Which Attribution Model Ties Your Ad Spend to Revenue?

Get attribution wrong and you’ll cut the spend that’s quietly doing the work while pouring money into the channel that just happened to be last in line. That’s the stake behind every model debate: a rule for deciding which channel earned the sale, with real budget riding on the answer.

The trouble is that “attribution model” gets discussed as if picking the right one is the whole job. It isn’t. The model is the easy part. The hard part is having data clean enough that any model produces an answer you can trust.

That distinction is the through-line of this comparison, and it’s worth grounding in one piece of real work. Agora rebuilt a client’s measurement and attribution setup so paid-media spend decisions could be tied back to actual revenue instead of platform-reported conversions. The lesson from that project: the model you choose likely matters far less than the foundation it runs on.

What an Attribution Model Is Doing

A typical customer journey runs through several touchpoints before a sale, a social ad, an organic search, a retargeting click, an email. An attribution model is simply a rule for splitting the credit for one sale across all of them.

Every model is a different rule. None is objectively “true”, they’re all simplifications. The goal isn’t philosophical correctness; it’s to choose the rule that leads you to better budget decisions than you’d make otherwise.

The Single-Touch Models

These assign 100% of the credit to one touchpoint. They’re simple, easy to explain, and badly distorting.

First-touch gives all credit to the first interaction. It over-rewards awareness channels and pretends the rest of the journey did nothing. Useful only if your single question is “what introduces people to us?”

Last-touch gives all credit to the final interaction before purchase. It’s the most common default and the most misleading, because it systematically over-credits bottom-of-funnel channels, branded search, retargeting, that mostly harvest demand other channels created. Cut your awareness spend based on last-touch and you’ll watch your “efficient” bottom-funnel channels slowly starve as the demand feeding them dries up.

The Multi-Touch Models

These spread credit across several touchpoints. More realistic, more complex.

Linear splits credit evenly across every touchpoint. Fair and simple, but it pretends a throwaway impression matters as much as the click that closed the sale.

Time-decay gives more credit to touchpoints closer to the purchase. Reasonable for short consideration cycles; still biased toward the bottom of the funnel.

Position-based (U-shaped) loads credit onto the first and last touch, often 40% each, and splits the remainder among the middle. A pragmatic compromise: it honors both demand creation and demand capture. It’s also an arbitrary split you chose, not one the data revealed.

Data-driven attribution (DDA) uses your account’s actual conversion patterns to assign credit, comparing paths that converted against paths that didn’t. In principle it’s the most accurate because the weights come from your data rather than a rule someone picked. In practice it’s only as good as the data underneath it, and it’s a black box, so you can’t fully audit why it credited what it did.

A Side-by-Side Comparison

Model	How credit is split	Main bias	Good for
First-touch	100% to first	Over-credits awareness	Understanding discovery
Last-touch	100% to last	Over-credits closing channels	Quick, simple reporting
Linear	Evenly across all	Treats all touches as equal	A fairer first step
Time-decay	More to recent touches	Favors bottom of funnel	Short sales cycles
Position-based	Heavy on first + last	Arbitrary weighting	Balancing create vs. capture
Data-driven	From your conversion data	Opaque; data-quality dependent	Mature, high-volume accounts

The Mistake Hiding Underneath the Model Debate

Notice that every model above takes the underlying conversion data as a given. They all assume you know which touchpoints really happened and which sales really occurred. That assumption is usually wrong, and it’s where attribution projects quietly fail.

If your tracking is missing a chunk of touchpoints, blocked by ad blockers, lost to consent, dropped before the event fired, then even a perfect model is dividing credit across an incomplete map. If each ad platform is reporting its own conversions by its own rules, you’re not running one attribution model; you’re running several contradictory ones at once and trying to reconcile dashboards that were never meant to agree.

This is why the Agora client’s work was a measurement and attribution rebuild, in that order. Choosing a model came near the end. First, the team established where the truth lived, the backend record of real orders, and made sure conversion data reached analytics intact and reconciled against the revenue the business had actually booked. Only on that foundation does the choice of model become a real decision rather than an argument between unreliable reports. Once the data was solid, spend decisions could finally be tied to actual revenue instead of platform-reported conversions, and the model could be used to allocate budget with confidence in the numbers feeding it.

What Switching Models Does to Your Budget

The stakes get concrete the moment you switch models, because the same channels change rank. Under last-touch, branded search and retargeting look like your best performers, they sit closest to the sale, so they collect the credit. The awareness channels that created the demand show almost nothing, because their touch happened early and something else got the final click.

Run the same data through a position-based or data-driven model and the picture shifts. The awareness channels that introduced the customer start receiving a share of the credit, and the “efficient” closers look a little less heroic, because now they share the win with everything that set it up. Nothing about the business changed; only the accounting did.

That shift is the whole point. Reallocate budget on last-touch numbers and you defund the top of the funnel, slowly starving the closers that depend on it. Seeing the same sales through a fairer model is what stops you from cutting the spend that was quietly doing the work.

How to Choose, Practically

Skip the search for the “best” model, it doesn’t exist. Choose based on where you are:

• If your data is messy or your volume is low, don’t start with data-driven attribution. It needs clean, high-volume data to mean anything; run it on a shaky foundation and it’ll launder bad inputs into confident-looking outputs. Start with position-based as a sane default while you fix the foundation.
• If you’re drowning in last-touch reporting, the single highest-value move is to stop using it for budget decisions. Even switching to position-based will reveal awareness channels you’ve been undervaluing.
• If you have clean data and real volume, data-driven attribution is worth adopting, but pair it with the gut check below.
• Always cross-check against blended reality. Whatever model you run, total attributed revenue should reconcile to what the business actually earned, and big budget shifts should be validated with holdout or geo tests rather than the model alone.

The honest summary: attribution models are tools for allocating credit, not machines for producing truth. The working principle is that clean, revenue-reconciled data with a simple model will likely beat broken tracking under a sophisticated one, the advertiser running a “simple” position-based model on trustworthy data is better positioned than the one running data-driven attribution on top of tracking they can’t trust. Fix what you’re measuring first. Then the question of how to split the credit becomes one you can actually answer.