What Is Marketing Mix Modeling In Ecommerce And How To Use It — Michael True | Why More Ad Channels Make Attribution Messy, Why Top-of-funnel Is Hard To Measure, How To Link Offline Sales To Digital Ads, How AI-driven MMM Give Daily Insights (#424)

In this episode, Mike True, Co-Founder and CEO of Prescient AI, shares how brands can cut through messy multi-channel ad data with marketing mix modeling. 

He explains how to see the true impact of top-funnel channels like YouTube, TV, and influencers, track offline sales, and optimize budgets daily.

Mike also reveals how AI-driven insights help brands scale confidently across DTC, marketplaces, and retail.

Topics discussed in this episode:  

• How marketing mix modeling works. 
• Why adding more ad channels makes attribution messy. 
• How the ‘last click’ skews your data. 
• Why top-of-funnel is hard to measure. 
• What MMM shows that others can’t. 
• How AI-driven MMM gives daily insights. 
• Why the real ‘source of truth’ is human. 
• How to link offline sales to digital ads. 
• Why omnichannel brands gain most from MMM. 
• What the MMM ‘aha moment’ looks like. 
• Why one metric won’t scale your business. 

Links & Resources 

Website: https://prescientai.com/
LinkedIn: https://www.linkedin.com/in/michaeljtrue/
LinkedIn: https://www.linkedin.com/company/prescient-ai/

Get access to more free resources by visiting the show notes at
https://tinyurl.com/y2v7t7fx

MORE RESOURCES

Subscribe to our FREE Newsletter: https://newsletter.ecommercecoffeebreak.com/

Free Store Optimization Beginners Guide: Instant PDF Download! 👉 https://ecommercecoffeebreak.com/essential-conversion-optimization-guide/

Rate & Review: Help others discover the show by rating the show on Apple Podcasts at https://podcasts.apple.com/us/podcast/ecommerce-coffee-break-digital-marketing-podcast-for/id1567749422

Advertise on ECB - https://ecommercecoffeebreak.com/podcast-sponsorship/

Follow the podcast to catch all the bonus episodes I am adding. Hit that follow button now!

Transcript

Speaker 1 (00:00.256)
The last click was the attribution robber. It's like driving down the highway and opening up the sunroofs and just letting the money fly out the window. We believe that the billboard drove this much sales. We believe that the newspaper drove this much sales. There is no form of measurement that is a source of truth, right? The source of truth is the human. And then when it works, I think it says, wow, like the math really started a math.

Hello and welcome to another episode of the Ecommerce Coffee Break podcast. Running ads on Google and Meta is simple until it's not. Once you add TikTok, YouTube, Amazon, influencers into the mix, things get messy fast. Your data stops lining up, roles gets blurry, and suddenly it's hard to know what actually drives your sales. So how to optimize ongoing ad performance when your marketing mix grows. That's what we're going to talk about today. And joining me for this is Mike True. He's the co-founder and CEO of Prescient AI.

And before launching Prescient in 2019, helped clients at any IBM and Oracle generating millions using AI and analytics. Mike knows how to turn messy data into clear answers. We have a lot to cover. Let's get started. Hi, Mike.

I'm doing well. Thanks a for me. How are you doing?

very well. Mike, most brands rely on Google and meta-ads and then things suddenly become messy when it comes to attribution and really to assign where the sales coming from. What's the problem there?

Speaker 1 (01:22.498)
When brands are launching on Meta and Google, I think it's a little bit easier. They're more click-based channels where you can set up pixels and leverage some of the leading MTA tools in the space like Northaim and Triple Whale and Rockerbox for MTA. And then as you start to go scale onto new channels, more top-of-funnel things like YouTube and TV podcasts, influencers, that are more view-based channels.

it's really hard to understand which one of those are actually driving performance. Because, you know, what somebody said to me the other day, like, you know, the last click was the attribution robber, right? the slutty. But it is actually, you know, I was looking at a client yesterday and they had a high, they have a pretty high AOV, probably about a thousand dollar price point, sending across nine channels. So a very diverse media mix.

selling on DTC, selling on Amazon, and then also selling offline and retail. Their four search campaigns, the ROAS, ranged from 298 to 175X. So looking at that, it's kind of a paralysis, if you're just basing it with the platforms they're reporting, because you know that people are going to click and add for that price point. I mean, some people might, but...

the majority of people are going to take some time, they're going to go searching. That's what makes it challenging.

I think what you mentioned there, attribution rubber ring, I think that's a very good word because it's sort of known that every platform, MetaGoogle, they try to assign the attribution to themselves because they won't have people spending more money on that specific platform. And that's where things really get messy. I have this in my own experience that you really can't say is like, I'm not sure that doesn't look right for me. But that's the number that you get. This is that if you add more channels, things become really difficult.

