In this interview, the guest (Max Kremer) explains how SaaS and e‑commerce companies can drive growth by focusing on the right marketing data. He argues that traditional last‑click or heuristic attribution is obsolete for modern multi‑touch customer journeys that involve dozens of interactions across channels (Facebook, Google Ads, email, video, retargeting, etc.). To fairly assign credit, he advocates using the Shapley value from cooperative game theory, which calculates each touchpoint’s marginal contribution based on statistical performance rather than arbitrary weights. He stresses the foundational need for clean, trackable data: every external link must carry well‑structured UTM parameters and a pre‑defined taxonomy, otherwise traffic appears as “direct” and attribution fails. As data volume grows, he describes the scalability challenge of capturing every click and keystroke, and shares how his company moved from fully managed AWS services to custom‑built EC2‑based pipelines, gaining fine‑grained storage control and cutting cloud costs to roughly one‑tenth of the original spend. The conversation ties together data hygiene, robust attribution methodology, and cost‑effective infrastructure as the three pillars for sustainable growth.
Chapters
Okay. Is there like, so if you have all these datasets, right, is there anything specifically now that maybe as a SaaS company that you find to be most important to To track specifically for generating growth and regarding on the customer behavior side? What are you seeing? You mentioned, you know, e-commerce stores, what are they tracking? What are they looking at?
So tracking everything and trying to be everything to everyone is sort of a double-edged sword. Analytics and in general, you know, martech, marketing tech, it's a huge space.
Yeah.
And it's a noisy space. So trying to be just another analytics player, we quickly realized, well, that's not going to work.
Got it.
So, you know, you have, of course, Google Analytics and you have enterprise versions of GA. You have the Adobes of the world, you have IBMs, you have, you know, all the incumbents. So to stand out, we really focused on marketing data. So we started pulling down AdWords performance data, Facebook ads performance data, and mashing it with our data. So now we're taking what you're spending on ads, right? Mashing that into the customer journey and building an attribution solution. So now you're not, you know, the model for understanding if an ad is working or not working has been like, okay, well, what are the— what's the conversion rate?
Okay.
How many people clicked on this ad and bought? But it doesn't work like that anymore. It worked like that 10 years ago when you Googled something and then bought it. But now you— the way people shop online is completely different. They go on Facebook, they'll click an ad, then they'll watch a video on YouTube. Then maybe they'll subscribe to a newsletter and then they'll get an email. and then maybe you'll retarget them and then they'll buy.
Exactly.
So what, all that journey and you're still giving credit to that retargeting ad that converted them, right? You have to measure your entire funnel.
Right.
And that's where we started tackling this attribution problem, 'cause people were still doing it in a very archaic manner. They were using last click, they were using first and last click, They were saying, okay, well, what if there's 10 clicks? Let's just divide it by 10. And, you know, those heuristic models are completely arbitrary. They don't make sense.
Right. So how do you kind of, you know, track that? And how do you give that marketing attribution correctly? So you have 10 different, I think, what's the average now? I think it's like 7 or 9 kind of times you have to hit somebody with some kind of marketing context.
Exactly. Something like 6 and 11.
6 and 11.
marketing touches, around a dozen, let's say, on the high end before someone converts.
So, I mean, you basically have to be everywhere and then keep hitting them on different angles. And then where do you give— where do you suggest giving that attribution? Where's the right place to do it?
So there's only one right way to do it. And actually the problem was solved back in the '60s by Lloyd Shapley, who won a Nobel Prize in mathematics for the Shapley value formula. And this is part of collaborative game theory. So which is a branch of mathematics. So collaborative game theory is all about how players play in a coalition, right? So if you think about hockey, right, you have the guy that scores, but the guy that gets the assist also deserves some credit. So the way I like to The analogy I like to use is a card game. Okay. So let's say we're playing a card game together. Okay. Me and you are playing and we win a bunch of money. We're playing against the house. We're a team. Someone else joins our team. We win an even bigger pot. Someone else joins our team. We win an even bigger pot. How do we split fairly our total winnings? Okay.
So, I mean, so one thing you mentioned could be equally, right? So one option is divided by the 4 of us, right?
