of the day, so much of what we
do is doing things well, but so much of the impact
we cause is by choosing the right things.
And that's what we do, right? We help you understand
(07:31):
what it is that you should be investing in. So when you go and put
resources behind it, you actually see it pay out. And what do
performance metrics from advertising platforms like Google or Facebook
fall often to provide accurate insights?
Yeah, if you think about it, you have. They're
the source of truth for certain things. So they can certainly tell you
(07:53):
how much you're spending with them, they can certainly tell you how many clicks
they provided, they can show you how many impressions and reach
they had, but they have no idea how much revenue they
actually drove. So what happens is we put
this pixel on your website and they see, oh, somebody
(08:14):
clicked an ad or even viewed an ad and then purchased.
Therefore we drove the purchase. But 61% of
conversions are shared by more than one platform.
And then 58% of conversions are non incremental. So
they would have happened anyway without the ad. So what that
means is that maybe you got an email for
(08:36):
20% off for something you wanted to buy, you went ahead and bought it,
and then without you realizing this, you are like
browsing like scrolling on Instagram early that day and you happen to see
an ad, but you didn't even click on it. Well, Instagram is going to take
credit for that. I think
what's important to understand is that ad platforms are in the
(08:58):
business of selling you ads, not necessarily in the business
of measuring and tracking. That has to be your responsibility
as a company to make sure that you have data
that you can trust. Is there like a somewhat easier way of
figuring that part out. Because I know that's like the big piece of the puzzle.
Because like you said, meta is going to be like, oh, we did it. Or
(09:20):
Google's going to be like, oh, we did it. And you're like, well, you helped,
but you didn't really fully drive the sale, of
course. So I'll tell you two things
people are doing that are very
unreliable, and then two things that actually work. So what doesn't work
is looking at platform data, trusting that, or
(09:44):
using what's known as multi touch attribution, which essentially is
trying to piece all this into a cohesive customer journey,
and then based on that,
arbitrarily attributing credit to different touch points. So if you saw
an ad on Facebook, then click Google and then did this and this and this,
like, okay, that's four channels. Let's split the credit
(10:07):
in quarters, 25% for each. That's even fair.
Well, I mean, who it is to say that the first touch
point had exactly as much influence as the second one, the
third one, and so on? I mean, it's kind of ridiculous.
So that's, you know, the first thing I mentioned
is not having an attribution platform. The second one is probably
(10:30):
even worse because if you have that kind of attribution that is
arbitrary, you have a false sense of confidence in data that is
actually highly inaccurate. So there's really two things
that we could do. One is called media mix modeling. So
media mix modeling essentially helps you understand what
is the most likely scenario for
(10:52):
the revenue you got given the variables provided.
So given that you spend like last week, you spent this much
on YouTube and this much on this and this much on that, and you got
those sales, what is the most likely explanation for, you
know, how much each one contributed to? So the way, the way it's done is
looking at variance in spend and how it correlates to
(11:16):
revenue. So if something is performing really well and you
double your investment well, your sales should go up.
If something has no impact, you double your investment or cut it in half
or pos it all together, it should have no impact on your
revenue. So what we
do is we look at, for example, if you sell in the United States, that's
(11:38):
50 states. So each day you have 50
observations of what happened
when spend increase or decrease
for different states and what happened to your revenue
and that. We usually have three years worth of
data. So times 50 daily observations, that's 15,000
(12:00):
observations. Very clear patterns emerge in terms of
what happened in the past as you increase your decrease
spend for different channels. The
other way that you could do this is through
incrementality experiments. So incrementality, or
it's also called like a market market test where what you
(12:22):
do is you say for, I want to
test, for example, Google Ads. So if I'm running it and spending 10k
a day, I'm going to pause it in Colorado and Michigan
and I'm going to see for the next three weeks or so what's going to
happen to sales in those states. Or I could say I'm
going to double my, my, my budget in those states
(12:44):
and see what happens. So when you do that, essentially you create
an ideal environment for an experiment because you're
saying, you know, all other things are equal. Every, everybody is getting, you
know, my, my emails, my text messages, my social media. The
only difference is that some states are either not seeing my ad or
seeing my ad more often. And
(13:08):
so one way, the first way, media mix modeling
doesn't require that you do those experiments, but when you do both
combined, that's extremely powerful. Because with media mix
modeling, you get very, very close
to the reality in the world. Then you run experiments,
use those experiments to calibrate your model so it keeps getting more and
(13:30):
more accurate. So it is
not easy. Right. And we work with companies that are spending
250,000 or more on ads per month,
some, you know, several million. So that's a, you know,
highly sophisticated approach to a really
big problem. But if
(13:53):
you are on a budget and you still want to
know how well your ads are going, a very easy way of doing
that is to actually do go ahead and
pause your ads every now and then and see what
happens or double down and you see what
happens. It's really, if you can't see
(14:16):
a clear impact or when you make such a drastic change,
that should be an indication that, you know, it's not
giving you the results that it's promising. And so how has
the death of the third party cookies impacted marketing attribution and
what adjustments do brands need to make? Yeah, so I think
that it's important to know that, you know, pre Internet
(14:39):
we already had, you know, incrementality
testing, we had media mix modeling. And then
when the Internet came out, it just allowed us to track
individual users. And that's the key difference between
the two methods that I talked about and what the Internet
promised for a while. Well, that kind
(15:02):
of worked for a little bit. It
didn't actually allow us to measure real impact, but it
allowed us to see that a visitor came from email or a
certain ad campaign and so on. Now
the problem in the last couple of years is that because of ad
blockers and privacy settings in our browsers,
(15:25):
these pixels are not working as well as they used to.
