Campaign attribution analysis

Campaign attribution analysis is the practice of assigning credit for a conversion across the marketing touchpoints a customer interacted with on the way to buying. A customer might click a paid search ad, read an email a week later, and convert after a retargeting display ad. Attribution decides how much of that one sale each touchpoint earns. Under a last-touch model the display ad takes all the credit; under a linear model the three touchpoints split it evenly.

The analysis matters because marketing budgets are allocated on the back of it. Get the model wrong and you over-fund the channel that happens to sit last in the journey while starving the channels that created the demand. The goal is not a single perfect number but a consistent, well-understood view of which campaigns genuinely move customers toward a conversion rate you can trust.

Attribution works by giving every touchpoint in a conversion path a credit weight, where the weights for one conversion sum to one. Add up the weights a channel earns across all conversions in the period and you get its attributed conversions. Multiply by deal or order value and you get attributed revenue, which is what spend should be judged against.

The weights come from the model you choose, and the model is the single biggest decision in the analysis. The inputs below are what you assemble before any model can run.

1. 1

   Touchpoint data

   Every tracked interaction tied to a user identity, from ad clicks to email opens to site visits, with timestamps so the path can be ordered.

2. 2

   Conversion events

   The actions that count as success, such as a signup, a qualified lead, or a purchase, with their value.

3. 3

   Lookback window

   How far back before the conversion a touchpoint can still earn credit. A 30-day window and a 90-day window produce different results.

4. 4

   Attribution model

   The rule that distributes credit: last-touch, first-touch, linear, time-decay, position-based, or data-driven. This decides every weight.

An attribution report tells you which channel earned the credit, but the number alone hides what drove it. A metric tree decomposes attributed conversions into the levers beneath them, so a shift in a channel result points to a reachable cause rather than a black-box model output. Attributed conversions sit at the top, and the drivers that build them sit underneath.

When an attributed result changes, the tree separates a reach problem from a quality problem from an efficiency problem. More impressions, a better click-through rate, or a stronger landing page all lift attributed conversions, but each sits with a different team and a different fix.

There is no benchmark for an attribution number itself, because it depends entirely on the model and the journey. The useful comparison is how much the credit for a channel shifts between models, which tells you how exposed your decisions are to the model choice. The table below shows how common models distribute credit, so you can read the same data through each lens.

Improving attribution is less about finding the one true model and more about making the analysis trustworthy enough to act on. That means cleaner identity data, a model matched to your sales cycle, and a habit of validating attributed results against real lift. The levers below target the branches that make the number reliable.

1. 1

   Trusting last-touch by default

   Last-touch over-credits the closing channel and starves the campaigns that created demand. It is a starting point, not an answer.

2. 2

   Ignoring the lookback window

   Too short a window drops early touches; too long a window credits irrelevant ones. The window quietly reshapes every result.

3. 3

   Confusing attribution with incrementality

   Attributed credit is not the same as causal lift. A channel can earn credit for conversions that would have happened anyway.

4. 4

   Comparing across models

   Putting a last-touch number next to a linear number in the same table makes one channel look better for no real reason.