Anyone who has ever been in a meeting discussing a Marketing Mix Model (MMM) will have heard the word “adstock”. It’s one of those terms analysts use often, and it can sound more mysterious than it is.
The concept is simple – adstock is a device we use in MMM to represent how the impact of advertising dissipates over time. The further the consumer gets in time from the message (or impact), the less the effect is estimated to be.
It’s called adstock because we like to think that advertising is building a “stock” of impact, and then that stock decays. It’s a useful model because, in practice, the response to advertising usually does follow a decaying pattern like this.
When an analyst builds an MMM, they will most likely use the adstock model to estimate the impact of advertising over time. Geometric adstock is itself a simple distributed lag. More flexible lag shapes are available – polynomial distributed lags (PDLs), or the Weibull and delayed adstock forms used in Robyn and Meridian – and they can capture effects that build before they fade, but they have more parameters and need more data to fit well.
To estimate the adstock series, the analyst will take your data – typically weekly TV impacts, gross rating points (GRPs) or spend – and apply an adstock (carryover) rate to the series: each week keeps that share of the previous week’s stock and adds the new activity. With a weekly rate r, the half-life is roughly ln(0.5)/ln(r) weeks – a 50% rate halves in one week, a rate of about 84% in four. The theory is that some media will have a longer impact on our memory structures than other ads, simply because sometimes we pay more attention. Think an epic ad in the cinema versus that ad you just scrolled past in your feed. Memorable ads will tend to have higher adstock rates, whilst “wallpaper” ads are quickly forgotten and carry little or no adstock. These rates translate into business impact – the effect carries over for longer when adstock rates are higher.
Easier to feel than to describe. Drop a weekly spend pattern in below, then nudge the rate up and down to see how the curve responds.
A few things worth noticing as you play with it. A short heavy burst with a high adstock rate keeps working for weeks after the spend stops. The same burst at a low adstock rate is mostly gone by the following week. And a small top-up in a later week behaves very differently against an empty tail than against a still-warm one – this is the bit MMM is meant to capture, and the bit that makes the choice of adstock rate matter when the model is read.
If you’d like to talk through what adstock rates are doing in your own model, drop a line to hello@themmmdoctor.com.