AI-Assisted Marketing Mix Modeling 101

Which marketing efforts are the biggest contributors to your sales, market share, and profitability? How can you best use this information to optimize your marketing efforts? Digital marketers have been working to successfully answer this question for a long time now.?

It’s a bigger challenge than ever, thanks to the widespread use of “omnichannel marketing,” especially when you’re selling a B2B long-sales-cycle product or service. There can be quite a few actions customers can take before they finally decide it’s time to talk to a salesperson.

Marketing Mix Modeling (MMM) is the science of using regression analysis, time series analysis, and machine learning algorithms to answer those important questions.?

MMM is rising in importance and is also benefiting from the analytical power of new AI tools. However, it’s important to say here that it is not a perfect tool. As with all analysis, there’s a “garbage in, garbage out” factor. There will be times when your data analyst needs to make a “best guess” entry, especially when you’re trying to assign a numerical value to an aspect of your business that is difficult to quantify.?

Our goal with this article is to make MMM understandable so you can put it to work for your company.?

AI-Assisted Marketing Mix Modeling 101

The term “marketing mix” is usually seen now as the collection of channels and methods (SEO, social, content, email marketing, online advertising, etc.) that you’re using to attract and educate your customers.?

Back in 1949, when the term was introduced by Niel Borden,1 a professor at Harvard Business School, he defined the marketing mix as the set of marketing elements that a firm could use to achieve its objectives, including “product, price, place, and promotion.”?

• “Product” includes the design, features, quality, packaging, branding, and other associated services or warranties.2
• “Price” is the price of your product or service, obviously.
• “Place” is where your messages appear or the product is sold. Messages and products now appear almost exclusively online (sites, social, email), but distribution, retail locations, and logistics are also part of the mix for products sold in physical stores.
• “Promotion” is also primarily carried out online, in a variety of channels.?

The goals of Marketing Mix Modeling

You want your marketing to work. You want to see how your marketing efforts are impacting your sales. You want to be able to use reliable data to make solid marketing decisions. MMM can help.

Pros of MMM

When it is done properly, MMM will improve your:

• Budget allocation. You will be able to invest more heavily in the marketing efforts that result in the highest return, and reduce spending on the efforts that haven’t yielded the highest results.
• Decision-making. “Data-driven decisions” is becoming a popular term, and for good reason. It’s sadly too easy to fall into a trap of false beliefs in the absence of reliable data. Decisions based on these false beliefs lead to disappointing outcomes. When you have the right data, decisions involving brand promises, customer expectations, and budget allocation are much easier to make, and lead to better outcomes. You can also more accurately forecast.
• Internal discussions about marketing. Without proper data, marketing discussions are subjective, dominated by “we’ve always done it this way,” or “this is how we did it at my last company,” or “that wouldn’t appeal to me at all” (when in fact, it might be precisely what? customers find appealing). Data comes to the rescue here.?

It’s important to note that even the most sophisticated data analysis is not a substitute for interviewing customers who have already purchased your products. Understanding their very specific Mindset when they set out to buy (consisting of their desires, concerns, and questions) will make sure that your messages appeal precisely to your target audience and support the steps in their buying process.?

For that reason, we tend to think of MMM more of a help choosing channels (where your message will appear) rather than the messages themselves.?

Challenges/Cons of MMM

1. Complexity. MMM may be challenging to interpret and requires advanced statistical skills.
2. Lack of Real-Time Models. MMM is driven by historical data, so insights are not always current, making it difficult to incorporate sudden changes in market conditions.
3. Inaccurate Attribution. Accurately attributing the impact of an independent variable to dependent variables.
4. Data Quality & Availability. Data may be incomplete, inconsistent, or inaccurately recorded, making it difficult to conduct meaningful analysis.
5. Data Granularity & Timeliness. Limited granularity may obscure important insights, while delays in data availability may hinder real-time decision-making.
6. Cross-Channel Integration. Integrating data and modeling across multiple marketing channels presents challenges due to differences in measurement methodologies, data formats, and attribution models.
7. Privacy & Compliance Concerns. Ensuring data security, anonymization, and compliance with legal requirements is essential to protect customer privacy and mitigate legal risks.

How Marketing Mix Modeling works

OK, time to get into the nitty-gritty.?

What needs to go into the model is much of what we’ve talked about. Marketing activities and their respective costs, pricing info, distribution channel info, and external factors such as economic indicators and competitive marketing activities.?

What you choose for input will be unique to your company, industry, the types of products or services that you sell, and the marketing channels and methods that you are using. Your market might be seasonable. Internal or external issues might come into play.?

Each marketing channel has its own metrics.?

