Jeremy says: “Analytics like MMMs, or ‘Media Mix Models’, should be the core of your marketing strategy and budget, to understand truly what is working and what isn't.”

I thought we're talking about M&Ms there for a second. So, what is a Media Mix Model?

“I would say a Media Mix Mode (if you're somebody that works in marketing and branding and advertising) should be as sweet as an M&M.

MMM stands for Media Mix Model, or Marketing Mix Model. You might have heard it referred to as ‘econometrics’ beforehand, as well. That's also another long word that people find incredibly challenging to say.

It's a glorified statistical model, without going into the depths of mathematics and statistics. It allows you to measure the impact that each channel has on your overall marketing performance.

If you're trying to get more sales or more leads through the door – if you've got 100 leads this week, how many of them have come from search? How many of them come from social? How many of them come from offline media?

However, it also allows you to look at how many sales you're getting without marketing, because not everything that people buy or every lead that comes through the door is because we've done marketing. Sometimes, people buy things because they buy things.”

What does a typical Media Mix Model look like? Is there a typical model and is it in any way similar to an attribution model?

“Oh, the attribution versus econometrics debate, or attribution versus MMM debate, is a really interesting one.

Is there a typical Media Mix Model? I think there's a typical structure of a Media Mix Model, but it's going to vary from one brand to the next. It's also going to vary on the type of product you're selling.

If we go back to that example of ‘people buy things because they buy things’, Imagine you are one of the biggest bread retailers in your country. We won't advertise any of them here, but you don't walk into the supermarket shelves and say, ‘I'm going to pick up this brand of bread because I've just seen the ad for it outside’. You pick it up because you need the bread. So, actually, a Media Mix Model for a bread brand will probably say that most of your sales don't come from marketing.

Whereas, if you're selling something that maybe is a more considered purchase, where it's a really highly competitive market – say a car, for example – advertising is quite key in industries like that. It can make the difference from one sale to the next, but the structure of that model is going to be very different. It will be more marketing driven than non-marketing driven.

But also, the lag time from lead to sale will be very different to bread. You don't think about it; you just go and pick it up. Whereas, with a car, you look at one brand, you look at another brand, and then you make your decision three or four weeks later.

The structure is the same – it's the marketing plus non-marketing impact – but the models will slightly differ in that way.”

I've certainly seen lots of bread adverts on TV. I’m old enough to remember the old Hovis adverts. Are you saying that marketing for bread doesn't work?

“Marketing for bread does work, but a lot of people don't really think about it when they're in the bread aisle in the supermarket, do they?

If you usually buy Hovis, maybe a small proportion of people will think, ‘Oh, I just saw the ad for Warburtons, we'll try that instead.’ You just buy what you buy every single week. It's a bit like toothpaste as well. There are loads of toothpaste ads, but actually, once you've stuck to one brand, you don't really go and change. Whereas, with other things…

We built MMMs recently for software brands. That's quite a competitive market. If your software is no good, then somebody is going to go and buy your competitor next time. We've built it for leisure brands. If you think about things like gyms, there's a massively competitive market for gyms and fitness, and if you're not getting the service you want from one brand, again, you can quite easily go to another brand.

This is where MMMs become really useful. There are so many other questions that you can dive deeper into once you've got a fully structured MMM built too.”

Where does one begin putting together a Media Mix Model?

“The best place to start is data. It's always the best place to start. This is also where a lot of people get things wrong. It's like, ‘Yes, we've got data. Let's analyse some data,’ and actually it's the wrong approach. It should be: well, what's the question?

You start with some data and you start with: What's our key performance indicator? Normally, it's leads or sales – or if you're a charity, donations. How has that tracked over the past two to three years? The reason you look at two to three years of history is you want to understand what seasonality looks like, particularly for charities. There's going to be a massive seasonal trend around Christmas, when everyone gives more money to charity.

Bread, maybe not so much seasonality, but every brand has a pattern, and you need to understand what the pattern of that KPI looks like over time. Then, you want to understand how you've been advertising over the same time period.

What does your advertising data look like? How much have you spent by channel by week, over the course of that period? Then, to build the non-marketing side, you'll look at economic factors. Google Trends is really useful for this. What does Google Trends tell us about this market, and how people are searching for it?

Beyond that, what about things like the Consumer Price Index (so, inflation)? How's the market changing? How does this type of product respond to changes in the cost of living or changes in the market? Also, competitor advertising. If your competitors advertise, is that better for you? Is that worse for you?

In the case of something like cars, it could actually be better because, if people are test driving other cars and realising that they're not quite as good as yours, they might come back and buy from you.

It's data, data, data. I would say 80% of a Media Mix Modelling project is getting the right data. Once you've got that, the model is actually quite easy to build.”

In terms of building it, is there any particular software that you favour at the moment?

“There are a few different ways of doing it. There is what we will call ‘point-and-click technology’. It's a bit like an off-the-shelf product. You go and buy a product, you can put in all of your data and information, and it builds the model for you.

That's what some people might call a black box. You don't really know what's happening under the hood. You put some data in; you get something out. Is the model good? Who knows?

We build all of our Media Mix Models from scratch, and we do that in something called R or Python. Probably people here have heard of Python. R is just another coding language that you might say is similar to Python, but it's slightly more niche. Maybe fewer marketers use it. As a marketer, I've used R throughout all of my marketing life.

The advantage of doing that is you can actually write the model equation or formula or – we won't go into regression and statistics too deeply, but you can write that yourself and you have more control over it. If you understand how these models work, you can understand how to move parameters within that model to make it the best version of understanding the data.”

We mentioned the word attribution, and you alluded to the fact that your Media Mix Model and attribution were perhaps comparable but distinctly different.

What are the key differences and why would you want one over the other?

