What an attribution model does
Most customers visit more than once before they buy, often from different channels. An attribution model is the rule you apply to those visits to decide which channel gets credit for the sale, and how much. It is the core choice in any marketing attribution setup.
The model never changes your total revenue. It only changes how that revenue is divided between channels, campaigns and pages.
Types of attribution models
| Model | How credit is split |
|---|---|
| First-touch | 100% to the first tracked visit |
| Last-touch | 100% to the last visit before the conversion |
| Linear | Equal parts to every visit |
| Time-decay | More to recent visits, halving with each half-life back in time |
| Position-based | 40% first, 40% last, 20% shared by the middle |
| Data-driven | Learned from your own converting and non-converting paths |
The first two are single-touch models. The rest are multi-touch attribution models. Every model also needs an attribution window: how far back a touchpoint can be and still get credit.
Ad platforms have moved toward data-driven models. Google Analytics 4 and Google Ads no longer support first click, linear, time decay or position based, and offer data-driven and last click instead.
Example: one sale, four models
A $1,000 sale follows four visits: a Reddit post, a Google search, a newsletter and a direct visit to the pricing page. The models split the $1,000 like this:
| Visit | First-touch | Last-touch | Linear | Position-based |
|---|---|---|---|---|
| $1,000 | $0 | $250 | $400 | |
| Google search | $0 | $0 | $250 | $100 |
| Newsletter | $0 | $0 | $250 | $100 |
| Direct | $0 | $1,000 | $250 | $400 |
Note that last-touch hands the whole sale to Direct, which tells you nothing about what to invest in. That is one reason some tools ignore direct visits when another source exists.
How to choose a model
- Want to know what brings new customers? Use first-touch.
- Want to know what closes deals? Use last-touch, ignoring direct visits.
- Long, many-touch journeys and enough data? Try a multi-touch model, but check how sensitive the result is to the model.
- Whatever you choose, keep it fixed so you can compare months.
Common mistakes
- Comparing numbers from tools with different models as if they should match.
- Picking the model that makes your favorite channel look best.
- Treating any model as proof of cause. Models divide credit; they don't measure what would have happened without a channel.
Which attribution model MetricRaven uses
MetricRaven uses first-touch attribution on real payments. Each payment is credited to the customer's first tracked visit, the earliest pageview across every visitor ID linked to them, in every report. It doesn't offer a model switch, so a channel's revenue means the same thing everywhere.
To see the touches first-touch leaves out, use assisted revenue in the Meta Ads and Google Ads reports, Paths to payment for the pages buyers saw across visits, and the Pages card's Viewed before paying credit. How visits get their channel is in Campaign & referral links.
Questions
- What are the main types of attribution models?
- First-touch, last-touch, linear, time-decay, position-based and data-driven. The first two give all credit to one visit; the others share it across the visits before a conversion.
- What attribution model does Google Analytics 4 use?
- GA4 offers data-driven attribution, paid and organic last click, and Google paid channels last click. First click, linear, time decay and position based are no longer available.
- Does the attribution model change total revenue?
- No. It only changes how the same revenue is divided between channels, campaigns and pages.