Right. But just because you joined the team doesn't mean you did anything, right? So how do we measure that? Well, the way we do it is we look at your marginal contribution, right? So how did the team do before you? And then how did the team do after you? Right? So if you, if you take that analogy and map it to the marketing world, the players are the touchpoints.
Mm-hmm.
Okay, so you have your Facebook, AdWords, email, those are all your touches. And the total pot, the winning is your conversions, your sales, and you use the statistics. You don't try and attribute each individual sale. You use the statistical performance of the different channels, how they work together as a coalition to create sales.
And so would you give like a weight to each kind of I guess that's how you look at it, right? Like maybe double the weight for the retargeting, maybe half the weight for the initial piece of content they read.
It's not weighted. That's the beauty of it is you don't have to come up with arbitrary heuristics. The statistical performance gives you the weighting. It's kind of like a fancy weighted average.
Got it. Yeah.
The weighting is based on the performance.
Based on the data. So by this one person joining the team, if you're, winning streak just automatically kind of started going, instead of increasing, you know, every game, by them joining it over time, their value increases.
Yeah. You know that they're contributing marginally to, and they're boosting yourselves. They're providing lift.
Nice. So when it comes to data, I think one thing that's tricky is, you know, good data versus bad. I think that's always important. You want quality data to make better decisions, right? How do you guys, as TrialFire, or how do you suggest to people to make sure that what they gather is clean and verified? And then can you give an example between what's good and bad data for specifically marketing analytics that you see?
Right. So I'm sure you've seen that when you post links to a website somewhere and you want to measure the performance of those, you have to decorate those links with query parameters, UTM parameters they call them. That's sort of the de facto standard is these UTM parameters. It came from the GA world and kind of almost every marketing platform uses them today. And it's just a way of putting some metadata on those links. And it's just a matter of being regimented about building a scheme for, maintaining those parameters and sort of the taxonomy of what they mean. If you have links coming in everywhere and a link doesn't have those parameters on it, then it just looks like direct traffic essentially.
Mm-hmm.
So that's the first step. That was a bit of a learning experience for us that so many companies don't have their shit together in terms of like, maintaining that, right? You need to have sort of a clean framework of how you're gonna— like, your taxonomy for your marketing needs to be clearly defined ahead of time.
And so you suggest that, like, same thing you mentioned, like, would you have that everywhere? Like, you know, you're going on a podcast, you put a link there. Do you, you know— Absolutely.
Absolutely. So we stress that, like, don't put a link anywhere Anywhere in the world, on the internet, in your email signature, without those parameters.
Really?
So that way, that way, you know, what is driving traffic? And that's the start point. If you don't know where the traffic's coming from, then you have no visibility.
That's right. You can't give proper attribution. How do you do that as you scale, right? Because, you know, as analytics, there's so much more data that you have to manage and you have to, as it grows, what's your guys' strategy for Data analytics as you scale, and what's the system you provide to organizations or suggest to them?
So the scalability challenge for us was definitely prevalent. If you think about a fairly decent-sized organization with a lot of traffic, and you're talking about capturing not just every page view, but every single click and every single keystroke, then you have a lot of data. And initially, we just tried to piggyback our scalability on a scalable data infrastructure, and we depended on Amazon to do that. So we were using, of course, like everyone else, we were using AWS, not like a server under my desk or anything, right? So everything was on the cloud, but we started to just— we were kind of drowning in the fees, right? Like, sure, we can turn up the scale and okay, we can scale out and they make it really easy to do that, but it's also very costly.
It adds up pretty quick.
Yeah, it adds up real quick and we weren't really getting the performance we wanted. And a couple years ago, we realized that we have to fix this because it's eating too much of our— eating into too much of our profits. The Amazon bill was just getting out of hand. And we kind of had to, instead of using their data services, we kind of had to develop our own. Mm-hmm. It's still on Amazon, but now we have our own EC2 instances. We have very fine-grained control of the underlying storage devices that we're using. And we don't have like the fancy UI for like clicking and scaling things, but the cost is like, I would say, You know, we were paying, it's at least a 10th of what we were paying before. Wow. Wow. Yeah. Yeah. Yeah. Yeah.
Yeah. Okay. That's a much worth, much worth transition there.
