What is B2B Media Mix Modeling?

While media mix models (MMMs) have been around for decades, they’re surprisingly not very prevalent in B2B marketing. And if you’re not familiar with what a media mix model is, don’t worry, we’ll get to the definition and how it can help B2B marketers. When I used MMM software previously, my company was one of the very few B2B users. And since joining Align BI, I’ve yet to run into a B2B prospect using MMMs.

Here’s the problem with that—MMMs are likely the most powerful tool for budget allocation, planning, and ROI calculating. And they’re a perfect tool for B2B marketers.  

In today’s post, I’ll share what I’ve learned about MMMs and a few tips for making them work for B2Bs. You should walk away with some actionable ideas on how MMMs can work for you. 

First, let’s get into the definition. What is a media mix model? 

I recently jumped into media mix modeling too quickly with a prospect, not realizing they didn’t fully understand what a MMM was (again confirming that MMMs haven’t found their way into B2B marketing yet). So let’s dive into a high-level definition of a MMM. For a deeper dive and even tips on how to build your own, here are a few of my favorite articles on what they are and how to build them:

There’s one concept you need to get out of your mind before we begin. In B2B marketing, everything we do lives in a funnel with attribution of each opportunity (oppty) to channels, assets, and campaigns. Leave that funnel behind now. There are NO rules or algorithmic attribution in a MMM. Think of attribution as micro and MMMs as macro.

A media mix model is a regression and optimization problem. With it, we are identifying the relationship between the spend in our marketing channels (independent or explanatory variables, for those that remember college statistics) with an outcome (dependent variable) in our funnel. But this relationship does not require any attribution. We want to see if general changes in marketing channel volumes have a relationship with changes to our desired outcome (dependent variable).

There can be nuances to this relationship that can be included in the model. For example, this relationship likely has a lag (an increase in spend in search for example might increase my leads that same week, but my opportunities might take a few weeks to see any change). Or there may be a diminishing returns curve. Marketing spend often has what is called an AdStock effect. This means that marketing will continue to influence your prospects for some period after the interaction. Sometimes there might not be a measurable relationship at all. Or worse, there could be a negative relationship – the more you spend in a channel the less of an outcome you’ll see. (Although it’s probably more likely that if you see this your channel is just highly correlated with another channel.) A MMM can identify and capture all of these relationships.

But remember, when looking at this change, we aren’t looking at an individual opportunity’s activity history. We are taking the totals in each channel for a given timeframe (day, week, month) and looking for a relationship in our dependent variable (funnel stage). This is fundamentally different from attribution. 

Think of it this way: attribution is trying to assign credit for each individual opportunity to a single or multiple marketing activities. It’s at the micro level. MMMs are looking for a relationship between marketing channels and opportunities at a macro level. “If I spend $1M in search next week, how much, and when, do I expect results.” 

You might be saying to yourself, “well can’t I do that with attribution if I know my cost per and time between stages?” Yes. But MMM still has advantages here:

  • A MMM can identify a diminishing returns curve more easily
  • A MMM can identify lag and adstock carryover effects in the relationship (as mentioned previously)
  • A MMM avoids rules-based logic and instead identifies a statistical relationship (think a more accurate contribution number)
  • A MMM can look at how other factors like the economy, competitors, discounts, etc. impact sales
  • A MMM can include marketing channels that don’t create clicks and responses on a contact history (think brand, PR, print, radio, etc.)

But the biggest advantage to MMMs? You can find an optimal spend mix! 

Once you’ve landed on a final model you feed the model equation into an optimization model that will identify the ideal mix, based on ROI curves, to maximize your dependent variable. Armed with this data you can go to your business and sales executives with a budget request, optimized to support their goals. Or, based on a budget they give you, you can provide an optimized channel mix strategy and a contribution number you’re confident you can actually hit. (With attribution, you know what’s received credit, but often don’t know if that’s a product of an effective channel or just channel effort / volume. Thus it can be harder to optimize.)

I’ve never felt more confident in front of sales and finance than I did when we finally had a media mix model we all believed in. And I know you can experience that same feeling too.

Now that we know what an MMM is and why it’s helpful, let’s get to the B2B tips. I’ve built MMMs using software (ScanmarQED) and from scratch in R / Python. Here are my 5 recommendations when building a media mix model for B2B marketing:

1. Dependent variable: use pipeline / opportunities

Don’t use closed-won business as your dependent variable. In B2C, the purchase is typically much closer in time to marketing activities. But in B2B, this process can take more than a year.

Instead, build closer to what marketing effects and avoid hockey stick trends by using opportunity creation as your dependent variable. I pick oppty creation instead of qualified leads because nowadays, executives don’t really care how many leads you bring in. They want to know how you’re helping create the pipeline. Honestly, I think I’d prefer to build these against qualified leads so that the relationship was even tighter. But opptys work and have you talking the same language as sales.

