How time decay works
Time decay shares a conversion across every touchpoint in the journey, but not evenly. The nearer a touch happened to the conversion, the larger the share it keeps; the further back it sits, the smaller its share becomes. Credit fades as it ages rather than disappearing at a cut-off.
The fade is governed by a half-life: a period after which a touch is worth half what a touch at the moment of conversion would be worth. Shorten that period and the model becomes close to a final-click rule; lengthen it and the model drifts towards an even split. It is the one shared model with a dial you can match to how your customers actually decide, which is why the half-life setting deserves as much thought as the model choice itself.
As with any shared model, credit arrives in fractions. A single sale is spread across several channels, and the parts add back up to one real conversion.
Why time decay matters
It fits the way most considered purchases feel. Something introduced you, several things kept you interested, and the touches in the final days did the persuading. Time decay records that shape without pretending an ad seen weeks ago mattered as much as the one seen an hour before the enquiry, and without writing that old ad off entirely.
It is also the most defensible shared model to present to someone paying the bills, because the reasoning is intuitive. Recent contact influenced the decision more. Nobody needs to trust a machine-learned weighting to accept that.
Common mistakes with time decay
Leaving the half-life at the default is the usual error. A default tuned for a quick online purchase, applied to an education agency where families deliberate for months, crushes the credit of everything that happens early in the process and quietly hands the account to retargeting and branded search.
The other mistake is reading it as a measure of influence rather than of recency. The model does not know which touch persuaded anybody; it only knows which one happened last. A channel that always appears late will always look strong under it, whether or not it changed a single mind.
How to act on it
Set the half-life against your real sales cycle before you read a single report. Look at how long passes between first contact and enquiry for the deals you actually win, and choose a period in that neighbourhood rather than the one the tool ships with.
Then check what the weighting is rewarding. If your late-stage channels are dominating, ask whether they are creating demand or collecting it — and test by pausing one rather than by adjusting the model until the answer looks better. Where paths look shorter than the real customer journey, the problem is usually measurement rather than modelling, and that is a job for your analytics and tracking setup. Compare the same period under linear attribution to see how much of the picture is being driven by recency alone.
