Are Your Google Ads Lying to You? - Marginal ROAS Explained | Growing Ecommerce

Episode Transcript

00:00:00 Mike:
Welcome to another episode of Growing Ecommerce. I’m one of your hosts, Mike Ryan, and with me today is Chris.

00:00:15 Chris:
Hey, happy to be here.

00:00:17 Mike:
What’s on your mind today, Chris? Or we can talk about the tub you installed.

00:00:22 Chris:
Oh, yeah. That was the next thing. You’d love it; it’s great—a tub and shower. I think the children love it even more than I do. We’re going way off topic here, but as my daughter’s phrase goes: “pitchy bocce ball.” Kibaki. Can you figure out what that means?

00:00:44 Mike:
No.

00:00:47 Chris:
All right, I’ll do it.

00:00:49 Mike:
Enlighten me.

00:00:52 Chris:
It’s a blend of English, German, and baby gibberish. She’s saying something like “splish splash.” For non-German speakers, this is a childish way of saying “butt cheek.” So, it’s her way of saying she’s splashing around naked, and for her, it’s like a battle cry.

00:01:08 Mike:
Did you buy the tub online, just out of curiosity?

00:01:14 Chris:
No, we ordered that through the plumber. But I buy almost everything online; I’m a loyalist to e-commerce. However, there are still cases where the good old buying behavior of going into stores to feel and see things is still there for me—like if I have to sit on it, like a chair or a couch.

00:01:54 Mike:
By the way, we got a new couch and we bought it online.

00:01:58 Chris:
Online? How brave. I can’t imagine.

00:02:02 Mike:
It was a good choice. We bought it from Westwing. It’s an amazing online shop with great products.

00:02:13 Mike:
So, small talk aside, what do we have? Do the letters SSOT mean anything to you?

00:02:22 Chris:
No.

00:02:24 Mike:
Single Source of Truth.

00:02:27 Chris:
Holy shit. I was going to say how generous of you, but wow.

00:02:40 Mike:
This is a great topic. Let’s delve into it. I’ve been observing that the number of advertisers who are measuring and optimizing based on a data source that is not the ad platform has been growing and growing over the past year. What do you think about this, Chris?

00:03:19 Chris:
Yes, for sure. Give us an example.

00:03:26 Mike:
I have a couple of examples. There’s no right or wrong here, just an opinion. I’m a very biased person, but I talked to a couple of decision-makers from huge online stores. It’s no surprise that these companies look at their own data in the back end, which is fine. Where I see a huge challenge is when you have your backend data set reflecting the true impact of your channels, yet these super advanced retailers take optimization decisions based on the data of the given ad platform.

00:04:35 Mike:
For instance, take Google PMax. If you try to optimize your PMax revenue based on Google Ads data, but the final evaluation of the PMax impact on your bottom and top line is different in your backend data set, that’s a breaking point. It bothers me because it’s a fundamental issue. I’ve seen big enterprise-level clients where the data simply doesn’t match up, and they are aware of it, but there is sometimes just no solution for it.

00:05:37 Chris:
I agree, but I view it in a pretty positive light as “growing pains.” If I think back several years to the days of last-click attribution, each channel or platform was its own little silo. You were optimizing exclusively within those walls.

00:06:27 Chris:
Back then, if you added up the revenue of all your platforms, it reached at least 150% because every platform was taking credit for the same purchases. We’ve seen maturity moving toward multi-touch attribution and Marketing Mix Modeling (MMM) becoming more popular in the last few years. A big inflection point was when Apple “nerfed” Meta ads with iOS 14.5 anti-tracking. That broke everyone’s Meta campaigns and the way they measured.

00:07:30 Chris:
Tools like Triple Whale, Rockerbox, and Northbeam flooded in—I called them “magic pixels.” They helped fix reporting, but then advertisers had to take actions in a platform that had no idea what was going on. There’s a disconnect. Over time, looking at other sources of truth is positive, but there is an optimization gulf. You’ll see advertisers adding an x percentage multiplier on their in-platform ROAS so it corresponds to their backend ROAS.

00:08:42 Chris:
At smec, we work with one huge client—a trusted relationship for almost ten years. They have a pretty advanced understanding of the true impact of every channel within and beyond Google. Our team optimizes based on a daily report they provide. They understand what it means if the ROAS in the back end increases by four percentage points and how to translate that to the Google Ads ecosystem. It feels clunky in the year 2025, but it’s better than just relying on ad platform data.

00:10:06 Chris:
I wonder if there is something more advanced out there. Advanced retailers love a testing mindset. We did a comprehensive geo-test with a client where Google Ads data suggested the new setup won by a wide margin. However, the backend data did not support those results. It makes it so hard to take optimization decisions when you move the needle on one end and don’t see a corresponding activity. It’s a substantial issue that threatens the platforms because they are no longer trusted as the source of truth.

00:12:10 Mike:
To Meta’s defense, they threw a lot of money at engineering and restored a lot of the fidelity of their tracking. In the intervening time, those third-party tools earned a lot of trust. Now, Meta is integrating with them, which might be the way of the future. Google isn’t there yet.

