What is Attribution

Common mistakes

Trusting a single model as truth, ignoring the growing gap caused by privacy changes and cookie loss, and forgetting that a lot of demand (word of mouth, offline, dark social) is never tracked at all. Modern measurement pairs attribution with incrementality testing.

Why it matters

Attribution is how you assign credit for a conversion across the many touchpoints a customer interacts with before buying — a search here, a social ad there, an email, a branded search, then the purchase. It matters because your attribution model directly shapes budget decisions: a last-click model hands all the credit to the final touch (often branded search or direct), systematically under-crediting the upper-funnel channels that created the demand, and can lead you to defund exactly what is driving growth. Getting attribution wrong quietly misallocates your entire marketing budget.

• Assigns conversion credit across the multiple touchpoints in a journey
• Your model shapes budget decisions — the wrong one misallocates spend
• Last-click under-credits demand-creating, upper-funnel channels

Models and their trade-offs

No model is "correct"; each tells a different story. Last-click over-credits the closing channel and is simple but misleading; first-click over-credits discovery; linear spreads credit evenly; time-decay weights recent touches more; position-based (U-shaped) rewards the first and last touches. Data-driven attribution uses your actual conversion paths to assign fractional credit and is generally the most accurate where you have the volume. The practical stance is to know which model your reports use, avoid trusting last-click alone for budget decisions, and increasingly rely on incrementality tests (does turning a channel off actually reduce conversions?) as the ground truth privacy changes make attribution harder.

• Last-click: simple but over-credits the closer; first-click over-credits discovery
• Linear, time-decay and position-based spread credit differently
• Data-driven attribution is usually most accurate at sufficient volume
• Incrementality testing is the ground truth as tracking gets harder

In practice

A marketing team, judging channels by last-click, sees branded search and direct traffic as their top performers and considers cutting the 'inefficient' upper-funnel display and YouTube spend. Before doing so, they run an incrementality test — pausing the display campaign in a set of regions and comparing conversions against untouched regions. Conversions in the paused regions fall, revealing that display was creating demand that later showed up as branded search and direct in the last-click report. Cutting it would have quietly reduced total conversions while making the last-click numbers look even better. They switch budget decisions to data-driven attribution supported by incrementality tests, and stop trusting last-click for anything strategic. The example captures attribution's core danger: the model you choose shapes where you invest, and last-click systematically defunds the very channels that create demand.