So right now 42% of conversions are being blocked.
So that means that you don't have accurate data to work with.
It also means that the platforms that use
algorithms to essentially optimize their own performance
to figure out, you know, who should, what kind of consumers
(15:47):
should we be targeting, what kind of creatives should we be showing
them they're not getting the data they need to perform their best?
So the, there's two issues with third
party cookies. You know, one is our reporting just became
incredibly unreliable. The second one is
our platforms. Ad platforms can really do
(16:10):
what they're supposed to do if we don't collect the data ourselves as
first party data and then stream them to the
platforms. If you just take the pixel they give you,
put it on your website, your performance is going to be
far from optimal. What factors lead to
inaccurate revenue reporting from advertising platforms?
(16:34):
Well, there's a couple of them that we already talked about. So
one is pixels only work 58% of the times,
so sometimes they just don't capture the true value of
the conversions they actually should be getting credit
for. Then there's duplication with
(16:55):
multiple platforms claiming the same sales.
There's the issue that only
58% of conversions are incremental. So
you know, one third of conversions would have happened anyway. I think
that that's, that's a big one. And also like the
(17:17):
pixel essentially sees somebody viewed an ad and clicked on
that and purchased it, doesn't have
any clue about everything else that is going on in your world, any other
marketing you might be doing. So yeah, it's really a
variety of all the things combined that makes,
makes them that they're still the source of truth for again, clicks, impression,
(17:40):
spend, anything that they own. But your source of truth
for revenue is your sales, right? Like you have money in the bank, you
know, how much order, how many orders you get, how much revenue you get.
And yeah, no, no one
platform can tell you how much revenue they actually
drove. And why do some businesses rely on the
(18:03):
last click attribution? And what are the dangers of doing
so? Well, the why is because it's easy, right?
The same with the pixels. There's a difference between
what's easy or even most common and what's actually
the best practice. So if you think about the
customer journey, most of the time there are
(18:26):
multiple touch points before a purchase, there
are some exceptions for, you know, some impulse purchases, but most
of the time people just, you know, learn about it, come to your
website later and sign up for your emails and then
maybe they see an ad. And eventually it usually takes about seven
touch points for somebody to purchase. So if you're giving
(18:48):
credit to the last one, you're really ignoring everything that
preceded that. And you know, you would be over
investing in that last touch point. But really
it's like the first one and first few that drove people
to the purchase eventually, you know, so you, it's, it's, it's
easy, but it's, you know, highly
(19:09):
inaccurate and it can lead to all the wrong
decisions. And how is AI being leveraged in marketing
attribution and what are its current limitations?
Yeah, so with,
I think that the, the beauty of AI is that,
I mean, you know, what do you call a data head? But most people
(19:32):
aren't, you know, and especially when it comes to
marketing, a lot of people
can read a report and see trends over time.
