Marketing Mix Modeling
Last updated
Last updated
Marketing mix modeling, often referred to as MMM, is an advanced form of marketing measurement that uses machine learning statistical algorithms to estimate the relationship between marketing activities (e.g. LinkedIn ads, events) on outcomes (e.g. leads, demos, sales).
The goal?
To find out which marketing activities influence or change consumer behavior (incrementality) and which don’t, and by how much.
Marketing mix modeling works in four steps:
Target goal
Data collection
Regression analysis
Interpretation
First, we select a target outcome (metric) we want to measure marketing’s influence over.
Depending on your business model, sales cycle, and data, that could be closed/won new business, sales qualified opportunities, or leads.
For example, since MMM requires volume data to estimate incrementality with confidence, if you drive 2,500 opportunities in a given time period compared to only 100 closed/won new customers, opportunities would make for a better target metric.
Or say you have a long sales cycle (12 months).
Since there’s a significant time lag between marketing activity and closed/won, it’s best to choose an earlier stage target metric like leads, not closed/won new sales. Too much time lag between activity and outcome introduces changing dynamics that the model won’t be able to measure accurately.
Once we choose our target outcome, we collect and organize all of your historical spend and performance data across all of your marketing channels: online and offline, paid and organic, brand and performance.
What does historical mean? At least two years of organized data.
In addition to spend and performance data, we collect any other data that may bias outcomes.
For example, say you sell video conferencing software and part of your historical data includes heightened demand from the Covid era. If we didn’t collect that data and incorporate it into the model, your results would be heavily skewed.
With a target outcome set and data collected, now we can start modeling the data.
Marketing mix modeling uses a type of statistical analysis called regression analysis.
A regression analysis analyzes historical data to estimate the relationship between marketing inputs and business outcomes. More specifically, it helps us understand how changes in one variable might affect another and if there’s a clear connection between a marketing activity and an outcome.
For example, how do changes in radio, events, and LinkedIn ads spend and performance (independent variables) influence booked demos (dependent variable)?
Only a regression analysis can pick apart each of those three independent variables and estimate their respective influence on booked demos.
Last, we interpret, tune, and action the data.
Since no marketing mix model is perfect on its own, it’s important to feed the model with more definitive data to ensure accuracy and assess performance.
For example, say the model tells you that certain types of campaigns on Linkedin aren’t incremental. To validate those results, you can run an incrementality experiment on said campaigns, then input those results into the model to train it with causal data.
Over time, the model gets more accurate, making forecasting, budget optimization, and scenario planning more precise.
So what are the benefits of marketing mix modeling?
Marketing mix modeling uses aggregate data at the channel-level, not user data at the person-level. Which means it’s not subject to privacy laws or regulations. It’s literally future-proof.
MMM can evaluate nearly all channels- online and offline- giving it more coverage than any other measurement method.
Brand and performance like paid ads vs. cTV
Paid and organic like Linkedin vs. email
Online and offline like events vs. digital media
MMM doesn’t just measure the past so you can react; it helps you predict and action the future.
For example, MMM can do the following:
Scenario planning: MMM simulates different results at different budgets automatically so you can better forecast the future.
Budget optimization: MMM makes recommendations for budget allocation across all your channels
Saturation points: MMM can tell you if you’re about to hit a point of diminishing returns with different ad channels so you can cut spend before you start wasting it.
Incrementality refers to the additional impact a specific marketing activity has (or doesn’t have) on a goal beyond what would have happened naturally.
For example, let’s say a control group who hasn’t been exposed to your ad buys 100 times in a month, and a treatment group who was exposed to your ad buys 150 times in the same month.
The ad in this example was directly responsible for 50 incremental sales that you wouldn’t have gotten without it.
Marketing mix modeling estimated incrementality across your entire marketing mix, which means you can start optimizing spend for activities you know influence behavior the most while cutting spend to activities that waste money.
Since the marketing you do today accumulates over time and influences decisions months or years later when buyers move in-market, marketing mix modeling is able to account for that delay and give long term marketing the credit it deserves.
What about the limitations?
First, marketing mix modeling requires volume data to work accurately.
If you don’t have historical data (2 years) and large media spend (min. $3M/annually), the model will struggle to produce confident results.
Second, since MMM works with aggregate data, it struggles to zoom in on campaigns, creative, and content.
It can tell you which channels perform best and by how much (and sometimes campaigns if there’s enough data), But you’ll need something like lift analysis or MTA to help you figure out what specifically in that channel isn’t working.
Right at the top.
Marketing mix modeling is the most sophisticated and most difficult measurement method in the stack. But it’s also the most rewarding with the highest ROI.
That’s because MMM is built for diverse marketing mixes and large spending, and when you’re doing that much marketing, it gets really hard to tell which programs drive how much influence.
For example, as a startup, everything is incremental. That’s because nobody knows you exist and you can only afford 1-2 channels. Therefore, everything “works.”
As for larger brands? Not so much.
Much of a larger brand's spend has zero impact on sales, they just don’t know how much.
Why? Because most of the market knows you exist already and you’re spending millions across dozens of channels to reach them. Inevitably, not all of those activities are going to influence their behavior, even if they engage with them.
Without regression analysis, it’s impossible to identify incrementality across such a diverse mix.
Hence, the high return on investment.
A good marketing mix model is going to recoup many times more the revenue in wasted spend and opportunity cost than what it cost you to actually conduct.
Not everyone, obviously.
Given the data requirements of MMM, it’s best for mid-market and enterprise brands with healthy historical data and a sizable annual spend.
And if you have a really long sales cycle, like, say, 12-months, then MMM will struggle to capture those delayed effects.
Think of marketing mix modeling as the last stage of sophistication when it comes to measurement: until your budget and marketing mix grows with time and size, MMM may not be for you.
However, nothing helps a marketing mix model work effectively like organized historical data.
So if you don’t have the budget or channel diversity to meet the data requirements yet, it’s never too late to start collecting and organizing your data. When the time comes, you can train your model on clean data and achieve time to value faster.
Want to explore marketing mix modeling in more detail but don’t know where to start?
Schedule a demo (or reach out to your CS manager at HockeyStack) and we’ll walk you through our process, whether or not you’re a good fit, and how to get started.