00:13:18 Mike:
That’s why Google is investing in things like Robyn and Meridian. If they aren’t the trusted source at the channel level, they need to own the cross-channel conversation between the CMO and CFO. Marketing Mix Modeling has been around the corner for a few years and will remain a very hot topic.

00:14:04 Chris:
You have to trust your MMM. It provides a valid single point of truth to understand cross-platform correlations. Every online retailer should take this step because looking at platform data that doesn’t match backend data has massive tactical implications.

00:15:33 Mike:
MMM is solving different problems; it’s not tactical. It helps you determine things like your saturation curve. Multi-touch attribution is falling out of fashion because of its dependence on user-level data.

00:15:56 Chris:
MMM might suggest spending 20% more on PMax, but that isn’t tactical advice on which specific campaign needs that budget. There need to be more synergies between strategic MMM and the “last mile” expertise we provide. We’ve brought up the problem today, even if we haven’t solved it.

00:17:15 Chris:
The solution to the problem is the problem of tomorrow.

00:17:16 Mike:
I’m going to think about that with a six-pack of beer. Company policy says we’re not supposed to drink before 4:00 PM, but for our Christmas episode, spiked eggnog is the only proper choice.

00:18:15 Mike:
Before we go down that rabbit hole, I have another topic. MMM often helps you understand your marginal ROAS per channel. Marginal ROAS is the idea of what you get back for the next dollar of ad spend. You can plot this to see a diminishing return curve that eventually flattens out.

00:18:47 Chris:
I love the concept. It’s very important to know because you can always put more money in, but the relation of what you get back matters. Is marginal ROAS widely adopted in the e-commerce world? Are retailers actually looking at it to understand if they are buying revenue at a margin-burning rate?

00:19:30 Mike:
Yes, to an extent. Google Ads has tools like Keyword Planner that show this curve, although it’s their estimation. Google has gotten better at this. However, many clients are just happy hitting their average ROAS target. If a campaign is at an 8.3 ROAS against an 8.0 target, they think everything is okay because the marginal ROAS isn’t visible.

00:21:51 Mike:
In practice, people look at impression share, but you don’t want 100% impression share because you’re eventually getting impressions that are no longer profitable. Google has been bringing up marginal ROAS lately to build a story about why cross-network advertising makes sense.

00:23:07 Mike:
They argue that if you look at each ad network as its own silo—Search, Shopping, Display—you might miss valuable conversions lingering just beyond the borders. They suggest your next conversion in Search might be more expensive than an incremental conversion in Shopping.

00:24:36 Chris:
It’s fascinating how Google simplifies complex topics. If I have a CPA target of 6 on Search, and the next conversion costs 7, it won’t be bought. But there might be a conversion available for a CPA of 4 within Shopping.

00:25:25 Mike:
Exactly. They compare it to fishing in a bunch of little ponds instead of the ocean. In principle, you get more conversions and better efficiency. But here’s the problem: Google says PMax bidding logic finds the cheapest conversion wherever it is. My understanding was that PMax was full-funnel, but it turned out to be very bottom-funnel and remarketing-heavy.

00:28:02 Chris:
The Channel Performance Report shows there is a lot of YouTube and Display feed-based advertising, which is essentially dynamic remarketing.

00:28:25 Mike:
This teamwork or multi-touch attribution story is a bit gone because these channels are essentially in competition with each other. Google tells us to “go easy” on channel reporting because the ROAS of one channel doesn’t tell the full picture. They say it’s about marginal ROAS, not average ROAS. But if it’s truly a competition for the cheapest conversion, the segmented view should be the holy grail.

00:30:46 Chris:
It’s a contradiction. Google has done a lot of right things lately by giving more control back to experts, and the marginal ROAS idea is the right way to look at spend. But if their claim is true, then the segmented view of each channel should be taken with the utmost seriousness.

00:31:51 Mike:
They argue you get the best of both worlds—more revenue and better efficiency. But when I look at the numbers, these other channels appear to be far overshooting where they should have gone on the curve. It might be a failure to communicate how complex this auction environment is with intra-day seasonality.

00:35:12 Chris:
It’s a hell of a claim. If they can execute on it, the segmented view of channel performance becomes highly relevant to understand what those incremental conversions actually cost.

00:36:09 Mike:
We’ll learn more about this soon. Soon we will be able to see channel performance at the MCC level, allowing us to find trends and aggregate data across many accounts.

00:36:56 Chris:
Time flies. We only have two episodes left this year. Next will be a Black Friday recap.

00:37:16 Mike:
I’ll bring the data and tips for that. Then we have the Christmas episode on December 23rd. We’ll show up in festive outfits.

00:37:51 Mike:
I love Christmas. I have a podcast playlist just for Christmas episodes. I love the positive vibes, and the US studios do a great job with the decorations.

00:38:22 Chris:
Nothing beats an Austrian Christmas market, though. I was in Vienna on November 1st and the Ritz-Carlton already had decorations up. They went full Mariah Carey.

00:39:02 Mike:
We shall also go for Mariah. This has been another episode of Growing Ecommerce brought to you by Smarter Ecommerce, also known as smec. You can learn more at smarter-ecommerce.com.

00:39:28 Mike:
If you enjoy this podcast, please share, follow, and give us a review. It’s always a pleasure.

00:39:33 Chris:
Likewise. Thanks a lot.