But going beyond and really getting
deep into, you know, what will be a data science project and looking at,
you know, probability distributions, confidence intervals and
(19:55):
how to make sense of all that, how to interpret that, what kind of decisions
to make based on that,
it just takes a little bit of training to do that. And I think that
that's what the beauty with AI. Not only does
it have access to all your data across all channels, can see the whole
picture and can provide instant answers, but it
(20:17):
could iterate with you and walk you step by step
and maybe offer suggestions that you wouldn't have
thought about otherwise. So for example, you could ask about your conversion rate,
you know, why did it drop last week? And you know, maybe just, you know,
just tell you it dropped from 4% to 3%. But it
notices that most of that drop happened on
(20:40):
mobile devices for a specific landing page. Right. So you
can, you can say that and then offer to break it down by landing page
or device. So it's
essentially giving you the ability to be a
highly trained data analyst, even if you don't have that background.
(21:00):
And I would say that you also ask about the limitations.
I think that it's such a young technology that there's like still
a little bit of hallucinations and
especially when it comes to math sometimes, you know,
surprisingly it gets really basic stuff wrong, you know, so
(21:21):
you really need to like check and see. I mean, we, with our
AI models, we train them to provide
all the intermediate steps. So you, if there's something off
in the reasoning, you know, you, you can catch it
because you're following along rather than just getting the, the final answer.
So I think it's really important to, to know that and also to
(21:44):
know that nobody could just give you
the right strategy. You know, like you still have to,
you know, you can rely on this to, to, to, to get answers, but at
the end of the day, only you know what's best for your business, what's going
to work for you. So, you know, just take everything AI gives you
with a grain of salt, challenge it. But yeah,
(22:07):
I am very excited about everything that AI
is allowing us to do. And how can businesses identify which marketing
investments are effective and which ones are draining the budget
unnecessarily? Because we're always trying to save money. Yeah. So
again, if you have the budget for a proper attribution
and testing platform,
(22:30):
definitely that's the answer right
now. If you aren't really investing a lot
in ads and need a more,
I guess, like homemade way of doing it, like a low
budget, then you can for sure do a
holdout test where you just pause the channel or double
(22:52):
down and just see how well it does. You know,
it's not going to be as accurate, but it's certainly going
to be much more informative than what the
platforms are reporting. Gotcha. And how do
you see the role of marketing attributions evolving in the next few
years? I think what's happening already is that
(23:15):
marketers are becoming more and more savvy about the limitations
of the pixels that they've been using. They understand
that they really need an
independent way of measuring success rather than just
taking what the platforms are reporting as face
value. And I think that that's essentially going
(23:37):
to help us just waste less and make more, you know.
And a side effect of that is we
won't have to bother people with ads that are not relevant to them.
So I think that what makes for good advertising
for companies also makes for a good user experience for
users on their phones or computers looking
(24:01):
at information ads that are relevant to them.
And how is the shift to privacy conscious marketing, for
example, GDPR or ccpa, involving
the future of marketing data and AI powered attribution?
So there's a lot to CPA
and gdpr, but essentially what it means is
(24:27):
to collect certain data, you have to let people know that your
GDPR is a little more,
it has more restrictions and more requirements. So for
gdpr, for example, you explicitly have to opt in before
any Cookies can be used with ccpa.
(24:48):
You see all those banners where you just accept and then just move on.
And then what you do
with that data, you have to disclose how you're going
to use it. You can't sell it, you can't
rent it.
(25:09):
So I think that
the quick answer to that is that
I don't see much of a link between AI
and what data we collect. I mean, I think it's up to us,
it's up to each company to understand
(25:32):
what do I want to be compliant with, what
markets am I in. That's going to also inform what compliance you
need, what are the local laws, what do I need to
disclose, how do I need to store the data? But I
don't see AI really messing with
that. I think that those are the two separate
(25:55):
paths and those
are more geared towards the way the data is collected and stored,
not so much how it's processed and consumed. Got you. And then
what are the key questions marketers should ask when evaluating
their current data and attribution practices?
First and foremost is what approach you use.
(26:18):
If somebody tell you that they use multi
touch attribution or mta, run away as far as can,
that's not what you need. If somebody
tells you their they're
doing media mix modeling, then the questions I would ask would
be around the model creation process and
(26:41):
validation process. Do they take into account what makes your
business unique or do they give you the same model they give
the guy before you? What's your validation
process looking like? How are you going to make sure that
the model you create is not just overfitting past data, but you can
predict a number of different scenarios, even ones that
(27:04):
it's never seen before. And anybody who's
a real data scientist should be able to answer those
questions. And yeah, then
there's kind of the vibe check. Make sure that you like the people you're working
with and you see them as a long term partner in your business.