With online advertising, for example, it’s impressions, clicks, engagement, site visitor sessions, the various ways that customers have interacted with your content, and actual sales in a business-to-consumer situation.?

Even in a long-sales-cycle business-to-business buying process, if you have a strong CRM and reliable input from salespeople, you can track a customer’s activity all the way through to a sale to see your ROAS (Return on Ad Spend).

Each type of product or service has its own sales cycle. The speed with which a person decides to buy from you depends on the amount of scrutiny that the person applies to the purchase.?

The entire exercise of identifying the appropriate KPIs will be much easier if you have already mapped out the typical buying process for your type of products and services.?

You will need someone who enjoys this kind of analysis to own the process of building your list of KPIs. Your list will be unique to your company, offerings, and situation.?

Two categories of sales may come into play in your model:?

• Base Sales. These are sales that occur with no marketing. A referral customer or a returning customer are examples.
• Incremental Sales. Sales that come about as a result of marketing efforts. They are usually broken into two types: “above the line” and “below the line.”
• Above the line: Broad-reach marketing efforts that help with brand-building and creating awareness.
• Below the line: Highly targeted campaigns, focusing on customers who are most likely to buy the product or service and are already in the market.?

Laying the groundwork carefully here will ensure the best outcome.?

What should come out of the model are metrics related to sales, market share, customer acquisition and retention, and profitability. Some products or services may be more profitable than others; being able to tie campaigns to profitability is about as good as analysis gets.?

Inputs include how much you’re spending on various marketing channels, your current pricing and possible changes in pricing, distribution channel allocations, and external factors such as economic conditions, seasonality, weather patterns, and industry trends.?

Outputs include predicted sales over a given period, market share compared to the competition, and customer acquisition and retention.?

The tools MMM practitioners use

Here are just a few of the tools that MMM practitioners use. This is definitely not a complete list, but it will give you an idea of the resources and processes used for MMM.?

There are a number of companies coming on the scene that use AI to take these tools and build them into a campaign mix platform. One we have been pleased with so far is TripleWhale.com.?

Regression analysis

One of the tools used is multi-linear regression (MLR), which models the linear relationship between a “dependent” variable and “independent” variables.?

Dependent variables are the ones that you’re turning into your KPIs, such as revenue, market share, and customer lifetime value, the amounts you’re spending on a marketing channel, the price of your products, etc.?

Independent variables are those that somehow have an influence on the dependent variables, such as the amount you’re spending on advertising, the price of your product or service, competitive activity, economic conditions, and more.?

Time-Series Forecasting

This tool is used to forecast future values derived from historical data points collected over a time period and collected at regular intervals. Basically you will be predicting future values using previously collected data.?

You will be looking for trends, while filtering out the information that is less relevant (the “noise”). Various machine learning models can be used for this.

Applications include forecasting sales, allocating budgets, planning campaigns, predicting ROI, and playing “what if” with different marketing mix combinations to understand possible outcomes.?

As with all MMM efforts, the biggest challenge is compensating for missing data and factoring in external market factors.?

Logistic Regression

This technique can be used:

• As a “go/no go” (binary) tool used to predict the possibility of a specific event, such as a customer making a purchase—or not, whether a campaign will be successful, whether a customer will continue to subscribe, or whether a customer will purchase after visiting a website.
• As a way of classifying customers into groups3, according to their characteristics or behaviors
• Input can include customer satisfaction scores, frequency of usage, and number and type of customer support activity.
• To assess how likely a certain type of customer or certain types of customer behaviors lead to a purchase.?

The larger the sample size, the more you can depend on the results.?

Basic steps to implementing MMM4

1. Set your goals.
2. Collect the data, being careful to make sure it is as accurate as possible.
3. Clean the data.
4. Select the best models to achieve your goals.
5. Select the best variables to feed the models.
6. Run the models.
7. Analyze and tweak the models if needed.
8. Make recommendations based on the output.
9. Implement the changes.
10. Monitor and continue to optimize the process.

Conclusion

Obviously, these methods are rigorous and require deep thinking and proper use of the tools. If there is too much “guessing” on the input side, the output will lead management to make decisions that are too far from reality to be successful.?

Marketing can be data-driven, but you need to hire someone—in-house or as a contractor—to put these tools and methods to work for your company. Or, rely on the companies that have combined these tools on platforms that can be used by marketers without a degree in statistics.?

1https://en.wikipedia.org/wiki/Neil_H._Borden

2https://en.wikipedia.org/wiki/Marketing_mix

3https://www.techtarget.com/searchbusinessanalytics/definition/logistic-regression

4https://www.datacamp.com/tutorial/decoding-marketing-mix-modeling-a-complete-guide