“I think attribution has had a long history of being used to measure the effectiveness of different marketing channels. The issue with attribution is that it's becoming harder and harder, because if you think about, say, Google versus Facebook versus X versus other platforms, everybody's closing the walls. So, it becomes harder to share data.

If you're looking at that full customer journey, somebody's first seen an ad on Facebook, then they've seen something on X, and then they convert by searching for you on Google. You don't have that full journey because Google, and Facebook, and X aren’t talking to each other. You've then got the added complexity that, if you're advertising offline as well as online, if you add something like radio or TV into the mix, and somebody's heard a radio ad or seen a TV ad, how does that interact with Google and Facebook? You can't measure that.

This is where a Media Mix Model is better because it looks at the entire picture. Rather than, ‘We've got a hundred sales. Let's go through them one-by-one. This one's from Google. This one's from Facebook. This one's from TV.’ It does it the other way around. It says, ‘We've got a hundred. What proportion are from Google? What proportion are from Facebook? What proportion are from TV?’

That's becoming (I would say, has become, over the last couple of years) the new gold standard for measurement. As attribution becomes harder, people are looking for other ways to robustly measure the impact of marketing contribution. We're seeing an uptick in the number of MMMs that we're building and the number of questions that we're getting: ‘We want to build an MMM. Where do we start?’”

I guess, from what you're saying, a Media Mix Model could perhaps enable being a little bit more predictive in terms of what you should be doing in your future marketing strategy as well.

How do you use analytics to assist with determining what your marketing strategy should be?

“This is where MMMs are really powerful, when you look at the future marketing strategy. If you go back to the way I described it at the very beginning, all it is is an equation. It's an equation that says ‘this amount of your Google plus this amount of your Facebook plus this amount of your TV, etc.’ For every factor that you put into your model, it will give you the number of sales that you get this week.

Because you've got that equation, what you can do is you can build it into a tool – a dashboard. We'll call it a spreadsheet. I know marketers love a spreadsheet, so let's keep it really simple. You can put it into a spreadsheet, and in that spreadsheet, you can say, ‘If I spend 10K on this channel and 10K on this channel and 20K on this channel, what result do I expect to see?’ and you can mock up different scenarios. ‘Okay, if I've got my 40K to spend, if I spent it 5, 5, 30 instead of 10, 10, 20, how would that differ in terms of the number of leads or sales coming through the door?’

It becomes quite powerful because you can use it to plan the most optimal strategy. Hence, it's called a Media Mix Model. What is the best media mix across all of the channels to get the result that we want?

Let's say you are just running search, for example. If you've just got Google Ads or Google search or something on a search engine, if you haven't got anything on another channel (like Facebook or a TV ad or potentially some YouTube ads), what's feeding that search? Why are people going to search for you on Google or any other search engine?

That's where the Media Mix Model is telling you, ‘Okay, search when YouTube is on does this well, but when YouTube and Facebook are on, it does this well.’”

I think that SEOs naturally want to embrace analytics to see how what they're doing has performed/is performing, and where opportunities may lie for organic rankings in the future – but, obviously, other marketing departments aren't necessarily so deep into analytics.

How do you encourage other marketing departments to embrace analytics and MMMs, and become more data-driven in their approach?

“I think it's a challenge. We certainly see it a lot where people say, ‘I've got 20 years of experience in this industry, and because of that, I know that the answer is X.’ Actually, sometimes, you've got to break away from the personality and the experience and start to think about doing things in a different way.

One of the things we always encourage people to do is be brave. Be brave and think differently because data is actually your most crucial asset. Data tells you what's happened in the past, and if it's happened in the past, it's probably more than likely to happen again.

If you think about marketing, in any form, marketing is about attracting people. Whilst people might be unpredictable, we are creatures of habit, so we all behave in similar ways. If we behaved in a certain way in the past, we're likely to behave in that way again. By looking at the data and analysing how people are behaving, you maximise your potential chance of success.

This is why models like this are really, really powerful and useful. If you just say, ‘Well, we've done it the same way for 20 years. Let's keep doing the same thing,’ the most likely thing is your competitor will build the MMM. Your competitor will understand the data, and they will be the winners because they're using data to inform their future decision-making. Whereas, you're staying in the past and not.

Talking about doing things a little bit differently, I guess one of the challenges with being truly data driven is that the past doesn't necessarily always represent what is going to be the most effective thing to do in the future. You have different marketing channels that are up and coming and may not necessarily have driven any traffic in the past.

How do you ensure that you're not missing out on those opportunities?

“Regular updates. The whole point of the model is it's dynamic. You don't just build a Media Mix Model and say, ‘Right, that's done. This is how we're going to plan media from now on.’ You update the model regularly. You update it every time you change your marketing strategy.

Or, if you're not changing your marketing strategy too often, I'd probably say update it a few times a year. Anything from two to four times per year, when you've got a relatively stable strategy, just to account for changes in the economy, changes in the way that people are behaving, etc.

If your marketing strategy is changing all the time, that regular update allows you to observe the change in people's behaviour as a result of your strategy. Let's say, for example, you wake up tomorrow morning and say, ‘I'm going to do something crazy. I'm going to put a zero at the end of my YouTube budget and I'm going to spend whatever I was spending beforehand with a zero at the end of it.’ If you wait six months to update that model, you don't know if it's working.

Do it quickly. Make the change, update the model a few months later, and see if it's worked. If it does work, keep doing it. If not, revert back or go somewhere in the middle of previous strategy to current strategy.”

Jeremy, what's the key takeaway from the tip you shared today?

“The key takeaway is that, if you want to grow as a business and if you want to save money on your marketing, use an MMM because you will save time and you will save money.”

Jeremy Horne is Director and Founder at Datacove. Find out more over at Datacove.co.uk.