Got you. And then how can small businesses with limited budgets and resources
(27:27):
still apply the concepts of accurate attribution without getting
overwhelmed? Yeah, so I think it depends on what we
mean by small budgets. You know, so if we are talking about like
you're spending less than $10,000 a month on ads,
you don't really need any of this. If you're spending less than
(27:47):
50,000, maybe do one of these like
holdout tests, like homemade tests that we
talked about where you just turn off a channel or double it down
and then see what happens. But if you're spending 50,000 or
more. Yeah, like you. You need something a
little more sophisticated than that. Got you. And then
(28:10):
people listening to this episode, they're wondering where can they find you online to learn
more? You can learn more at data speaks AI or you
can find me on LinkedIn. I'm sure we can put the link in the show
notes. Yeah. Feel free to reach out if you have any
questions. All right, any final thoughts for listeners?
I'm going to go have some tea because I realized that I'm having too much
(28:31):
coffee. So I'm going to balance that out. That's fair. That's
fair. But thank you, Zeke, for joining Digital Coffee Marketing Brew and sharing your
knowledge on AI and data. Thank you, Brad,
and thank you for listening as always. Please subscribe to this podcast and all your
favorite podcasting apps. Leave a five star review. Really just help the rankings
and let me know how I am doing in of terms. Join me next week.
(28:54):
Let's talk about what's going on in the PR marketing industry. All right,
guys, stay safe. Get to understanding your data and your marketing attributions and
figuring out where you're actually getting all the sales and see you next week
later.
Five out of six marketers believes it's absolutely essential
to make data driven decisions to get the best results
from your advertising. Only one in six actually
trusts the data that you use to make
decisions. And at the end of the day, so much of what
we do is doing things well, but so much of the
(00:22):
impact we cause is by choosing the right things.
But 61% of conversions are shared by more than one
platform and then 58% of conversions are non
incremental. So they would have happened anyway without the ad.
(00:49):
And welcome to a new
episode of Digital Coffee Marketing Brew. And I'm your host,
Brett Deister. If you please subscribe to this podcast and all your favorite podcasting apps,
leave a five star review really does help with the rankings and let me
know how I am doing. But this week
I have Zeke with me and he is a
(01:11):
serial entrepreneur, the founder of Data Speaks, an AI
powered analytics platform that helps companies
identify what drives her sales and invest in the right
marketing strategies. And he has a background in economics
and data science. Zeke has spent the last 20 years developing AI machine
learning and data analytics solution, embedding hundreds of companies
(01:33):
to make data driven decisions and accelerate growth.
So welcome to the show, Zeke. Thank you, Brad. Thank you for having me. All
right, the first question is all my guest is, are you a coffee or tea
drinker? I'm both. I have coffee in the morning
and then usually tea before going to, going to bed.
(01:53):
But I would say that I drink coffee way more than a drink
tea. Do you have like a specific favorite of coffee or is it just
whatever you can get your hands on? I love
Italian coffee, espresso roasts.
I like usually
medium roast, South American
(02:17):
coffee brewed in Italy if it makes sense. You
know, so there's a couple of brands, Lavazza that
I, you know, really, really like is smooth and, and
just strong but not too strong. Yeah. So you specifically like
espresso? Because when I hear Italian coffee, it's basically always espresso.
Yeah, yeah, it, it's. I,
(02:40):
you know, was born and raised in Argentina. We have a lot of, you know,
Italian influence and it's just stronger, you know. So if you go to
Starb, you know, any kind of coffee
shop, it usually feels like watered down to anybody
who's from either Italy or Argentina. So I like
my coffee a little stronger. So you more like the double shot or
(03:02):
like when it goes into your. It's more of a double shot than a single
shot because I think Starbucks does a single shot and you
usually like the double shot or more? I would say
yeah, and. Yeah, and
a little darker rose than most people have it.
Fair enough. But I gave a brief summary of your expertise. Can you give our
(03:23):
listeners a little bit more about what you do? Sure. So
historically, I would say I'm a serial entrepreneur. I've
had many businesses in the tech space.
Right now I'm the CEO and founder of Data Speaks. You
kind of touched on that. But we help marketing teams
(03:44):
understand what channels and campaigns drive their sales
so they can invest in the right marketing strategies.
Gotcha. Can you explain what marketing attribution is and why it's such
a crucial, yet challenging area for marketers? Yeah, of course.
