mgamzec / ds-of-marketing

Data Science in Marketing

1. EDA- Performance Analysis

What kind of questions would you ask? 

A highly effective concept for marketing campaigns of this sort is to look at leading and lagging indicators. So you want to look at paid search impressions. This is a great leading indicator that will help us to know how many users are finding our brand and our promotions via certain keywords. 

Next, we need to assess CTR = click-through rate, which will help us to understand how effectively our ads are at addressing what customers are really looking for. 

And we also need to assess CPA, which stands for cost per acquisition, so that we can truly understand our return on investment and our conversion. So we just talked through a number of our TMTMTMs, or what I call, our metrics that matter the most. And again, leading and lagging indicators that help our marketing campaigns to ladder up to business impact. You can think of the customer journey here almost like one of those board games you played as a kid. Each roll of the dice, each move you make should move you forward along the path, should move your customer and your market forward along that path. If we move our CTR rate forward, chances are we've made some informed decisions now here that will also have a positive impact on our cost per acquisition, so on and so forth. So in the following videos we are first going to assess what you call the shape of our data. This is to say, we're going to visualize what the data looks like so that we can see if there are any patterns. Those patterns can help us to create a narrative and can help us to create an intuition for what deeper analysis and specific models might reveal. Then, in each video with each of our analysis platforms, we'll model a matrix so we can see relative performance at a high level. This is a data visualization that is indispensable to assessing marketing's impact.

2. Inference and Regression Analysis

Have you ever noticed how one thing can somehow influence another? - Linear Regression

In other words, two items or two events can be correlated. For example, ice cream sales go up as the temperature goes up. Remember however that correlation does not imply causation. For example, there's a strong positive correlation between per capita consumption of mozzarella cheese and the number of civil engineering doctorates awarded, but that doesn't mean that cheese consumption affects the affinity for postgrad academia in civil engineering. Regression analysis is a primary statistical technique in understanding the relationship between things. In our case, those things are marketing output and business outcomes. 

In linear regression analysis, we assess at least two data points, an independent variable and a dependent variable. Here's a quick definition for each type. A dependent variable is a thing that may be influenced by some other thing. A rainy day often means umbrellas are being used. In this case, the use of the umbrella is the dependent variable. Its use depends on it raining. An independent variable, on the other hand, is a thing that might influence some other thing. It does the influencing. So in the previous analogy with the rainy day, the rain is the independent variable. Think of it this way, it's going to rain whether folks choose to use an umbrella or not. It's independent. In the marketing world, there's a similar dynamic at play. The more people that experience your marketing, see a television commercial, for example, the more interest and the more engagement there is with your brand. So in this case, the amount of marketing is the independent variable and consumers' attention is the dependent variable. Now, let's get to work. Imagine that I'm walking through a casino in Vegas with one of my clients. We're on our way to the MAGIC show, which is a biannual event where all the apparel brands go to show their fashions for the upcoming season. So you've got brands from Nike to Quiksilver and they're all there, making deals with their current customers and creating new ones. This is how most clothing makes its way onto the clothing rack in your favorite store. My buddy stops at the roulette table, drops $100 on red, and it's the equivalent of a coin toss. It's 50 50 odds. The wheel spins, it lands on red, we have a winner. Luck is on our side. And this win has my client feeling bullish and as we continue to make our way to the exhibit hall, we discuss plans for a Times Square takeover campaign. This is a big investment, but it's a competitive business, and the winners have to make big bets to build consumer perception that drives demand. It's our job to make sure those are safe bets as well. Our client in this branded lifestyle space has made similar investments in the past and we have some really good data on hand to analyze, data that will help us to see what sort of response we have achieved using different channels, like broadcast media, out-of-home billboard, for example. Our assignment is to leverage that data to clarify the impact on those investments. Now, regression analysis is really going to be our friend here. What it allows for us to do is to find those relationships between those marketing outputs and the business outcomes that we need. So, what we're going to do in the next few videos is look at how to model this data.

