Adstock is the idea that advertising continues to influence buying behaviour after the ad has aired or been shown. A marketing mix model uses adstock to spread the effect of a week's spend across the following weeks, using a decay rate that reflects how quickly the effect fades.
The concept was introduced by Simon Broadbent in the 1970s and remains one of the most important building blocks in any MMM. Without adstock, a model would only look at what media spend and sales did in the same week, which misses the reality that advertising builds memory and intent over time.
How adstock works in practice
Imagine you run a TV campaign in week 1 and nothing else in weeks 2 and 3. Without adstock, the model would see no TV activity in weeks 2 and 3, even though viewers who saw the ad are still influenced by it. Adstock solves this by carrying a portion of week 1's spend forward into subsequent weeks, with the amount declining each week according to a decay rate.
If the decay rate is 50 percent, then 50 percent of week 1's effective spend carries into week 2, 25 percent into week 3, 12.5 percent into week 4, and so on. The model then uses these transformed spend figures, rather than the raw spend, as its input variable.
Decay rates vary by channel
- TV and radio: typically longer decay, 50 to 70 percent carry-over per week, because broadcast media builds memory strongly
- Out-of-home: moderate decay, as exposure is repeated passively over a campaign period
- Paid search: very short decay, often near zero, because people search when they are already in a buying mindset
- Paid social: short to medium decay, depending on whether the campaign is brand or performance focused
- Print: moderate decay with a spike effect around publication date
Why adstock changes your ROI numbers
Applying adstock means the model attributes some of next week's sales to this week's advertising. That increases the measured effectiveness of channels with long carry-over effects, such as TV. It reduces the measured effectiveness of channels with short carry-over, such as paid search, relative to a naive model that only looks at the same week.
Getting the decay rate wrong can significantly distort ROI comparisons between brand-building and performance channels. If you underestimate TV's carry-over, you will understate its long-term value and potentially underinvest in it. This is one reason why the adstock assumptions should be reviewed and agreed with you, not set silently by the modeller.
Geometric decay versus Weibull adstock
The simplest form of adstock uses geometric decay, where the same percentage of effectiveness carries over each week. More advanced models use a Weibull distribution (a flexible mathematical curve) that can capture delayed effects. With Weibull adstock, the peak effect of an ad can occur one or two weeks after it ran, before decaying. This better reflects how some formats, particularly brand campaigns, build awareness over time before driving purchase.
Modern open-source tools such as Google Meridian and Meta Robyn both support Weibull or similar flexible adstock functions. Ask your modelling partner which form they use and how they test which one fits your data.
Questions to ask about adstock assumptions
Every MMM contains explicit or implicit adstock assumptions. If your provider cannot tell you what decay rates they used for each channel, or if they used the same rate for every channel, treat that as a red flag. The decay rates should be estimated from your data or calibrated against industry benchmarks, and they should be documented.
Does adstock only apply to traditional media, or does it apply to digital too?
It applies to all channels, but the decay rates differ considerably. Digital performance channels such as paid search have very short carry-over because the click happens immediately. Brand-focused digital formats such as YouTube or connected TV have longer decay, similar to traditional TV.
Can we set adstock parameters ourselves, or does the model estimate them?
In Bayesian MMM, adstock parameters are typically estimated from the data with priors informed by industry benchmarks. In traditional MMM, they are sometimes set by the modeller based on experience. Either way, you should see and approve the final values before accepting the model output.
If we stop advertising entirely, how long does the adstock effect last?
It depends on the decay rate. With a 50 percent weekly decay, 97 percent of the original effect has dissipated within 8 weeks. With a 70 percent weekly decay, meaningful carry-over can persist for 20 or more weeks. Your modeller can show you the exact decay curve for each channel.