So what happens is when you are
advertising a lot, you know, you have a presence on multiple channels,
(04:06):
you're running video ads on YouTube, you are
on search, Google, Bing, you're doing
social, Snapchat, Meta, Instagram,
Pinterest, and so on. So you're doing all these different
things. And then at the end of the day, your store sells,
say, you know, $200,000 a week of
(04:28):
widgets. And what you want to know is you want to understand
how much of that was influenced by each channel
and campaign, because otherwise it's,
you know, money going out, money coming in, but you're not really
understanding the ROI of each of these investments. So
what we found is that in the US 54%
(04:51):
of marketing budgets are spent on ads. So more than
half. And while five out of six
marketers believes it's absolutely essential to
make data driven decisions to get the best results
from your advertising, only one in six actually
trusts their the data that you use to make
(05:12):
decisions. And that's because they put all these pixels on the
website for Google Ads, Facebook ads, and
they make decisions based on those. And those can potentially
be a source of truth because each of those only
sees activity from that platform. So if the Facebook
ads, pixel sees activity from Facebook, but
(05:35):
it doesn't really know that you're also running ads on Google, that you
are sending emails, text messages. So you
really need an independent way to track all the activity and
how that then impacts your bottom line.
So what you're basically saying is it's fragmented data that you need to
centralize so you can understand exactly where everything is
(05:58):
going. It's kind of like podcasting, because podcasting
there is a lot of different third party data, but it's not all
centralized either. So I have to look at three or four different ones
just to figure out, like, what's going on. And the thing is, like Even,
even if you were so right in that it's
really a puzzle and each of those pieces helps you see the full
(06:20):
picture. But none of those can
help you figure out the actual revenue it drove. Because you
need to understand, essentially, it's called incrementality.
You want to understand the incremental impact of your ads. So if you go
and increase your spend
by say $10,000 a month for
(06:43):
YouTube, you start running more YouTube ads. You want to understand
how much revenue that's going to drive. And
that's really what good marketing teams are trying to do all the time.
Figuring out where is the best
place to invest my money, what return am I going to get from that?
And how much can I scale
(07:06):
each channel and do I have a way to actually measure that, test it
and see it through a real life experiment?
You know, and at the end of the day, so much of what we
do is doing things well, but so much of the impact
we cause is by choosing the right things.
And that's what we do, right? We help you understand
(07:31):
what it is that you should be investing in. So when you go and put
resources behind it, you actually see it pay out. And what do
performance metrics from advertising platforms like Google or Facebook
fall often to provide accurate insights?
Yeah, if you think about it, you have. They're
the source of truth for certain things. So they can certainly tell you
(07:53):
how much you're spending with them, they can certainly tell you how many clicks
they provided, they can show you how many impressions and reach
they had, but they have no idea how much revenue they
actually drove. So what happens is we put
this pixel on your website and they see, oh, somebody
(08:14):
clicked an ad or even viewed an ad and then purchased.
Therefore we drove the purchase. But 61% of
conversions are shared by more than one platform.
And then 58% of conversions are non incremental. So
they would have happened anyway without the ad. So what that
means is that maybe you got an email for
(08:36):
20% off for something you wanted to buy, you went ahead and bought it,
and then without you realizing this, you are like
browsing like scrolling on Instagram early that day and you happen to see
an ad, but you didn't even click on it. Well, Instagram is going to take
credit for that. I think
what's important to understand is that ad platforms are in the
(08:58):
business of selling you ads, not necessarily in the business
of measuring and tracking. That has to be your responsibility
as a company to make sure that you have data
that you can trust. Is there like a somewhat easier way of
figuring that part out. Because I know that's like the big piece of the puzzle.
Because like you said, meta is going to be like, oh, we did it. Or
(09:20):
Google's going to be like, oh, we did it. And you're like, well, you helped,
but you didn't really fully drive the sale, of
course. So I'll tell you two things
people are doing that are very
unreliable, and then two things that actually work. So what doesn't work
is looking at platform data, trusting that, or
(09:44):
using what's known as multi touch attribution, which essentially is
trying to piece all this into a cohesive customer journey,
and then based on that,
arbitrarily attributing credit to different touch points. So if you saw
an ad on Facebook, then click Google and then did this and this and this,
like, okay, that's four channels. Let's split the credit
(10:07):
in quarters, 25% for each. That's even fair.
Well, I mean, who it is to say that the first touch
point had exactly as much influence as the second one, the
third one, and so on? I mean, it's kind of ridiculous.