Speaker 2 (03:25.89)
What's the price of just getting onto more channels and losing control?

What's the like the the price of getting on those channels? I mean, you're it's not like a state. It's like driving driving down the highway and opening up the sunroof and just letting the money fly out the window. You can't measure what you can't manage. I can't manage what you can't measure. It's one of those things where if you're going to go. Put a top of funnel strategy in place, it's it's very important to have the right, you know.

workflows and infrastructure and measurement types to triangulate that measurement alongside other forms of measurement. And it's very important. But yeah, I we see this often where clients will come to us and they'll have spent six months on scaling the top of funnel and didn't have the proper form of measurement. And when they onboard, we can go back historically and they're like, wow, like intuitively, I thought this was actually, you know, at its peak.

And there was either more room to scale or, you know, they were oversaturated and were spending too much money. it's largely missed opportunities.

Now, running a business by gut feeling when it comes to marketing spend is probably not a good idea. So you're running with something that's called marketing mix modeling, MMM. A lot of our listeners might not have heard about it. Tell me what it is.

Speaker 1 (04:54.476)
And of course, some MMMs are probabilistic. They're statistical models. They've been around since the early 60s where they had people that had a billboard. They maybe did TV, newspapers, catalogs, and people walked into a store and they purchased a Gatorade. Although you can't really track any clicks there, so they would try to put as much revenue data as they could find from the store and as much data as you could find about these channels. And they would run a model and say, we think...

we believe that the billboard drove this much sales. We believe that the newspaper drove this much sales. And essentially it was a form of attribution. the other part of an MMM is going to tell you how much essentially will predict future performance if you make changes to your budgets. So if you increase spend on Meta, if you decrease spend on Google or you scale spend on TikTok, what is that sort of predicted outcome that you could expect?

So yeah, I mean, I traditionally most people might think of an M &M is like a Nielsen, right? Where they run this one is big, you know, EDF report once a year, there's a bunch of scientists that come together and they will give you recommendations on how you should reallocate your medium exits.

I think a lot of DTC brands found as marketeers are not really data scientists and you're coming from the side of AI, of analytics. With all the data points that we have to handle nowadays, how does your system help them to really get a real point of view on what's happening in their marketing?

Yeah, it's that, I you position that perfectly is it is a perspective and it's a point of view. I've said this since I've joined the e-commerce industry was there is no form of measurement that is a source of truth, right? Do you have different measurements that are designed to, you know, be value based off of, you know, different media mixes and different goals, right? The source of truth is the human, right? And so it's a variety of ways to slice it and dice it. So for us as

Speaker 1 (06:58.542)
is pressing day I when we started researching our models in 2020, it was really under that theme of, well, how can we give brands and organizations access to state of the art machine learning models without having to build infrastructure, hire engineers, hire data scientists, build the model, train the model, and then maintain the models. And so we built a platform where we wanted to onboard a brand in under 20 minutes and just all that has

It was funny, used to, like our first few clients, they would ask them to time it with a stopwatch and they would send this. It was very effective for our next fundraising round to show the seed of it. But yeah, we can ingest all that historical data and the models will train for each brand specific to that brand. So it's not sharing data. And there's no humans from our side really involved. All of the models are able to learn using.

the brand's historical data. So things in the model you look at, we have to look at all of like their sales data from Shopify, for Amazon and for retail. So we want to think of things like revenue by day, your customers by day, orders by day. And then we'll look at obviously their website data. So a big one for us is looking at session data from GA, and then the associated conversions, whether it was paid organic and direct. And then we'll want to look at the data from the ad platforms. And so

We'll look, obviously, their channel, the campaign, spend impressions, click sessions, and then whatever reported revenue is coming from the three ad platforms. And then we try to figure out things like what is the seasonality of the business, right? What is that like, what is just like the trend of their, the word of mouth, like the brand of the business, the brand equity. And then the last thing you want to figure out is, you know, what is sort of the buying cycle of that product, consideration, cycling, content, committers, and so on.

$40 price point or a $3,000 price point. We bring all of those signals together, which normally you'd have a bunch of research scientists and data scientists come together to analyze those. We've made it in the self-service way that's fast, that's quick, and it runs every day.

Speaker 2 (09:10.584)
Can you give me an example how a brand actually works on a day-to-day basis with your system? So who's logging in? What are they checking? And what do they basically read from the data?

Yeah. So we, we have a lot of performance teams for our teams and then a lot of the agencies as well. And I guess I could give you an example of a brand that is selling on Shopify and selling on Amazon as well. And it's a CPG brand that heavily relies on subscription. So LTV to CAC. They'll be able to log into our platform and essentially look at what we believe that CAC is, new customers is, or ROAS is, and then compare that to what the ad platforms are saying.