3. Prediction

Decision Trees

Hollywood loves making movies about prediction. There was Moneyball, which predicted athletes' performance in baseball. There are countless movies about someone trying to outwit the casinos or really their own luck. There's the story of the housing bubble and how one commodities trader predicted the recession and made billions buying against foreclosed mortgages. Predictions make for a great story and, in this chapter, I'm going to teach you a popular machine learning technique for predicting events and outcomes called decision trees. A decision tree takes a set of data and it splits that data continuously until it has a predicted outcome. How many times have you heard someone say, "If I only had a crystal ball, I could predict the future"? Well, we're going to do just that, and walk you through the process step by step so you can grasp exactly how to go about it. Now brick and mortar retailers in the US do an estimated 4 trillion in sales each year. That number might be surprising given the power of e-commerce these days, but old school retail is still a significant economic engine. To that point, in this chapter, we're going to help a retailer determine where their next store and expansion of the retail business should be located. And we're going to do so using, you guessed it, prediction. What our prediction algorithm is going to do is assess the ability for different predictors to influence the outcome. In the case of our retailer, the potential predictors that we have available to us in our data set are weather, the radius of complimentary establishments, the population, the number of cars that drive by each store, and the unemployment rate for the store's geographic location. These predictors can help us to understand the conditions necessary for our retailer to make a safe bet on their next location. We just have to figure out which ones matter. So, let's get started with our prediction analysis. These predictors can help us to understand the conditions necessary for our retailer to make a safe bet on their next location.

4. Cluster Analysis

• Cohort Analysis

Think about your closet at home. Do you keep certain types of clothing like T-shirts or dress shirts, jeans and slacks, organized into groups or is it a mishmash? How about a sock drawer, do you have one of those? Whether you keep things organized in groups or not is your own personal preference, and this is certainly not intended to be a critique on your own personal organizational habits. To each their own as they say, but the idea of a sock drawer, the concept of all things in their place as they say. Those sorts of organizational devices help things in general to stay organized, to stay efficient. Need a pair of socks? Well if you have a sock drawer, you know where to go. Now, we do this kind of thing all the time, right? We sort similar things into groups, so we can make sense of them. This is what cluster analysis is all about. Sorting your data into groups. Now, this is actually a very powerful concept when applied to marketing. In the marketing world, we talk about market segments, and competitive sets, and business units, or brand extensions. All these ways in which we refer to our marketing strategy are organized into groups. Each of the different types of cluster analysis algorithms group data based on certain variables. In other words they create and organize your data into different buckets. This can be used for consumer segmentation, or to identify which types of ad units are driving the greatest conversion or brand lift. It could help you to identify what people are responding to. Our cluster analysis case study comes from the world of consumer packaged goods, CPG. You might imagine the client is a popular beverage brand, or a challenger brand in the personal care space. They have come to us seeking our guidance on their segmentation strategy. Now, take off your marketing hat for a moment. How many times in your lifetime as a consumer have you experienced an ad that was irrelevant to either your taste or your needs? It happens a lot, and it can be annoying on a personal level, and it can be a real waste of marketing dollars on a business level. Now, put your marketing hat back on. One of the great things about technological advancements in the marketing space is the ability to target your message to those that it will find the most resonance with. Marketing can be extremely valuable to the right customer, and providing that value can in turn drive revenue, and that's what we want. So, how do we go about doing this? Well, we have to determine where our marketing will provide value. Where it will be welcomed and be useful. We need the right message being delivered at the right time to the right person. So we have to understand which groups, what are known as consumer cohorts, represent which portion of the market so that we can evaluate and prioritize. This approach can also inform our messaging and our creative with consumer insights. So, the way we go about doing this is with cluster analysis. For example, imagine our CPG client leverages email marketing as one of their channels. Analysis has shown that the right kind of personalization in email messaging can dramatically improve performance. We might have a large group in our database with families, and their needs will likely be different from other groups. Safety and wellbeing for example might resonate most with this group. That same database might have a large group of recent college graduates, and value might resonate most with them. So you see, I'm talking a lot about groups. Our client may have millions of customer records in their database, and if we can organize that information into the right groups, we can do marketing personalization at scale.

5. Conjoint Analysis

Conjoint analizi, kaynaklarımızı nereye odaklayacağımızı bilmemize yardımcı olur. Verilerde göreceğiniz kategorilerden biri, differentiation kategorisidir. Müşterimizin sosyal medyada henüz hiç deneyimlemediği şeydir.