So that's, you know, the first thing I mentioned
is not having an attribution platform. The second one is probably
(10:30):
even worse because if you have that kind of attribution that is
arbitrary, you have a false sense of confidence in data that is
actually highly inaccurate. So there's really two things
that we could do. One is called media mix modeling. So
media mix modeling essentially helps you understand what
is the most likely scenario for
(10:52):
the revenue you got given the variables provided.
So given that you spend like last week, you spent this much
on YouTube and this much on this and this much on that, and you got
those sales, what is the most likely explanation for, you
know, how much each one contributed to? So the way, the way it's done is
looking at variance in spend and how it correlates to
(11:16):
revenue. So if something is performing really well and you
double your investment well, your sales should go up.
If something has no impact, you double your investment or cut it in half
or pos it all together, it should have no impact on your
revenue. So what we
do is we look at, for example, if you sell in the United States, that's
(11:38):
50 states. So each day you have 50
observations of what happened
when spend increase or decrease
for different states and what happened to your revenue
and that. We usually have three years worth of
data. So times 50 daily observations, that's 15,000
(12:00):
observations. Very clear patterns emerge in terms of
what happened in the past as you increase your decrease
spend for different channels. The
other way that you could do this is through
incrementality experiments. So incrementality, or
it's also called like a market market test where what you
(12:22):
do is you say for, I want to
test, for example, Google Ads. So if I'm running it and spending 10k
a day, I'm going to pause it in Colorado and Michigan
and I'm going to see for the next three weeks or so what's going to
happen to sales in those states. Or I could say I'm
going to double my, my, my budget in those states
(12:44):
and see what happens. So when you do that, essentially you create
an ideal environment for an experiment because you're
saying, you know, all other things are equal. Every, everybody is getting, you
know, my, my emails, my text messages, my social media. The
only difference is that some states are either not seeing my ad or
seeing my ad more often. And
(13:08):
so one way, the first way, media mix modeling
doesn't require that you do those experiments, but when you do both
combined, that's extremely powerful. Because with media mix
modeling, you get very, very close
to the reality in the world. Then you run experiments,
use those experiments to calibrate your model so it keeps getting more and
(13:30):
more accurate. So it is
not easy. Right. And we work with companies that are spending
250,000 or more on ads per month,
some, you know, several million. So that's a, you know,
highly sophisticated approach to a really
big problem. But if
(13:53):
you are on a budget and you still want to
know how well your ads are going, a very easy way of doing
that is to actually do go ahead and
pause your ads every now and then and see what
happens or double down and you see what
happens. It's really, if you can't see
(14:16):
a clear impact or when you make such a drastic change,
that should be an indication that, you know, it's not
giving you the results that it's promising. And so how has
the death of the third party cookies impacted marketing attribution and
what adjustments do brands need to make? Yeah, so I think
that it's important to know that, you know, pre Internet
(14:39):
we already had, you know, incrementality
testing, we had media mix modeling. And then
when the Internet came out, it just allowed us to track
individual users. And that's the key difference between
the two methods that I talked about and what the Internet
promised for a while. Well, that kind
(15:02):
of worked for a little bit. It
didn't actually allow us to measure real impact, but it
allowed us to see that a visitor came from email or a
certain ad campaign and so on. Now
the problem in the last couple of years is that because of ad
blockers and privacy settings in our browsers,
(15:25):
these pixels are not working as well as they used to.
So right now 42% of conversions are being blocked.
So that means that you don't have accurate data to work with.
It also means that the platforms that use
algorithms to essentially optimize their own performance
to figure out, you know, who should, what kind of consumers
(15:47):
should we be targeting, what kind of creatives should we be showing
them they're not getting the data they need to perform their best?
So the, there's two issues with third
party cookies. You know, one is our reporting just became
incredibly unreliable. The second one is
our platforms. Ad platforms can really do
(16:10):
what they're supposed to do if we don't collect the data ourselves as
first party data and then stream them to the
platforms. If you just take the pixel they give you,
put it on your website, your performance is going to be
far from optimal. What factors lead to
inaccurate revenue reporting from advertising platforms?
(16:34):
Well, there's a couple of them that we already talked about. So
one is pixels only work 58% of the times,
so sometimes they just don't capture the true value of
the conversions they actually should be getting credit
for. Then there's duplication with
(16:55):
multiple platforms claiming the same sales.
