A customer sees your ad on Instagram. They forget about it. Two weeks later they Google your product name, land on your blog, and sign up for your newsletter. A month after that, they click a promotional email and buy.
Which channel gets credit for the sale?
That question, the attribution question, is probably the most argued-about topic in marketing analytics.
Why attribution is hard
The fundamental problem is that customers don’t experience marketing in neat, isolated buckets. They encounter your brand across multiple touchpoints, over time, with varying levels of attention. The decision to buy is usually the result of all of them together, not any single one.
But tracking systems need to assign credit somewhere. So they pick a rule and that rule shapes how you see the world.
The four common models (and what each gets wrong)
First-touch attribution gives 100% credit to the first interaction. Good for understanding awareness channels. Terrible for understanding what closes deals, the first touch might have been a display ad that barely registered.
Last-touch attribution gives 100% credit to the final interaction before purchase. This is what most ad platforms use by default, which is why they all look like heroes. It ignores everything that built intent before the final click.
Linear attribution splits credit equally across all touchpoints. More honest than the first two, but treats a 2-second display impression the same as a 10-minute product demo read.
Time-decay attribution weights recent interactions more heavily than older ones. Directionally sensible (recency matters) but assumes a model about customer psychology that may or may not be true for your audience.
None of these is “correct.” They’re all simplifications of a messy reality.
The incrementality question is the right one
Here’s the question that cuts through the noise: what would have happened if this marketing didn’t exist?
Would the customer have found you anyway? Would they have bought from a competitor? Would they have bought at all?
This is called incrementality: the marginal lift your marketing actually created, net of what would have happened otherwise. It’s the number that actually matters for budget decisions.
The hard part: measuring incrementality requires experiments. Run a campaign to 50% of your audience, hold the other 50% out, measure the difference. That’s a true test of what your marketing is doing.
Most companies don’t do this because it means “wasting” budget on the holdout group. But the cost of not knowing is usually much higher.
What to do in practice
You can’t run incrementality tests for every channel, every campaign, every week. So here’s a practical hierarchy:
For major budget decisions: Consider holdout experiments. Accept the short-term cost as the price of knowing what actually works. Do this at least annually for your biggest channels.
For ongoing tracking: Pick one attribution model and stick to it. The model’s choice matters less than consistency, if you change models every quarter, you can’t track trends. Document your model so everyone interprets reports the same way.
For campaign-level decisions: Look at assisted conversions, not just last-touch. Most analytics platforms show you how often a channel appeared in the path to conversion, not just whether it closed it. A channel that’s never the last touch but appears in 60% of conversion paths is doing something important.
For channels that can’t easily be tracked: Media mix modeling can help. So can simple surveys (“how did you hear about us?”). Neither is perfect, but directional data beats none.
The mindset shift
The attribution problem is ultimately a precision problem: you want to know exactly which dollar caused which sale, and you can’t. The uncertainty isn’t solvable with better software.
The useful reframe: attribution is a tool for portfolio management, not forensic accounting. You’re not trying to find the single cause of every sale. You’re trying to understand roughly what’s working so you can allocate resources better over time.
That goal is achievable, even without a perfect attribution model.
Spend more on channels where incrementality tests show lift. Diversify so you’re not dependent on any single channel’s self-reported numbers. Review the full customer journey, not just the last click. And hold your attribution data loosely, it’s a flashlight in a dark room, not a perfect map.