Latest / Marketing Automation with Fexingo: HubSpot, Marketo, and Automated Customer Journeys / How Marketing Attribution Models Hide Real ROI
Transcript
- Lucas: So you run a LinkedIn ad campaign, HubSpot tells you it generated fifty thousand dollars in pipeline, you double down on the budget, and your CFO is happy. But what if that fifty thousand was going to close anyway, regardless of your ad? Luna: That's the attribution question, isn't it? Most marketers don't want to ask it because the answer might mean their campaign didn't actually do anything. Lucas: Exactly. And the thing is, marketing automation platforms like HubSpot and Marketo — they default to last-touch attribution. It's the simplest model to implement, and it's also the one that makes every campaign look like a hero. Luna: Last-touch meaning the last interaction before a deal closes gets all the credit. Lucas: Right. So if someone downloads an eBook from a LinkedIn ad, then two weeks later clicks a promotional email and buys, the email gets 100 percent of the credit, even if the LinkedIn ad was what got them in the door. It's misleading, but it's worse than that — it actively drives bad budget decisions. Luna: I've seen that play out. I worked with a B2B SaaS company — let's call them CloudMetrics — they were spending about two hundred thousand a quarter on LinkedIn ads. Their last-touch attribution showed a four-to-one return on ad spend. Lucas: Which sounds fantastic. What happened when they actually tested it? Luna: They ran a holdout test. Geographically randomized — they paused ads in three regions for three months. And the incremental lift from those ads? Less than ten percent. So their real ROAS was barely above one-to-one. Lucas: That's a brutal but incredibly common outcome. The thing is, last-touch is just one model. There's first-touch — which gives all credit to the first interaction. That makes your top of funnel content look amazing, but it gives no credit to the nurture sequence that actually converted the lead. Luna: And then there's linear — splits credit equally across every touchpoint. Better than single-touch, but also assumes every interaction is equally important, which we know isn't true. Lucas: Time decay is usually the most realistic of the built-in models. It gives more credit to touches closer to the conversion. But even that is a formula, not a truth. The real question is: did that touch actually cause the conversion, or was it just correlated? Luna: And correlation is the trap. Especially with automated workflows. You set up a nurture sequence, and you see that leads who open three emails convert at double the rate. But maybe those leads were already more interested. The emails didn't cause it — they just selected for it. Lucas: That's selection bias in action. And most marketing automation dashboards don't account for it. They show you a conversion funnel, and the numbers look clean. But the funnel is built on a false premise: that every step in the sequence added value. Luna: So what's the alternative? I mean, we can't run holdout tests on every campaign. Lucas: No, but you can run a few. Start with your highest-spend channel. Run a geo holdout or a time-based holdout for four to six weeks. Measure the difference in actual conversions, not attributed conversions. That gives you your incrementality baseline. Luna: Incrementality. That's the term — what did you actually cause, versus what would have happened anyway. Lucas: Exactly. And once you know your incremental lift, you can build a simple adjustment factor. Let's say your LinkedIn ads have an incremental lift of fifteen percent. That means for every dollar of attributed revenue, only fifteen cents is real. Apply that factor to your future campaigns, and you'll get a much more honest picture of ROI. Luna: But that requires discipline. Most teams I talk to say they don't have time for holdout tests. They'd rather just report the numbers the platform gives them. Lucas: And that's exactly how you end up with the CloudMetrics situation — spending two hundred thousand a quarter on ads that add almost no value. The time cost of running one holdout test is maybe ten hours of setup and analysis. That's nothing compared to the wasted spend it can reveal. Luna: What about Facebook? They have their own attribution system, and I've heard it's notoriously generous. Lucas: Facebook's default attribution window is a one-day click or twenty-eight-day view. So if someone sees an ad, doesn't click, but buys three weeks later after searching for your brand, Facebook still claims that conversion. It's designed to maximize the number of conversions reported to the advertiser. And it works — it makes the platform look good. Luna: But it's not necessarily fraud. It's just a choice about what to count. Lucas: Right. And every model is a choice. The problem is when you forget that it's a choice and start treating the output as objective fact. I've seen teams build entire quarterly plans around last-touch attribution data from HubSpot, never questioning whether those numbers reflect reality. Luna: So what's the one thing you'd tell a marketer to fix first? Lucas: Run one holdout test on your biggest paid channel. Doesn't matter if it's Google, LinkedIn, or Facebook. Just do it. And then compare the incremental lift to the attributed lift. The gap will tell you exactly how much your current model is overstating performance. Luna: And if the gap is huge, you have an honest conversation with your CFO about what the real ROI is. Lucas: Exactly. And you might end up reallocating budget to channels that actually drive incrementality. Email, for example, often has a much higher incremental lift than paid social, because it's reaching people who already opted in. Luna: Which brings us back to the automation platform. Most of these tools have the data to run this analysis — they just don't surface it as a default report. You have to build it yourself. Lucas: And that's the gap between using marketing automation as a tool and using it as a strategy. The tool will default to last-touch because it's easy. The strategist asks: easy for whom? And what are we missing? Luna: Speak of which — we cover this kind of stuff every week on the show, no ads, no sponsors. Just the frameworks we actually use. If that's been valuable to you, there's a way to support it. Buy me a coffee dot com slash fexingo. It's a small gesture, but it adds up, and it lets us keep the episodes ad-free. Lucas: Yeah, we love putting these together. And listener support is what keeps it independent. So if you've gotten something out of it, that link is buy me a coffee dot com slash fexingo. Okay — back to attribution. Luna: So we've talked about holdout tests. But what about multi-touch attribution models? Some platforms offer algorithmic attribution — where the model uses machine learning to assign credit. Lucas: Algorithmic is a step up, but it's still a black box. The model might be more accurate than time decay, but you have no idea why it assigns credit the way it does. And if you can't explain it to your CFO, it's hard to act on. Luna: Right. So transparency matters. Even if it's less 'sophisticated.' Lucas: I'd rather have a simple model I understand and can stress-test than a complex model I have to trust blindly. That's why I like the 'shapley value' approach — it's a game-theory method that fairly distributes credit across all touchpoints based on their marginal contribution. It's more transparent than neural-network-based approaches, and it's been around for decades. Luna: But it's not built into most marketing automation platforms. You'd have to export the data and run it separately. Lucas: True. But that's changing. There are tools now — like Rockerbox or Northbeam — that are built specifically for incrementality and multi-touch attribution. They integrate with HubSpot and Marketo. The data pipeline is there. Luna: It sounds like the real shift is moving from reporting to experimentation. Lucas: That's exactly it. Reporting tells you what happened. Experimentation tells you what you caused. And when you base decisions on causation rather than correlation, your marketing ROI goes up — not by a little, but by orders of magnitude. Luna: Alright, I'm going to go run a holdout test on our own campaigns this afternoon. Lucas: Do it. And let us know what you find. That's the kind of data that changes how a whole team allocates budget.