Marketing mix modeling small business applications use statistical analysis to determine how different marketing channels contribute to total sales. By processing aggregate weekly data from the past two years, this method identifies the return on investment for each channel while accounting for external factors like holidays or price changes. It offers a privacy-safe way for mid-size brands to optimize their media spend without relying on individual user tracking or cookies.
Why is marketing mix modeling gaining traction now?
The digital landscape has changed significantly due to privacy regulations and the deprecation of third-party cookies. According to industry trends, 46.9% of US marketers are increasing their investment in marketing mix modeling as reported by various vendor surveys. This shift happens because traditional tracking methods like pixels often fail to capture the full customer journey across different devices and offline touchpoints.
For a mid-size business, relying solely on last-click attribution can lead to poor budget decisions. You might see a high return in Google Ads, but those users may have first discovered your brand through a Meta video ad. MMM looks at the big picture by analyzing how fluctuations in spend correlate with fluctuations in revenue over time.
What data does a mid-size business need for MMM?
To build a reliable model, you generally need approximately two years of weekly data. This duration is necessary to account for seasonality, such as the surge in sales during Black Friday or the typical summer slowdown. If you only use six months of data, the model might incorrectly attribute a natural seasonal peak to a specific ad campaign that happened to be running at the same time.
The data required is aggregate, meaning you do not need individual customer emails or names. You need the total amount spent on each channel per week and the total number of conversions or revenue generated. You also need to include non-marketing variables that could influence sales, such as economic indicators or even local weather patterns if you have physical storefronts.
- Media Data: Weekly spend and impressions for Meta, Google Ads, LinkedIn, and YouTube.
- Sales Data: Weekly revenue or order volume from Shopify, Amazon, or your CRM.
- Baseline Data: Dates for major holidays, price changes, and product launches.
- External Data: Competitor activity levels or regional economic trends.
How do open-source tools lower the entry barrier?
In the past, marketing mix modeling was an expensive service provided only by high-end consultancies. This has changed with the release of open-source libraries like Robyn from Meta and Meridian from Google. These tools allow businesses with internal data analysts or specialized agencies to run complex statistical models without paying massive licensing fees.
Robyn uses an automated approach to handle complex calculations like adstock (the lingering effect of ads) and saturation (the point where spending more money yields fewer results). Meridian, which is Python-based, offers a similar framework that focuses on transparency and allows for a deeper understanding of how different media channels interact with each other.
Understanding adstock and saturation effects
A common pattern in marketing is that the impact of an advertisement is not always immediate. This is known as the adstock effect. If a potential customer sees a YouTube ad on a Monday, they might not make a purchase until Thursday. The model uses decay functions to account for this time lag, ensuring that the credit for a sale is distributed correctly over the days following the ad exposure.
Saturation is another critical concept for businesses with mid-size budgets. Every channel has a point of diminishing returns. For example, a Shopify store selling organic coffee might find that spending $2,000 a week on Instagram generates $10,000 in sales. However, increasing that spend to $10,000 might only generate $20,000 in sales. MMM helps identify these saturation points so you can move budget to more efficient channels.
A comparison of MMM and multi-touch attribution
It is helpful to view MMM and Multi-Touch Attribution (MTA) as complementary tools rather than competitors. While MTA focuses on the individual user journey and short-term clicks, MMM provides a macro view of the entire business ecosystem.
- Data Privacy: MMM uses aggregate data (privacy-safe); MTA uses user-level data (privacy-sensitive).
- Scope: MMM includes offline factors and brand equity; MTA is mostly limited to digital clicks.
- Time Horizon: MMM is best for long-term strategic planning; MTA is better for daily campaign adjustments.
- Technical Focus: MMM uses regression analysis; MTA uses tracking pixels and cookies.
Steps to prepare your marketing data for modeling
The most time-consuming part of the process is data preparation. You must ensure that your data is clean and consistently formatted. A typical account should start by exporting historical spend from platforms like Google Ads and Meta Ads Manager at the weekly level.
Next, align this media data with your revenue data. If your fiscal week starts on a Monday but your marketing reports start on a Sunday, you will need to re-align them to ensure the correlations are accurate. You should also create a ‘holidays’ file that lists all major events that could skew your data, such as Labor Day or specific promotional periods unique to your business.
Variable selection for small business models
Do not try to include every single metric in your first model. Start with 5 to 10 key variables. These might include your top three spending channels, your average discount rate, and a seasonality index. Adding too many variables with limited data points can lead to ‘overfitting,’ where the model finds patterns that do not actually exist in reality.
Budget allocation scenarios for e-commerce
Consider a mid-size e-commerce brand that spends $50,000 per month across various platforms. By running an MMM using an open-source tool like Robyn, the brand might discover that their search ads are highly saturated, meaning additional spend there is inefficient. At the same time, the model might show that their video ads have a high ‘halo effect,’ driving indirect traffic to their website.
With these findings, the marketing manager can reallocate $10,000 from search to video. Since MMM accounts for the lag in video effectiveness, the manager will not be discouraged if they do not see an immediate spike in direct sales from the video ads. They will understand that the long-term impact will show up in the aggregate data over the following weeks.
Where to start this week
If you are ready to explore marketing mix modeling, begin by auditing your historical data. Check if you have at least 104 weeks of consistent spend and revenue records stored in your analytics or CRM system. This data is the foundation of any successful model.
Once your data is organized, you can leverage open-source documentation for Robyn or Meridian to understand the technical requirements for your data analyst. If you do not have an in-house analyst, consider a pilot project with an agency like Alien Road to build an initial model. This will help you identify which channels are truly driving growth and where you might be overspending.
Keep reading: Strategic Planning · Digital Marketing