In our last video, we learned that Imra at Calla and Ivy is looking to better understand where her marketing dollars are going and if her efforts are truly making an impact on her sales. She has been able to study the impact of individual market efforts, but she wants a more comprehensive look at how all of her marketing efforts are impacting her sales. In order to do that, she's going to use a marketing mixed model. Let's look at what marketing mix modeling is, how it works and how this can help Imra in this video. Marketing mix modeling, sometimes also referred to as media mix modeling or just MMM is a data driven statistical analysis that quantifies the incremental sales impact in ROI of marketing activities. It's an established measurement solution for holistic cross channel sales measurement that can take both offline and online data into account. MMM uses historical data to quantify the impact of a large set of variables on sales. These variables can be marketing or non marketing related. Like for instance the opening of a new store or the season during which you are evaluating data. The result is a model that helps marketers understand what influenced past sales and predict what might happen as a result of future marketing action. Let's look at the different steps involved in MMM. In most cases, marketing analysts will work with a research provider who does the actual modeling. But you need to understand the steps, so you can fully support this process. First, you start by identifying all the different variables that could impact sales, again this can be marketing and non marketing variables. As I'm sure you can imagine there can be a very large number of variables. This is an example of what a variable landscape could look like for an airline, the red variables directly related to marketing campaigns. You see things like TV ads, social media ads, PPC display and so on. The blue variables are other variables that may affect sales like seasonality, loyalty program membership and so forth. Once you have identified all the variables, data collection begins a good time to think of the osemn model again. Which you will remember if you completed the second course in this program. You'll have to identify how you can obtain the data for each one of these variables and you'll have to scrub them, make sure that they're clean. This may already be a large amount of data for any given month, but for a marketing mixed model to really understand the environment of your business. The general rule is to have at least two years worth of data. With two years of data, you know that you're not only capturing a solid sample size. But that you're accurately capturing large factors like seasonality and market conditions that may impact your sales as well. Even if you work with a vendor, you'll have to assist in this process. Later we'll also see how platforms like Facebook for instance may provide you with direct access to the data you need via an API. Then it's time for modeling, MMM uses a form of multi variant linear regression analysis to quantify the impact of all the different variables on the outcome or your key performance indicator. All the variables that may impact your KPI are considered independent variables. A marketing mixed model is a long mathematical equation which shows the statistical relationship between all the independent variables and your outcome variable. At high level The equation looks like this, KPIt stands for the key performance indicator at time T you want to model. Beta refers to the coefficients or what changed in the variable means for the KPI similar to the regression coefficients we talked about earlier. Beta zero refers to the base performance or what performance would look like if all the other variables were zero. So no media, no marketing, no promotions, no seasonality and so on. the goal for the modeler is to come up with the model that best fits your data and explains as much as possible of the outcome variable with the combination of independent variables. So the model should be giving us the closest possible prediction of the KPI. Based on the modeled combination of the independent variables while keeping the error in our model as small as possible. The closer the model predicts the KPI the smaller the error. Marketing mix models are never perfect, but they're very useful and the smaller the errors the more useful your model will be. As we said, MMM uses multi variant linear regression, but it's a rather complex form of multivariate regression. As you can imagine, it has to take into account a very large number of variables as well as the possible correlation between these variables as they don't all move independently. No need for you to learn those models as that is where you will get to help from data scientists and vendors. What you'll find when working on marketing mix modeling is that there's quite a bit of interaction that will happen between you and the modeling team you work with. You'll decide together on the variables and you will discuss the best ways in which you can get the data. Then the modeling team will come up with the actual statistical model that involves calculating the beta coefficients. Even in that modeling step, there will be some back and forth between the teams to come up with the best fit for your data. Once you have the model you can use it to predict what your KPI value would be based on the models coefficients. So you can evaluate what would happen if you distributed your spending differently across your marketing campaigns or your marketing variables in this model. Let's take a look at how this may work for Imra at Calla and Ivy. Remember Imra is looking to see which marketing efforts had an impact on her sales. And more specifically which marketing efforts in combination had an impact on her sales. The goal is to gain more transparency into what's working and what isn't. As a first step, Imra needs to decide which variables may have an influence on her sales. While she's interested in understanding how her marketing action influence sales, she should still take into account other non marketing related variables as they may have had an influence too. Let's first look at the marketing variables. Since Calla and Ivy is a small flower shop, we wouldn't have as many marketing efforts to include as a large company would. But Imra has tried a number of different marketing tactics over the past two years and so we want to include all of them. Every Facebook ad campaign she ran, every google ad campaign, she ran every print ad she took out. Every email newsletter she sent every organic engagement on social media, her in-shop promotions and each paid search ad. When thinking about these variables, Imra needs to further specify what about these campaigns will have influenced sales. She may add how much she spent in each of these channels, but she should also add things like the number of impressions in a campaign for instance. Then Imra thinks about the non marketing variables she needs to include. She decides to include the launch of her website, the price changes she made over the years, the opening of a new store in Amsterdam, the addition of the Belgian market for online sales. Now, Imra needs to collect data for all these variables. This may seem like a lot of data and it is, one thing to keep in mind as you anticipate using marketing mix modeling in the future is data collection. Are you or those you work with actively collecting data on your marketing efforts? Models are only as good as the data you put in, so be sure to make data collection of priority, Imra won't do the analysis herself. She hired the vendor to help and their team of statisticians is helping her define variables and the exact data she needs to collect and they help her clean the data too. Then they're ready to start the modeling work. We won't get into the details of how the exact analysis works for Imra, essentially the modelers run a very complex regression on all the data Imra and team provided. After several iterations, the modelers come back to Imra with a model that explains how much influence all the variables she defined had on her sales. In a very, very simple form, this model could look something like this. If there were only three variables, the model tells me that if Imra increases prices with $1, her sales units go down by 5000. If Google search ad impressions increase with a million, sales units go up by 500. And an increase of Facebook ad impressions by a million, increases the sales units by 4000. Of course Imras model looks more complex as she had many more variables she was assessing. But now that she has this model, she can use it to start planning how to best allocate her budget. As you can see, having a marketing mixed model can be very helpful for small and large businesses alike. But as we'll see in the next video, there are some challenges when it comes to running this type of analysis.
