Attribution
Time-decay attribution
Also called recency model.
Time-decay attribution is a multi-touch model that gives more credit to touches closer to the conversion. Older touches get less.
How it is measured
Credit falls off with time before conversion, commonly by halving every few days. The model computes a weight for each touch from its age and normalizes the weights so they sum to the order value.
Pick the half-life from your buying cycle. A 7-day half-life suits quick purchases, and a 30-day half-life fits considered ones.
Worked example
A hot-tub dealer has an order of $6,400 with touches at 28, 14, 6, and 1 day before purchase: a magazine ad, a search click, a brochure download, and a showroom-visit email. With a 7-day half-life the weights come out near 5, 14, 34, and 47 percent.
The email gets about $3,000 and the magazine ad about $320. The dealer decides the magazine is an early introduction worth keeping at a small budget, rather than something to credit with the sale.
How it differs
Time-decay attribution favors recent touches. Linear attribution treats all the same. Time-decay sits between last-click and linear in how much it respects the early path.
Common errors
Setting the half-life without looking at the cycle. Using it with very short windows. Treating weights as truth. Ignoring touches outside the window. Comparing with a model using a different half-life.
In practice
Choose the half-life from your median time to purchase and write it in the report. Compare against linear to see which channels depend on the model.