There's the issue that only
58% of conversions are incremental. So
you know, one third of conversions would have happened anyway. I think
that that's, that's a big one. And also like the
(17:17):
pixel essentially sees somebody viewed an ad and clicked on
that and purchased it, doesn't have
any clue about everything else that is going on in your world, any other
marketing you might be doing. So yeah, it's really a
variety of all the things combined that makes,
makes them that they're still the source of truth for again, clicks, impression,
(17:40):
spend, anything that they own. But your source of truth
for revenue is your sales, right? Like you have money in the bank, you
know, how much order, how many orders you get, how much revenue you get.
And yeah, no, no one
platform can tell you how much revenue they actually
drove. And why do some businesses rely on the
(18:03):
last click attribution? And what are the dangers of doing
so? Well, the why is because it's easy, right?
The same with the pixels. There's a difference between
what's easy or even most common and what's actually
the best practice. So if you think about the
customer journey, most of the time there are
(18:26):
multiple touch points before a purchase, there
are some exceptions for, you know, some impulse purchases, but most
of the time people just, you know, learn about it, come to your
website later and sign up for your emails and then
maybe they see an ad. And eventually it usually takes about seven
touch points for somebody to purchase. So if you're giving
(18:48):
credit to the last one, you're really ignoring everything that
preceded that. And you know, you would be over
investing in that last touch point. But really
it's like the first one and first few that drove people
to the purchase eventually, you know, so you, it's, it's, it's
easy, but it's, you know, highly
(19:09):
inaccurate and it can lead to all the wrong
decisions. And how is AI being leveraged in marketing
attribution and what are its current limitations?
Yeah, so with,
I think that the, the beauty of AI is that,
I mean, you know, what do you call a data head? But most people
(19:32):
aren't, you know, and especially when it comes to
marketing, a lot of people
can read a report and see trends over time.
But going beyond and really getting
deep into, you know, what will be a data science project and looking at,
you know, probability distributions, confidence intervals and
(19:55):
how to make sense of all that, how to interpret that, what kind of decisions
to make based on that,
it just takes a little bit of training to do that. And I think that
that's what the beauty with AI. Not only does
it have access to all your data across all channels, can see the whole
picture and can provide instant answers, but it
(20:17):
could iterate with you and walk you step by step
and maybe offer suggestions that you wouldn't have
thought about otherwise. So for example, you could ask about your conversion rate,
you know, why did it drop last week? And you know, maybe just, you know,
just tell you it dropped from 4% to 3%. But it
notices that most of that drop happened on
(20:40):
mobile devices for a specific landing page. Right. So you
can, you can say that and then offer to break it down by landing page
or device. So it's
essentially giving you the ability to be a
highly trained data analyst, even if you don't have that background.
(21:00):
And I would say that you also ask about the limitations.
I think that it's such a young technology that there's like still
a little bit of hallucinations and
especially when it comes to math sometimes, you know,
surprisingly it gets really basic stuff wrong, you know, so
(21:21):
you really need to like check and see. I mean, we, with our
AI models, we train them to provide
all the intermediate steps. So you, if there's something off
in the reasoning, you know, you, you can catch it
because you're following along rather than just getting the, the final answer.
So I think it's really important to, to know that and also to
(21:44):
know that nobody could just give you
the right strategy. You know, like you still have to,
you know, you can rely on this to, to, to, to get answers, but at
the end of the day, only you know what's best for your business, what's going
to work for you. So, you know, just take everything AI gives you
with a grain of salt, challenge it. But yeah,
(22:07):
I am very excited about everything that AI
is allowing us to do. And how can businesses identify which marketing
investments are effective and which ones are draining the budget
unnecessarily? Because we're always trying to save money. Yeah. So
again, if you have the budget for a proper attribution
and testing platform,
(22:30):
definitely that's the answer right
now. If you aren't really investing a lot
in ads and need a more,
I guess, like homemade way of doing it, like a low
budget, then you can for sure do a
holdout test where you just pause the channel or double
(22:52):
down and just see how well it does. You know,
it's not going to be as accurate, but it's certainly going
to be much more informative than what the
platforms are reporting. Gotcha. And how do
you see the role of marketing attributions evolving in the next few
years? I think what's happening already is that
(23:15):
marketers are becoming more and more savvy about the limitations
of the pixels that they've been using. They understand
that they really need an
independent way of measuring success rather than just
taking what the platforms are reporting as face
value. And I think that that's essentially going
(23:37):
to help us just waste less and make more, you know.
And a side effect of that is we
won't have to bother people with ads that are not relevant to them.
