Why Marketing Attribution Breaks on Subscription Models

Lucas: The thing about subscription models is - they break your attribution system in a way most marketers don't see coming. Luna: How so? I mean, a purchase is a purchase, right? Lucas: Not when the purchase is a recurring billing event.

Let me give you a concrete example. A SaaS company I spoke with - about two hundred million in ARR - was running a standard last-click attribution model. Their dashboard showed that retargeting ads were the top driver of new subscriptions. Nearly forty percent of attributed revenue.

Luna: That sounds great. So what's the problem? Lucas: The problem is, they started doing cohort-based analysis - looking at when people actually decided to sign up, not when the credit card got charged. And they found that most of those retargeting conversions were happening days after the user had already made a decision.

The real first touch was a podcast mention three weeks earlier. Luna: Wait - so the retargeting ad got credit for a conversion that would have happened anyway? Lucas: Exactly. And this isn't a one-off.

Subscription businesses have this weird temporal mismatch. The decision to subscribe doesn't happen at the same moment as the transaction. There's often a gap - sometimes hours, sometimes days, sometimes weeks - between when someone decides and when they actually enter their card details. Luna: Because you've got to get your wallet, or you're on your phone and you'll do it later, or you want to check a competitor first.

Lucas: Right. And during that gap, retargeting ads, email reminders, push notifications - they all get the attribution credit. But they're not driving the decision. They're just poking someone who's already decided.

The attribution model is measuring timing noise, not influence. Luna: So if you rely on last-click or even multi-touch with a short window, you're basically rewarding channels that happen to be present at the transaction moment. Lucas: Exactly. And the fix sounds simple but it's actually pretty hard: you need to align your attribution window with your average time to convert.

Most platforms default to a thirty-day cookie window. But for a subscription product with a long consideration cycle, that window might need to be sixty days, or ninety, or even longer. Luna: But isn't that arbitrary too? I mean, why choose sixty days instead of thirty?

Lucas: Good question. You shouldn't guess. You should actually measure your distribution of time from first touch to transaction. Look at your cohort data.

If you see that the median time to convert is forty-five days, then a thirty-day window is cutting off half of your early-stage influence. You're basically blind to the channels that plant the seed. Luna: And that's where first-touch attribution becomes valuable again, right? Lucas: It does, but with a caveat.

First-touch attribution overcorrects - it gives all the credit to the first channel, which might just be a generic awareness play. What you really want is a model that understands the full journey but weights the early and middle touches more than the last-minute nudge. Luna: So something like a custom multi-touch model with a recency curve? Lucas: Yeah, but even that can be tricky.

Because in subscription models, there's another layer: the billing cycle itself. Let's say someone signs up for a free trial. That's an event. Then they convert to paid thirty days later.

Most attribution systems count the free trial signup as a conversion - or they count the paid conversion separately. But the decision happened at the trial signup, not at the billing event. Luna: So you'd want to attribute the trial signup, not the paid upgrade? Lucas: Exactly.

Or at least, you need to be very careful about what you call a 'conversion'. If your attribution model is counting paid subscriptions, and your average trial to paid conversion rate is, say, twenty percent, then you're attributing an event that's downstream from the actual decision. The marketing that drove the trial is the real influence. Luna: And that changes which channels look effective.

Because the channels that drive trial signups might be different from the channels that drive paid conversions. Lucas: Right. In the case I mentioned, the podcast mention drove a lot of trial signups. But those trial users didn't convert to paid for weeks.

By the time they did, the retargeting ads had already taken credit. The podcast looked like it had zero direct conversions. Luna: So the company was probably underinvesting in podcast advertising and overinvesting in retargeting. Lucas: That's exactly what happened.

They shifted budget after the cohort analysis. And their cost per acquisition actually dropped by about fifteen percent over the next two quarters. Not because they spent less - but because they spent on the channels that actually drove decisions. Luna: It makes me wonder how many other subscription businesses are making the same mistake.

Lucas: A lot, I'd guess. Because the default attribution settings in most platforms are built for e-commerce, not for recurring billing. E-commerce has a short, discrete purchase cycle. You see an ad, you click, you buy - that's a clean attribution path.

Subscription models have a delayed decision loop. Luna: And the platforms don't really encourage you to change the window. They just show you the default numbers. Lucas: Right.

So you have to actively question your data. One thing I recommend: run a side-by-side comparison. Take your last-click attribution report and compare it against a cohort-based view that looks at first-touch and assisted conversions over a longer period. If the two reports tell different stories, you have a window problem.

Luna: And then you can adjust your attribution window accordingly. But is there a rule of thumb for what the window should be? Lucas: I'd say look at your data first. But a common pattern I've seen: for B2B SaaS with a free trial, the window often needs to be sixty to ninety days.

For consumer subscriptions - think streaming services, meal kits - it's shorter, maybe thirty to forty-five days. But again, verify. Luna: And what about the billing cycle itself? Should you attribute at the subscription start or at the first payment?

Lucas: I lean toward attributing at the subscription start - when the user commits. Because that's the decision moment. The first payment is just fulfillment. But I know some finance teams want to attribute at payment because that's when revenue hits the books.

That's a reporting alignment issue, not an attribution accuracy issue. You might need two views: one for marketing decisions, one for financial reporting. Luna: Two sets of truth. That's messy.

Lucas: It is messy. But it's better than using a single flawed view for both purposes. If you only look at payment-attributed data, you'll systematically underinvest in top of funnel channels that drive consideration. And over time, your pipeline dries up.

Luna: I've heard about companies that switch to subscription models and suddenly their attribution data gets worse. This explains why. Lucas: Yeah, it's a common pain point. A company might go from selling one-time software licenses to a SaaS model, and all of a sudden their attribution reports look completely different - not because the marketing changed, but because the measurement framework no longer fits.

Luna: So the key takeaway is: if you're a subscription business, don't trust your default attribution settings. Go dig into your own time to convert data. Lucas: And once you do, you might find that the channel you thought was your top performer is just a late-stage reminder, not a decision driver. And the channel you were about to cut might be the one actually bringing in customers.

It's a classic case of measuring the wrong thing. Luna: And that reminds me - if these marketing conversations have sparked something you've actually used, whether it's a fix to your attribution window or a new way of looking at your cohort data, listener support is what keeps this show ad-free and independent. You can help at buy me a coffee dot com slash fexingo. Lucas: Yeah, it genuinely makes a difference.

But back to the attribution question - one more thing I want to flag. Even if you set the right window, you still need to watch out for what I call 'false recency'. That's when a user touches your brand, then goes dark for weeks, then comes back via a different channel and converts quickly. The attribution model might give credit to the last channel, but the real work was done earlier.

Luna: So you need to look at assisted conversions, not just last-click. Lucas: Exactly. Assisted conversions - or better yet, a full path analysis. For subscription models, I think the most useful framework is a time-decay model with a longer tail.

Give more weight to touches in the first third of the journey, and less to touches in the last few hours. That tends to align better with how subscription decisions actually happen. Luna: And that's something you can set up in most analytics platforms, right? Lucas: Most of them allow custom attribution models.

But you have to know what you're looking for. The default is almost never right for subscriptions. So take the time to audit your own data. It pays off.

Luna: Alright, so next time someone says 'retargeting is our best channel' in a subscription business, ask them: 'What's your average time to convert?' Lucas: Exactly. And if they can't answer that, their attribution is probably lying to them.