Time-decay attribution and the fashion buying cycle

Make the decay setting visible

One possible rule halves the weight after a chosen number of days. With a seven-day half-life, an interaction seven days before purchase has half the raw weight of an interaction at the time of purchase. Other implementations can use different rules.

Write down the chosen half-life, time unit and event timestamp. A team cannot reproduce a model comparison when a report displays only time decay without the underlying settings.

Consider the collection and buying cycle

Recency weighting can look intuitive for a short promotional window. It may underrepresent the earlier discovery of a considered purchase. The relevant question is whether the chosen weighting assumption suits the decision being discussed.

A partywear purchase planned weeks ahead and an immediately purchased accessory need not share the same journey. Do not interpret a single timing assumption as a finding about all customers or all product categories.

Normalise the weights

Calculate raw weights for the eligible interactions, then divide each by the sum of those weights. Multiply the resulting shares by the purchase value. That normalisation keeps the allocated total equal to the value being distributed.

Run a sensitivity check with another reasonable timing assumption. If channel priorities change sharply, show that sensitivity to the decision-maker instead of presenting one setting as a uniquely correct answer.

Turn the guide into a useful review

1. Record the half-life and timestamp basis.
2. Choose a window that fits the investigation.
3. Normalise weights before assigning value.
4. Compare the conclusion under another plausible setting.