Why most marketing attribution models are broken
We spend enormous energy attributing conversions to channels, yet the models we rely on systematically misrepresent how customers actually decide. Here's what I think we're getting wrong.
Contents
Contents
There’s a peculiar ritual in most marketing teams. Every Monday morning, someone opens a dashboard, looks at which channel “drove” the most conversions last week, and makes decisions accordingly. More budget to paid search. Pull back on display. Email is performing — push it harder.
The problem isn’t the process. The problem is that the underlying model is almost certainly wrong.
The seduction of measurability
Attribution exists because someone, somewhere, needed to justify spend. And when you’re accountable for budget, measurability becomes deeply attractive. If you can point to a number and say “this channel produced X conversions,” you feel like you’re running a rational operation.
The trouble is that what gets measured gets optimised for — and most attribution models measure the wrong thing.
Last-click attribution, still the default in many tools, gives 100% of the credit to whatever a customer interacted with immediately before converting. It’s intuitive. It’s also a fiction. The person who clicked a paid search ad to buy didn’t materialise out of nowhere. They’d likely read a blog post six months ago, heard about you from a colleague, seen your product in a newsletter, and maybe watched a YouTube video before they ever typed your brand name into Google.
Last-click looks at the final step and says: that’s the one that worked. It’s like crediting the last person who knocked on a door for closing the sale, and ignoring everyone who made the introduction.
Multi-touch doesn’t fix this
The natural response is to move to multi-touch attribution — split credit across touchpoints. Linear models. Time-decay models. Position-based models. These are better, but they share the same foundational assumption: that the customer journey is a measurable sequence of discrete events, each of which can be assigned a causal weight.
That’s not really how decisions get made.
Consider the actual mechanics of a considered B2B purchase. The buyer sees your content shared on LinkedIn. Doesn’t click. Weeks later, a colleague mentions your product in passing. They Google you, read the website, close the tab. A month after that, a podcast they listen to interviews your founder. Now they’re genuinely interested. They sign up for a trial directly, typing your URL from memory.
In your attribution model, that looks like direct traffic. The podcast, the word of mouth, the LinkedIn impression — none of it registers. You conclude that direct is your strongest channel and that social and content are underperforming. You cut content investment. The pipeline slowly dries up, and you can’t explain why.
The dark matter of marketing
There’s a category of influence that attribution tools fundamentally can’t see. Some people call it dark social — shares that happen in Slack channels, DMs, email threads, WhatsApp groups. Private recommendations. Things people read but never click. The ambient presence of your brand in the places your customers actually spend time.
This isn’t a niche edge case. For many businesses, especially in B2B, dark social and unmeasured influence account for the majority of how buyers actually discover and form opinions about products. The trackable stuff is often just the last visible step in a longer, largely invisible process.
What attribution is actually good for
I’m not arguing you should abandon attribution. It’s useful — but for a narrower purpose than most teams use it for.
Attribution is a reasonable tool for tactical optimisation within channels. Which ad creative is outperforming? Which landing page converts better? Which email subject line drives more opens? These are questions where attribution data is relatively clean and the feedback loop is tight enough to be trustworthy.
It’s a poor tool for strategic allocation across channels, especially for channels that operate at the top of the funnel or through influence that doesn’t leave a trackable footprint. Using last-click or even multi-touch data to decide whether to invest in brand, content, or community is like using a thermometer to measure the weather a week from now.
A better mental model
The most honest thing you can say about marketing’s contribution to a pipeline is: we think these activities are building the conditions under which people become interested and eventually buy. That’s less satisfying than a dashboard, but it’s closer to the truth.
This means getting comfortable with a mixed measurement approach. Use attribution data for what it’s good at. Layer in surveys — ask customers how they first heard about you, what made them take it seriously, who else was involved in the decision. Run brand awareness studies. Pay attention to your search volume for branded terms over time. Model incrementality where you can.
Most importantly, resist the temptation to cut anything you can’t directly attribute. Some of the most valuable marketing is the kind that never shows up in a dashboard at all.
The companies that understand this tend to build stronger brands over time. The ones that optimise purely for what’s measurable often find themselves in a race to the bottom on paid acquisition costs, wondering why their unit economics keep getting worse.
This is the kind of thing I think about constantly working with clients at TopOut Group. If it resonated, subscribe below — I write about this stuff regularly.