328. MMM 😋
Cliff Notes on Marketing Mix Modeling, the data foundation used by big-time marketing teams
Hi there 👋
Welcome to all the new subscribers! If this is your first newsletter, you can scroll down for the actual marketing stuff. Before we get there, I was hoping to ask for some advice from the community: What are everyone’s thoughts on mechanical keyboards (like this one, pictured below)? I’m intrigued, but I worry that my typing will be incredibly loud and distracting. Maybe that’s the point. If you have a mechanical keyboard or don’t have one for particular reasons, I’d love to hear from you! Hit reply to chat with me.
Wishing you a great week ahead,
Say hi anytime at firstname.lastname@example.org. I’d love to hear from you.
Cliff Notes on Marketing Mix Modeling
I had a chance to chat with a couple very smart marketers the other day about this current season of startup life where valuations are down, efficiency is up, and marketing dollars are under scrutiny.
Their advice: the companies that thrive in a downturn are the companies that have measurement figured out.
Five years ago was the best time to double-down on measurement. Now is the second-best time.
Of course measurement can mean a lot of different things to a lot of different people. In my experience at Buffer and Oyster, it has meant business intelligence dashboards (Looker) with real-time product analytics (Mixpanel) supported by a team of data analysts and data scientists.
But now my eyes are turning toward a framework used by a lot of big-time, big-spending marketing orgs: Marketing Mix Modeling.
I thought I’d share my Cliff Notes with you as I’ve gone down the research rabbit hole. 🐇 🕳
What is Marketing Mix Modeling (MMM)?
Marketing mix modeling refers to analytical solutions that help marketers to understand and simulate the effect of advertising, and to optimize tactics and the delivery medium.
Marketing mix modeling can also help managers with P&L responsibility understand ROI by tactic. For example, if I have $1 to invest, I can get an X return from advertising and a Y return from a consumer promotion. This helps marketers make informed decisions on where to invest based on their measures of success.
MMM helps you in the following ways:
Predict the future (to the best of your ability) using some super advanced data techniques. And this ability to predict the future helps you make tradeoff decisions in the present about what you can expect from additional spend and investment across your channels.
See which channels are contributing high ROI today
Get better visibility into how all of your channels relate to your target outcome, whether that’s revenue or market share or anything in between
Can I see an example?
Sure thing. Here are some MMM screenshots from a very data-sciency project using Facebook data.
If this is a bit much, there’s also plenty of MMM that lives in spreadsheets and even a Udemy course on how to do MMM in Excel.
And here is a tutorial for building MMM using Python.
How is this different than multi-touch attribution?
Multi-touch attribution has long been a favorite way to assign value to all sorts of different channels. This methodology relies a lot on cookies to track user behavior from channel-to-channel, which you can then model and view on something like a Google Analytics dashboard.
(Multi-touch attribution has become less and less effective with all the cookie restrictions in place, like with iOS14+.)
Multi-touch attribution focuses on user behavior. MMM focuses on channel efficiency — beyond the world of cookies. MMM measures the impact of the typical, measurable channels plus the typically immeasurable ones like TV, radio, and newspapers.
What’s the science behind it?
Like any good data analysis, MMM begins with the inclusion of inputs and outputs
Inputs = either numerical inputs like budget, sessions, price, impressions, or categorical inputs with fixed inputs like day, week, month, promotion (these are generally a boolean of 1 or 0)
Output = What you are trying to predict. This can be revenue, market share, conversions, traffic, etc.
From there, the data science takes over.
A lot of this can be modeled in spreadsheets or in more robust software solutions (see below). The most robust of the bunch incorporates Artificial Intelligence and learning algorithms that get smarter the more data you feed them.
In addition to your standard linear and non-linear regression analysis, MMM can include advanced techniques like:
Carryover effect, which measures how effective a campaign will be if you keep running it again and again
Diminishing returns, which creates a concave shape showing the decrease in ROI as marketing activity increases
Hyperparametrization, which involves external values that control what the learning algorithm actually learns
MMM Decomposition, which breaks down sales (decomposes) into all of the contributing variables that drove sales.
Where can I get started?
If you’re a hands-on learner, then a resource like this tutorial from Cassandra will help you get your feet wet with MMM in a spreadsheet.
From there, it’s best to work with your data team on how to build out a proof-of-concept MMM program that you can tweak and refine together or to invest in some tools that can do the MMM job for you (or with you).
Before you begin, keep in mind that MMM is most effective if you’re running a large budget (which is one reason why we never considered it at Buffer). It’s easier to implement in B2C than in B2B, but I know plenty of B2B companies who have managed to do so - especially those with a B2C2B motion.
And most importantly, begin by understanding what it is you hope to measure and learn from your MMM investment. If there are clear outcomes and impact that you can’t see today, then MMM may be worth a closer look.
I’d like to learn more …
A Complete Guide to Building a Marketing Mix Model in 2021 by Terence Shin on Medium
Market Mix Modeling (MMM) - 101 by Ridhima Kumar on Medium
WTF is Marketing Mix Modeling? by Kristina Monilos at Digiday
Marketing mix modeling vs. attribution modeling: the prose and cons by Evan Kaeding and Pinja Virtanen at Supermetrics
Your content business model (part 12 of a 12-part brand and content workshop)
About this newsletter …
Hi, I’m Kevan, a marketing exec based in Boise, Idaho, who specializes in startup marketing and brand-building. I currently lead the marketing team at Oyster (we’re hiring!). I previously built brands at Buffer, Vox, and Polly. Each week, I share playbooks, case studies, stories, and links from inside the startup marketing world. Not yet subscribed? No worries. You can check out the archive, or sign up below:
Thank you for being here! 🙇♂️
I’m lucky to count folks from great brands like these (and many more) as part of this newsletter community.