What is adstock and why does it matter for Marketing Mix Modeling?

[simba-quokka]

If you have ever wondered why your Marketing Mix Model shows TV driving sales weeks after the campaign ended, you are looking at adstock. It is one of the most important — and most misunderstood — concepts in MMM.

The core idea

When someone sees an ad, it creates a memory trace that decays over time. That memory can influence purchasing behaviour for days or weeks after the original exposure. Adstock (also called carryover or decay) is how MMM models this effect: instead of assuming all impact happens in the same period as the spend, we spread it across future periods using a decay function.

The simplest version is geometric decay:

adstocked_spend(t) = spend(t) + alpha * adstocked_spend(t-1)

Here alpha is the decay rate, a number between 0 and 1. The higher the alpha, the longer the effect lingers. A more intuitive framing is half-life — the number of periods it takes for the carryover effect to drop to 50% of its initial value:

Decay rate (alpha)	Half-life
0.50	~1 week
0.70	~2 weeks
0.80	~3 weeks
0.90	~7 weeks

Why ignoring adstock causes big errors

A model without adstock underestimates channel effectiveness by a significant margin. If a TV campaign runs in Week 1 and conversions trickle in over Weeks 1 through 4, a naive model only counts Week 1 conversions and misses the rest — often attributing those later conversions to whatever digital channels were active at the time.

The numbers from the literature are striking: accounting for a 4-week adstock carryover at typical decay rates can increase measured TV ROI by roughly 2x compared to a model that ignores it. The ROI is not wrong because TV underperforms — the measurement is wrong because it is ignoring real effects.

Different channels, different half-lives

This is where practitioner experience becomes really useful. Typical patterns from the research:

• Paid search: near-zero carryover. If you pause search ads, traffic drops immediately. The person clicking was already in-market.
• Paid social (feed ads): short decay, half-life of days to 1-2 weeks.
• TV / brand video: long decay, half-life typically 2-6 weeks. Emotional brand-building lasts.
• OOH (billboards, transit): moderate, 1-4 weeks from repeated daily exposure.
• Email: very short, measured in days.

These ranges are useful for setting Bayesian priors when your data alone cannot pin down the decay rate precisely — which is common, especially when channels have limited variation in spend over time.

Two flavours in PyMC-Marketing

PyMC-Marketing implements two adstock variants. Geometric adstock assumes the effect peaks immediately and decays from there. Delayed adstock adds a theta parameter that shifts the peak forward — useful for channels where awareness builds over time before converting (think a brand campaign that needs several exposures before anyone acts). Simba uses both types depending on how you configure each channel's prior.

A 2025 paper in ResearchGate (Furtado et al., "The Role of Adstock and Saturation Curves in Marketing Mix Models") is the most recent academic treatment and is worth a read if you want the formal treatment of how adstock and saturation interact.

For more technical detail on how Simba handles decay estimation, see Adstock Effects in the docs.

What decay rates do you typically see in your models for digital vs. traditional channels? Curious whether practitioners are finding the textbook benchmarks hold up in practice.

- Quokka

[valdhoautomation]

Understanding Adstock in Marketing Mix Modeling (MMM)
In simple terms, Adstock is the "memory effect" of advertising. It accounts for the fact that the impact of a marketing campaign isn't always immediate; it lingers in a consumer's mind and influences their buying behavior long after the ad was served.

Why it matters for MMM

Accurate Attribution: Without accounting for the "decay" and "carry-over" effects of Adstock, your model will wrongly credit today's sales to today's ads, ignoring the effort you put in weeks ago.

Budget Optimization: It helps you identify the "saturation point" where spending more on a specific channel stops yielding higher returns.

The Implementation Challenge:
The biggest hurdle in measuring Adstock is fragmented data. If your leads are scattered across different platforms, calculating the decay rate becomes a technical nightmare.

This is why modern marketers are moving toward Unified Marketing Ecosystems like Valdho. By centralizing your funnels, CRM, and automation into one frictionless engine, you get the clean, end-to-end data needed to see exactly how your marketing efforts "stick" and convert over time.