Typically with channels like YouTube or TV or podcasts, we're showing more favorable results in the platforms. And so they're getting more confidence in how this top of funnel channel is really performing. And then what they'll use is the other portion of our platform. There's we call those halo effects, by the way. So what is the halo effect of, how is YouTube or app love and impacting sales on their Shopify store? But also what is the halo effect onto their Amazon store?

So that's the first part of the model is a measurement. And then what else they'll do is well, how much can I scale this campaign? So we will show them, you know, kind of the saturation curves, if you will. You know, if you spend $2 and make $10 on a campaign, it's pretty likely you can't spend $2 million and make $10 million on that same campaign. Eventually, you're to start to hit points of diminishing returns. What's unique to our product is we go down to an individual campaign level. And so they could look at a YouTube campaign and plot out

What is the appropriate spend for new customers on Amazon? And then they'll be able to go scale that spend. I would say a unique part of that is traditionally an MMM would run every 30 days and it's kind of have to wait. Our model will run every single day. We've figured out a way to do that at a computational cost. And so you're going to start getting results back on these lines. The lines will actually start to move as the data comes in each day. So they'll use it for more of the optimization.

Speaker 1 (11:22.606)
in real time.

I like the thinking and I like the way you do that. You basically what you just said is you can look at your YouTube channel and you know what's happening on Amazon. That's what traditional analytics cannot really tell you. There's a huge difference there. Now, before we started recording, you told me also offline conversions can be recorded. How does that work?

Yeah, this actually goes live next week. We've been building this for the last year with a couple of our brands. But you could think of there's a halo effect for people seeing digital campaigns and then purchasing in a retail location. One of our beauty brands that we've done some early testing with was Influencers was not looking good in their platforms for measurement on their website. And then our platform really wasn't looking too good on Amazon.

But in Alta Beauty, one of their best performing channels. And so it was a really big unlock there. You'll have folks that have owned and operated, right? So think of it with their own brick and mortar stores. But then you're gonna have brands, know, to why everybody over at Mary Ruth's, brands that are like crushing, you can go into Walmart and Target and all these different wholesale channels.

And then being able to measure the impact of what's happening in those offline sales. We have to factor in other inputs, if you will, we call them covariates, but things like, is there any in-store promotions going on or any contextual things? it's, you know, retail's a little, offline's a little more ambiguous than your digital, if that makes sense. You you see things, click, there's a lot of data associated to that. But, you know, while it's not perfect, why do these models get smarter?

Speaker 1 (13:07.266)
hopefully as a directionally accurate kind of guiding principle.

Tell me about who's your perfect customer, for whom would that work best?

Who would not work best as brands that are pretty much one or two advertising channels. We will typically recommend one of the MTAs for that. As a brand starts to go into territories of more view-based channels, brands that are omnichannel, so D2C, marketplaces, and then going obviously offline is really where these models will thrive. So the more complex, the better.

I was chatting to a friend of a friend yesterday, he has a brand and he's going on to his third channel. He's got a meta Google, he's going on to TikTok and he's like, and we're just getting ready to launch on Amazon. So this is one of the cases with their artificial intelligence is these systems get smarter, right? The more they learn. And so I did recommend, you see you can onboard because

He's in hyper growth mode and he has a plan to go into TV and all these different channels. so the faster you can start training these models, the smarter that they're going to get. So complex media nexus and people that have plans that's growing business, they know they're going to scale more and start training machine learning models.

Speaker 2 (14:19.822)
Mm-hmm.

Speaker 2 (14:26.882)
The question of that you mentioned before that you basically go through the history in the platform and use that data to optimize the MMM. I think that's a huge difference to how things worked in the past where it basically was in a moment when you switch it on and then it would start collecting. Is there any kind of homework that the merchant needs to do before they can get started?

I would, if it's a first time of looking at an MMM, all the time we're like, you know, my boss heard you guys on a podcast and I just YouTube a video before this call, I'd never heard of an MMM before. I think understanding the difference between deterministic attribution, which is more click-based multi-touch and then probabilistic attribution.

The number one thing is how do you operationalize this into your workflows? A lot of these brain steps and incredible growth on these two channels and they're very dialed in with the deterministic approach. And all right, well, and they have, you know, they have their whole kind of workflows with their agencies, their audience seems to creative. Everything's really dialed in. How do you take an MMM and operationalize that alongside?

your existing tools. Every single one of our clients that we have is running an MTA and runs us alongside them as well.

So does that mean for listeners that you will never have to look into Google Analytics and into your Google Ads platform again? Or how does that work on a day-to-day reporting basis?

Speaker 1 (15:54.254)
Listen, it's triangulation. You have to look at what your MTA shows. You have to look at what GA shows, the platform shows what we're showing. And try not to get analysis paralysis for trying to find some signals through the noise of feeling confident to go make changes to your spend. But I'll say this, I'll die on those heels. You cannot have a form of measurement as a source of truth, as a single form of measurement.