  • Sales organizations tend to have “hockey stick” closed deal trends, meaning they tend to close everything the last few weeks of each quarter. It’s really hard to find a relationship between marketing efforts and end of quarter hockey stick close trends.
  • To estimate ROI from opptys, you just need an average close rate and deal size to estimate what those opptys will be worth.

2. What explanatory metrics to use?

Go as high as you can as those metrics will tend to have the least “cost per” variability. You can use different metrics inside the same model: impressions, click, responses, articles, etc.

  • There is an argument for going with ease here too. B2B will have a lot of response data in the CRM. While clicks or impressions are typically better as they are closer to the spend and thus have less “cost per” variance, there’s something to be said for just pulling readily available data from your CRM. But if you can, go high.
  • Some channels will really only have responses, like content syndication and events. That’s fine.

3. Getting spend and results (especially for events) at the same level is often the hardest part.

Event data, in particular, is a larger piece of the spend pie in B2B. Getting event spend and event results at the same level can be hard. For example, the event spends recorded by finance aren’t always at the same level as the event categories in your CRM. To fix this I would either:

  • Require your event marketers to record their event costs on the CRM campaign where the results are tracked. This can be hard to enforce. Simple auditing of events with $0 spend should highlight your culprits though. Still, if you consider all of the tradeshows, strategic accounts, small events, lunches, etc., it can be a big list of events to be auditing / cleaning. If this is your business, then try the next tip.
  • Fuzzy match it all. Just do your best to roll event subcategories to event buckets from finance. It won’t be perfect, but better than nothing.
  • Other channels can also be hard to match. The recorded spend in finance rarely aligns with the in-market spend. And even certain channels like content syndication are typically a one time spend with leads that come in over time. For this you can either spread the spend evenly across the quarter, since spend is really only being used to calculate ROI and not as a dependent variable, or you can spread it in line with the channel’s metric. 
  • This can get a little annoying when each channel is different. So automating as much of this with Python or within your ETL tool is ideal.

4. You have a sales team in the process!

Pay attention to the relationship with the sales org.  If your inside sales team is a bottleneck, then more marketing spend won’t actually create more opptys, or your relationship will be loose. And are you certain that more opptys really equal more won when your AEs have quota, and might not want to exceed that quota? There are two ways to consider the effect of your sales team:

  • Add them to your model! We can enter variables like inside sales emails and phone calls, sales rep quota, etc., to find their impact on opptys.
  • Simulate what more opptys means for your closed won stage. If you can identify the close rate probability distribution is and how it changes with volume, you can simulate what more opptys really means for your bookings.

5. B2B marketing teams can have different goals and objectives from each other.

In B2C, companies without a sales team are typically trying to drive to a purchase. But in B2B, some teams / channels are tasked with creating pipeline and some are tasked with helping to close that pipeline (creation vs acceleration, and some are even tasked with post sale adoption and customer success).

  • I recommend putting everything in the model as most channels serve both objectives. It’s the content that is more likely to change.
  • Leave out specific campaigns / activities that are extremely targeted to closing activity. But how should you measure their effectiveness then? Probably attribution and looking at likelihood to close when those activities are on the history. I’m not sure a MMM for acceleration would work, as you then run into the hockey stick close and long close times.

A media mix model is amazing. It gives you marketing contribution, channel ROI curves, and spend allocation recommendations. Pay attention to the nuances of B2B listed above and you’ll be on your way to having some of the best conversations with your sales and business executive stakeholders that you’ve ever had. They’ll trust you and your recommendations more and you’ll be more confident building your marketing plan.

Imagine having a flat YoY budget but telling your stakeholders that you can support more pipeline because you’re MMM has found higher ROI returns with an improved channel mix. Or standing firm in your commitment to support only a certain level of pipe because you’ve got the data to show what’s possible based on historical results and optimized channels. It becomes a strong and really fun position to be in.

About Align BI

We recently built a media mix model at Align BI using Python and Tableau. Why Python and Tableau? Well, the hardest part of the entire process is often collecting and organizing the data. Python allows us to automate those steps. And Tableau, or any other BI visualization tool, allows us to push the results to the repository where the rest of your marketing dashboards live. We’re happy to help. And if you don’t mind organizing your data we’d be happy to throw your data into our model to see if we can find a signal and optimization opportunities. If there are, then we can work together to automate the collection of it and a cadence to run the optimizations.

Ready to give media mix models a try? Let’s talk. Contact us to learn more. 

© Align BI 2024 | Crafted by Reborn Consultants