So I think that what makes for good advertising
for companies also makes for a good user experience for
users on their phones or computers looking
(24:01):
at information ads that are relevant to them.
And how is the shift to privacy conscious marketing, for
example, GDPR or ccpa, involving
the future of marketing data and AI powered attribution?
So there's a lot to CPA
and gdpr, but essentially what it means is
(24:27):
to collect certain data, you have to let people know that your
GDPR is a little more,
it has more restrictions and more requirements. So for
gdpr, for example, you explicitly have to opt in before
any Cookies can be used with ccpa.
(24:48):
You see all those banners where you just accept and then just move on.
And then what you do
with that data, you have to disclose how you're going
to use it. You can't sell it, you can't
rent it.
(25:09):
So I think that
the quick answer to that is that
I don't see much of a link between AI
and what data we collect. I mean, I think it's up to us,
it's up to each company to understand
(25:32):
what do I want to be compliant with, what
markets am I in. That's going to also inform what compliance you
need, what are the local laws, what do I need to
disclose, how do I need to store the data? But I
don't see AI really messing with
that. I think that those are the two separate
(25:55):
paths and those
are more geared towards the way the data is collected and stored,
not so much how it's processed and consumed. Got you. And then
what are the key questions marketers should ask when evaluating
their current data and attribution practices?
First and foremost is what approach you use.
(26:18):
If somebody tell you that they use multi
touch attribution or mta, run away as far as can,
that's not what you need. If somebody
tells you their they're
doing media mix modeling, then the questions I would ask would
be around the model creation process and
(26:41):
validation process. Do they take into account what makes your
business unique or do they give you the same model they give
the guy before you? What's your validation
process looking like? How are you going to make sure that
the model you create is not just overfitting past data, but you can
predict a number of different scenarios, even ones that
(27:04):
it's never seen before. And anybody who's
a real data scientist should be able to answer those
questions. And yeah, then
there's kind of the vibe check. Make sure that you like the people you're working
with and you see them as a long term partner in your business.
Got you. And then how can small businesses with limited budgets and resources
(27:27):
still apply the concepts of accurate attribution without getting
overwhelmed? Yeah, so I think it depends on what we
mean by small budgets. You know, so if we are talking about like
you're spending less than $10,000 a month on ads,
you don't really need any of this. If you're spending less than
(27:47):
50,000, maybe do one of these like
holdout tests, like homemade tests that we
talked about where you just turn off a channel or double it down
and then see what happens. But if you're spending 50,000 or
more. Yeah, like you. You need something a
little more sophisticated than that. Got you. And then
(28:10):
people listening to this episode, they're wondering where can they find you online to learn
more? You can learn more at data speaks AI or you
can find me on LinkedIn. I'm sure we can put the link in the show
notes. Yeah. Feel free to reach out if you have any
questions. All right, any final thoughts for listeners?
I'm going to go have some tea because I realized that I'm having too much
(28:31):
coffee. So I'm going to balance that out. That's fair. That's
fair. But thank you, Zeke, for joining Digital Coffee Marketing Brew and sharing your
knowledge on AI and data. Thank you, Brad,
and thank you for listening as always. Please subscribe to this podcast and all your
favorite podcasting apps. Leave a five star review. Really just help the rankings
and let me know how I am doing in of terms. Join me next week.
(28:54):
Let's talk about what's going on in the PR marketing industry. All right,
guys, stay safe. Get to understanding your data and your marketing attributions and
figuring out where you're actually getting all the sales and see you next week
later.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
My Favorite Murder with Karen Kilgariff and Georgia Hardstark
My Favorite Murder is a true crime comedy podcast hosted by Karen Kilgariff and Georgia Hardstark. Each week, Karen and Georgia share compelling true crimes and hometown stories from friends and listeners. Since MFM launched in January of 2016, Karen and Georgia have shared their lifelong interest in true crime and have covered stories of infamous serial killers like the Night Stalker, mysterious cold cases, captivating cults, incredible survivor stories and important events from history like the Tulsa race massacre of 1921. My Favorite Murder is part of the Exactly Right podcast network that provides a platform for bold, creative voices to bring to life provocative, entertaining and relatable stories for audiences everywhere. The Exactly Right roster of podcasts covers a variety of topics including historic true crime, comedic interviews and news, science, pop culture and more. Podcasts on the network include Buried Bones with Kate Winkler Dawson and Paul Holes, That's Messed Up: An SVU Podcast, This Podcast Will Kill You, Bananas and more.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.