Digital Data in Marketing Mix Modeling

There are many businesses where majority of the sales are driven offline, but significant marketing dollars may be spent on online marketing. The key marketing questions that arises that many people struggles to answer are:

• How much should you spend and allocate your marketing dollars to what ad channels?
• What does a success looks like (or ROI) if you spend X amount to which Z ad?
• How much offline ads or online ads are contributing to the bottom line?

These questions are important even if you’re in an online/offline only business. The reason I’ve pointed out the business model where revenue share is split between offline and online channels is because it quite hard to answer these three questions depending on your data environment or availability.

Consumers touch points with your ads or brand could vary and come from many different sources. People can learn about your product through social media, research on manufacturer site, go to retail store, then shop online at a discount. Consumers could see the ad and go to the retail store directly and pick up the product, too. There is NO one single path.

This is very complex if you are trying to tackle trying to measure the effective of media mix within this eco system.

I’m actually working on a similar effort with a vendor specializing in marketing mix modeling, which is suppose to answer these three critical questions.

Before moving forward, here is a brief description of marketing mix modeling…via Wikipedia

“Marketing mix modeling is a term of art for the use of statistical analysis such as multivariate regressions on sales and marketing time series data to estimate the impact of various marketing tactics on sales and then forecast the impact of future sets of tactics. It is often used to optimize advertising mix and promotional tactics with respect to sales revenue or profit. The techniques were developed by econometricians and were first applied to consumer packaged goods, since manufacturers of those goods had access to good data on sales and marketing support.”

So what I’ve learned is that in modern erra of marketing mix modeling, there are more application of digital data than years ago when companies were using only TV ads, sales, spends, econometrics models, etc. Some of the digital data that would go into modeling are:
- Unique visitors
- Traffic to a key section of the site. i.e. product page traffic, cart page
- Organic search traffic like branded terms
- Product category traffic or search trend around those product related terms
- Digital Ad traffic from Banner Ads, Emails, Paid Search, etc.
- Social Media brand mentions online (could be positive sentiments)

These are examples only, and what goes into the model should be consulted and worked out to make sure the stake holders are in alignment. In most of the cases it has many dependencies on the experience of the partner you’re working with.

What I’ve learned from the data preparation for marketing mix modeling is that, it requires data that spans across multiple years, and better to have granular data in daily or weekly basis. Part of the reason is because in the modeling:
- it needs enough data to understand seasonality
- it needs enough data and trend to recognize correlation for small data that may have an impact on sales
- some macro trends may take only on rare occasions so having that data as one part on a timeline will allow it to be picked up as a signal
- build a good What-If scenario analysis model

What this means is that it is important to have your data methodology clean and consistent. As we digital folks know, digital data could be short lived for whatever reason. Switching vendors, site migrations where data are scattered in different databases or archived, change in tracking methodology, changes in conversion event, emergence of new data like Social Media, etc. If you’re not doing that now, I’d recommend you start thinking about it now and document everything you’re tracking.

One challenge that I had was getting social media data. Availability of social measures are pretty recent (I bet that applies to many companies), and many social measures are representing growth over time especially if companies put a lot of efforts into it in recent years. So it may or may not provide added value into the modeling’s output. So it requires some digging into narrowing down the data that shows relevant trends that gives you a good signal that may impact the bottom line. For example, instead of just tweets or mentions online, narrow down to branded terms or category terms.

There is this one study that illustrates this study pretty well. Study done by PC City, Google, and MarketShare partners. Primary goal for them was to understand the driver of their offline sales by marketing media mix. I thought this would be great example and a study many companies may consider performing.

Not sure how much effort went into social, or other digital data, but data models would be different by companies and industries. It seems like marketing mix modeling is more of an art than science because a lot of the learnings and recommendations vary by vendors, so I would recommend choosing the right partner with experience and know-how in your business.

Here is a link to the PDF for this study: 2010 PC City Marketing Mix ROI