So at the end of the day, it's the human who makes the decision to solve.

Always going to be the... Yeah, I mean, even with like the agentic AI and the agents and automation, some things that's going to happen, I believe that you always, you should always have a human in the loop, right? To make these decisions. Nobody knows this stuff. You can't even care what model and how good it's changing. I build a machine learning company, but nobody knows better than machine learning. They really don't. You know, you can have a really good, you know, mathematician mind that is solely focused on

you helping you when and getting you recommendations, but it's up to the human to make those decisions always. My opinion.

Good to hear. I would agree with you. think a lot of our listeners will have the same feeling that I want to give the steering wheel out of hand and have AI or any kind of system running the business automatically. And if you have really like the power to make the decision by yourself at the end of the day, for me, it would be a good feeling, a better feeling put it that way. So tell me about your pricing structure. How does that work?

Speaker 1 (17:22.99)
pricing structures per model. So think about that per sales channel. You have a DTC model and you'd have a marketplace model like Amazon. And then you have a retail model. So a model for each sales channel. And it's based off of your last trailing 12 months of revenue per sales channel. So if you're between 10 to 25 million, 25 to 50. And my thesis was

Some people are like, why don't you just charge a percentage of ad spent? That's kind of how, align with what an agency is offering, but I don't feel that the incentives are aligned there. You know, our platform is going to tell you how to go make more money. And if we tell you how to go make more money, then let's make more money together. So when they hit the next tier, it'll incrementally increase from there.

Okay, making more money is always good. When a brand is joining you, what did you experience? When is the aha moment coming for your clients to see that really what it does for them?

When they want, I think, the time to value, they have a lot of brains have this perception that these MMMs are very long time to onboard. All right. Well, they'll be like, wow, this was really fast. That's one of the things. And then when they go into the platform and they go back and they look at these top of funnel channels and they look at the performance, they're like, wow, right.

lot of times it matters their intuitions. A lot of times they're surprised and they want to see surprised. They want to see something that's different and they want to feel confident in that difference. I think the main wow factor is usually about week three or four, we'll start running optimization scenarios. Right? So essentially it's going to go, the model is going to go in there and scan all of their channels, their campaigns, their seasonality, the ratio between the top of funnel and bottom of funnel, upcoming seasonal moments.

Speaker 1 (19:19.116)
and it's going to tell them how to reallocate their budget, at least in a few days, what we believe that incremental impact is going to be. And then when it works, I think it says, wow, like, you know, the math really started to math.

I think that's definitely a highlight for every marketer or brand owner to see that there is so much out there that they didn't know about it and now they can optimize. Before we come to the end of our coffee break today, is there anything you want to share with our listeners that we haven't covered yet?

No, think, think it's, yeah, there's one thing I just say, a couple things, but I think it's important to understand the different forms of measurement where, know, MMM again is going to be a probabilistic model. It's going to give you topofunnel measurement and it's going to tell you what you should go do next. A multi-touch attribution is going to give you very deterministic, what is the click-based journey of your, of your clients or your customers. And then you have

know, surveys, right? So post purchase surveys where you can get forms of measurement. And then you have things like incrementality testing, right? We can do these holdout tests to say, this channel incremental at this time? Right? And so you have to understand all of these things are designed to work together. We call it triangulation. And then this is what a lot of our brains do. So my suggestion is as brands start to go and steal top of funny, they start to expand their, their Omni channel presence.

is to do their homework and really understand what is that right measurement stock because starting to grow your business with one form of measurement is incredibly challenging. And I've seen a lot of Durand's benefits in very meaningful ways by being thoughtful about it. And if there's anybody that's listening, I love to nerd out about this stuff. So if there's anybody that's just curious to talk about M &M, is there anybody curious to talk about artificial intelligence?

Speaker 1 (21:13.134)
I'm just coming up on my 11th year in the engineering artificial intelligence. so please don't hesitate to reach out.

I think there's very few people out there who really can stay there for 11 years in AI. So you start off, don't get me wrong, but you're sort of a dinosaur when it comes to that. And that's a good thing. And for all listeners, if you want to reach out to Mike, please do so. MMMs are really powerful, but you need to have the right mindset and you need to really understand what you're doing there. Mike, where can people go and find out more about you guys?

It's pressing to AI.com. I'm on pretty active on LinkedIn as well. yeah, feel free to contact us, contact me directly on LinkedIn or go right to the website and we can engage from there. So I'm based in Miami too, if anybody's ever done Miami, me a shout.

I will tell him.

I might pop in at some point. I will put all the links in the show notes and then you're just one click away. Mike, thanks so much to give us an overview of an MMM system and what it can do and a lot of people will reach out to you. Thanks so much for your time.

Speaker 1 (22:14.03)
Thank you very much. Take care. Tschüss.