If this newsletter was forwarded to you, subscribe here to get these numbers in your inbox every week. 

In this issue: attribution models, why they matter, and how to use them. Also, biases and traps. 👩‍🔬

The attribution problem

If you’re like me, your entire career has been plagued by attribution models. One-day last-click. Linear. Time-decay.

Most attribution models weren’t built for today’s ecommerce journey. They were built for simplicity, so C-suite and finance could comprehend the impact of marketing spend. They were built to make sense of the unknowable magic impact of marketing on human behavior.

The problem is, these models rarely capture the true journey of your customers. When your customers are bouncing around between channels before checking out, simplicity is the enemy of truth.

Here’s what most of us get wrong about attribution: we think like marketers.  

Under one attribution model, we see one bad campaign and kill it. We see one good campaign and scale it. Then we’re confused by conversion metrics drying up, acquisition getting more expensive, and budgets getting slashed. 

It’s dramatic, painful, and confusing. Most marketers feel this all the time. But data doesn’t care about your feelings. 

Instead of thinking like a marketer, you should think like a scientist.

Becoming a scientist marketer

“Thinking like a scientist” means removing bias from your decision-making. It means evaluating campaigns objectively based on how they drive incremental revenue, not just short-term spikes or platform-reported ROAS.

It means knowing there is no single source of truth in marketing analysis. There’s only different ways of viewing “what happened.” A scientist knows that you must analyze results from every angle to shorten the distance between what you’ve observed and the actual truth. 

Every model comes with inherent strengths and weaknesses. The best scientist marketers know these features and work around them. 

Your goal as a scientist marketer: drive new customers above all else. 

For example:

Let’s say Meta says Campaign A is crushing. High ROAS, low CAC, one-day last click, driving tons of revenue. Under that same attribution model, Campaign B is doing horrible. So you think you want to slash budget on Campaign B, until you change your model.

Looking at new customer percentage, you realize that Campaign A is mostly bringing returning visitors, which is why it’s doing so well. And on a longer lookback window (say 30-day click) you realize that Campaign B is actually driving tons of incremental new visitors to your site, filling your funnel so Campaign A can perform. 

If you’d slashed Campaign B, Campaign A would have dried up over time. I’ve seen this at every single company I’ve ever worked for, and you probably have too.

Understanding attribution like a scientist

Attribution models are just formulas, tools that assign “credit” to various marketing touchpoints in a customer’s journey. Each model comes with its own lens. Some favor the beginning of a customer’s journey. Some favor the end. Some give equal weight to every interaction.

At Northbeam, we’ve developed models for every kind of business, sales cycle, and marketing mix:

Last Touch: Great for short, high-intent funnels.

First Touch: Ideal when you’re investing heavily in awareness.

Linear: Useful when every touchpoint matters equally.

Last Non-Direct: Filters out the noise from direct traffic.

Clicks Only: A conservative, “hard data only” model.

Clicks + Modeled Views: Northbeam’s most robust model, blending click data with proprietary view attribution for a truer picture of influence.

Clicks + Deterministic Views: Gives view attribution with actual view data provided by platforms, giving you a stronger deterministic understanding of when and where a view is impacting your conversion performance. 

Each of these models has a unique moment where it shines. Slavishly committing to one will cost you time, burn your budget, and give you grief.

Understanding model bias 

Every model is biased, even the fancy ones. The real problem is when you don’t know what those biases are.

Some overvalue the last click. Others ignore view-through influence entirely. Many struggle to handle long sales cycles or returning customers. And if you’re only looking at in-platform attribution, you’re cut off even further thanks to the walled garden problem. 

Your job is to slice through that noise. To ask:

Who really drove this sale?

Which campaign sparked interest vs closed the deal?

How do we measure that difference?

That’s what Northbeam’s attribution models are built for. How do you know when to use each model?

This is dependent on the nature of your business and your goals in each marketing analysis. 

Selling $30 impulse buys? You’ll want tighter windows and a more deterministic model (Clicks Only).

Selling $300 skincare systems with a 45-day decision window? You need Clicks + Modeled Views with an indefinite lookback to measure full funnel lift.

Same goes for your payback period: If you want to know how long it takes for a customer to become profitable, you better be measuring revenue on an accrual basis, not just when the transaction shows up in Shopify.

Good attribution is about matching your model to your reality. Not the other way around.

TL;DR

They want you to think attribution is a vanity tool, when really it’s a tactical advantage.

The marketers winning in 2025 are the ones who know exactly what’s working, why it’s working, and where to put their next $10K of spend. They’ve started thinking like scientist marketers and demanding accurate, actionable data you can filter in dozens of ways. That’s real clarity. 

Help me help you by booking a demo here. 

Cheers,

-bryan 

↗️ Applovin's Q2 earnings call contained a lot of huge revelations for advertisers. We'll be breaking this down in our next issue, stay tuned. 

😴 Is hustle culture dead? AI finally delivering on the promises of less work for all of us, perhaps?

🤖 An AI prompt for fast, simple storytelling DTC ads. This one looks like a winner. 

🎙️ My presentation on attribution at CommerceNext 2025. A deep dive into why your numbers suck and how to fix it with "The New Rules of Growth." 

Love this email? Please forward it to your friends, coworkers, or Lip-Bu Tan. Subscribe here to get this data in your inbox every week.