Position-Based Attribution
Position-based attribution, also called U-shaped attribution, gives the most credit to the first and last touchpoints and divides the rest among the middle interactions.
Key takeaways
- Touchpoints are classified as first, last or middle and given fixed positional weights.
- The default forty-twenty-forty split is a configuration choice, not a measured property.
- A fixed middle pool divided among many touches shrinks each nurture step toward zero.
- Two-touch paths normalise back to an even split because no middle rows exist.
- Weights encode a stated belief about the funnel and should follow its real behaviour.
In depth
The model sorts every touchpoint into one of three classes by position, namely first, last or middle, and applies a fixed weight to each class. The familiar default gives forty percent to the first touch, forty to the last, and divides the remaining twenty evenly among however many middle touches exist. Those weights are a configuration choice rather than a property of the data, so thirty-forty-thirty or twenty-five-fifty-twenty-five are equally valid setups once you can defend the reasoning behind them.
The number of middle touches drives most of the variation, because the middle pool is fixed while the number of claimants is not. Twenty percent shared between two middle touches is ten each; shared among twenty it becomes one each, which effectively erases nurture channels from long paths. Widening the middle pool restores them but drains the two ends. Edge cases follow the same arithmetic: a two-touch path has no middle rows, so the weights normalise back to an even split.
Set the weights from how your funnel actually behaves rather than accepting the default. If the hardest part of the business is earning attention, weight the first position more heavily; if buyers arrive knowing what they want but stall on the decision, weight the last. In a quiz funnel the scorecard often occupies the last position for the lead conversion and a middle position for the eventual sale, so run the model against both events to see the quiz in each of its roles.
Positional weights ignore what actually happened at each step. A middle touch that answered the blocking objection scores the same single percent as an accidental page visit, and no amount of tuning fixes that, because the model has no way to tell the two apart. It also assumes the recorded first touch really was the first, which fails whenever discovery happened somewhere untracked. On paths with one or two touches the logic degenerates into a single-touch or even split.
Example in practice
How to measure it
Report credit by position class as three totals: first, middle and last. Where the middle pool is spread across many touches, also show credit per middle touch, because a channel holding a large share of middle appearances can still be paid almost nothing. Those three numbers show whether the configured split matches how the work is really distributed across your funnel.
Then test the weights for sensitivity. Recompute the channel ranking under two alternative splits, one favouring the first position and one favouring the last, and see which channels change places. Conclusions that survive all three weightings are safe to act on. Ones that flip between them are being produced by your configuration rather than by any difference in campaign performance.
Common mistakes
The common error is accepting the default split without asking whether the business matches it. Forty percent at each end assumes that discovery and closing are equally hard; for a company everybody already knows, that overpays the first touch dramatically. Write the assumption down as one sentence before setting the weights, and if you cannot defend that sentence to a colleague, stay on an equal split until you can.
The second is running the model where most journeys have only one or two recorded touches. With no middle rows it quietly becomes an even split between two channels, yet the report is still labelled position-based even though the positional logic never applied. Check the distribution of path lengths first. If half your paths hold a single touch, tracking rather than weighting is the thing to fix.
Frequently asked questions
Why is position-based attribution called U-shaped?
Because the credit forms a U when plotted across the journey, peaking at the first and last touchpoints and dipping in the middle. The standard split is 40% first, 40% last, and 20% shared by the rest.
When does position-based attribution work best?
It fits journeys where both creating the lead and closing it are clearly the most decisive moments. It is a strong default when you want to value demand creation and capture without ignoring nurturing entirely.
What weights should I use for position-based attribution?
Start from the default forty-twenty-forty only if discovery and closing look equally difficult in your business. Shift weight toward the first position when demand has to be created, and toward the last when demand already exists and the decision is the bottleneck. Whatever you choose, document the reasoning next to the report so the number can be challenged later.
How does position-based attribution split credit when there are no middle touchpoints?
The middle pool has no claimants, so the remaining weights are rescaled to sum to one. With the standard split a two-touch path ends up fifty-fifty rather than forty-forty, and a single-touch path takes the entire conversion. This is why very short paths make position-based and single-touch models produce nearly identical channel rankings.
Is position-based attribution the same as U-shaped attribution?
U-shaped is the best-known configuration of position-based attribution, the one that weights first and last equally and leaves a small middle share. Position-based is the wider family: change the weights and you get an asymmetric split, and add a third peak at a defined milestone and you get the W-shaped variant instead.
How does position-based attribution handle a very long path?
It compresses the middle. Twelve middle touches sharing a twenty percent pool receive under two percent each, so long nurture sequences look almost worthless next to the two ends. If long paths are normal in your funnel, either enlarge the middle share or group middle touches by channel before dividing, so a channel is paid once rather than per visit.
Can I change the position weights in my analytics tool or CRM?
Some CRMs and warehouse-based models let you set them directly; most advertising platforms do not, and several analytics products have retired rule-based options in favour of algorithmic ones. Where the weights are fixed, rebuild the calculation on your own path table. It is a simple query once each touch carries a source and a timestamp.
When should I not use position-based attribution?
Avoid it when journeys are short, when you have no reliable record of the first touch, or when the decisive moment reliably happens mid-path, such as a technical evaluation. In those cases the fixed weights actively mislead. A time-weighted model or a milestone-based one will describe the same journeys more honestly.