adstock playground

An ad keeps working after the week it ran. The first dollars work hardest. Slide the settings and watch both.

Spend pattern Weekly spend when on $ Max weekly lift at full saturation units

Carryover: share of last week's effect that stays Half-saturation: adstock where effect reaches half of max Steepness of the curve

Carryover is the geometric adstock rate. Half-saturation and steepness shape a Hill curve. These are modelling choices, not facts about your business. Slide them and watch what changes.

what the sliders say

Half-life of an ad

Total lift over 30 weeks

Lift per $1,000

Lift after spend stops

Share of all lift that lands after the last week you spent.

Week by week

Top: what you spend. Middle: the carryover stock, which is this week's spend plus what remains of earlier weeks. Bottom: the lift after the saturation curve.

Diminishing returns

For a steady weekly spend, the lift each week is on the vertical axis. The dot is your spend. The ring is double your spend.

How it works

The two ideas

Carryover. An ad keeps working after the week it ran. The stock for week t is that week's spend plus a share of last week's stock.

Saturation. The first dollars do more than the later ones. Lift is a Hill curve of the stock, so it rises quickly, then flattens.

stock(t) = spend(t) + carryover x stock(t - 1) lift(t) = max x stock(t)^a / (stock(t)^a + half^a) half-life = ln(0.5) / ln(carryover) weeks

Why this matters for modelling

Marketing mix models estimate these two shapes for each channel from your sales history. A model that ignores carryover blames the wrong week. One that ignores saturation says to put everything in the best channel. Seeing the shapes here makes the model output easier to question.

Read this before you use a number

This is a toy, not a model. Nothing here is estimated from data. The numbers show how the shapes behave and are not benchmarks for any channel.

Real carryover and saturation are unknown. They have to be estimated and tested, and different data can give different answers.

The curve shapes are modelling choices. Geometric carryover and a Hill curve are common. Real behaviour can differ.