6. Agile Marketing

Agile marketing

There are three strategic best practices that will provide you with the right foundation for data driven marketing. The first of these is agile marketing. Agile marketing is an iterative workflow implementation focused on speed, responsiveness, and maximizing performance. It's important for data driven marketing organizations for one key reason. If you're leveraging the data to make informed decisions about how to drive performance, then you have to be responsive. You have to take action on the data. You have to take action on the insights your data analysis reveals. Time and time again over the years, I've observed expensive marketing research efforts where the results get housed in a three ring binder collecting dust on a shelf somewhere. This is not the type of data analysis we've been discussing in this course. The shelf life for some of these insights can only be a few weeks, or a few days, or even a few hours. That requires not only a shift in thinking, but a shift in process. Think of it this way. All marketing consists of three things. Planning, acting, and tracking. The planning consists of strategy, the acting is the execution of your plans and programs, and the tracking is all about analyzing the right data to validate results and find hidden insights. Envision a triangle in your mind's eye. Each corner of this triangle represents one of these three elements. One corner is the plan, the next is the act, and the next is the track. Now, imagine if we were to evolve that triangle into more of a circle. Now the lines and the points separating each element gets blurred. It becomes difficult to see where one element begins and the other ends. The triangle approach creates a silo effect. Each element exists in its own corner, somewhat rigid. The circle represents an agile approach. The idea here is transparency and responsiveness, and collaboration. So, the question becomes are these three facets of your marketing organization organized in a triangular workflow, or a circular one that is as responsive as it needs to be? If you categorize yourself in the former, you are certainly not alone. It can be a challenge, but there are some great resources out there today to help you learn agile workflow, so that you can bring it into your own marketing group. There are devices like backlogs and sprint plans and standup meetings, and realtime collaboration tools, and these have all gained a lot of traction over the last few years. The right solution is typically one that borrows from these conventions and does it in a way that is iterative and open, so that you can find the solutions that are best for your group.

Design and conduct market experiments

I've analyzed many marketing campaigns over the years. Some were performing okay and others less so. My job is to help these campaigns and I find that using an approach called marketing campaign testing provides me with a nice set of tools that guarantee campaign success. Here's one way you might go about doing a marketing campaign test. I call it the MVC, or the minimum viable campaign. Now, the minimum viable campaign is a marketing campaign that invests the minimum amount of resources necessary to validate performance. So that a marketer can see the right opportunities to scale marketing investments and drive growth. Here's how it works. First, you put an MVC road map in place, which includes a hypothesis, stated objectives, requirements for data, and a resource plan. Second, you execute the campaign. And then third, you analyze the data from that execution to determine whether you can scale that program or pivot the effort. Now, you can take this approach with any channel and any campaign. You can do it with traditional media, social media, digital media. You can test creative messaging, call to action. Really any other variable you can conceive of. Let's borrow from one of our case studies to help put this in context. Pick any one you like. It can be our CPG client, or our airline client, or our retail client. Or, consider your own brand, your own company. And let's say we're going to implement a social media program MVC. So again, first we need a MVC road map. The hypothesis here could be that if we lift our share of voice, it will drive increase word of mouth and in turn, consumer consideration. Next, we need our stated objectives. Our objectives for this MVC are to lift our share of voice to either prove or disprove our hypothesis. Our requirements for data ultimately answer the question, how much data do we need and of what sort? Okay, so, here we know we need share of voice and a measure for consumer consideration. And our research plan has to address our budget, our timeline, and a project plan that clarifies who does what when. The MVC is a powerful concept that can reveal your best opportunities for marketing programs that will perform and drive growth.

Stakeholder alignment

As important as data is, you will only realize its full potential when you get everyone involved. Each member of your marketing team needs to be aligned and on board with a data first way of thinking. This is not to suggest a data only way of thinking, however. So keep that in mind. The best marketing is often unexpected and echoes a sentiment by Ralph Waldo Emerson. He once said, do not go where the path may lead, go instead where there is no path, and leave a trail. This is the sentiment that the best marketers know. Without blazing a trail, it can be difficult if not impossible, to break through the clutter and have you message heard. It's important to have your creative resources in alignment then with your data driven marketing. Also important to have your marketing directors and your marketing managers and lo and behold, please make sure your strategy team gets it too. It will have an impact on everyone involved and it will require some change management. Again keep that in mind and just plan for it. One of the biggest hurdles for organizations these days is execution, getting things done on strategy as effectively and as efficiently as you can. To combat this challenge, we have to hold our teams in our programs and one another really, to accountability. So inventory who your stakeholders are, who your teams are, really who's involved, and take the time to reach over that proverbial aisle to establish the necessary shared understanding and systems to support your success. Everyone within the marketing organization should understand the skills taught in this course. That's one of the only ways you'll be able to remove some of the uncertainty that comes from conversations around data science in marketing organizations today. Anyone with a stake in the outcome should be able to feel comfortable and confident. Talking the talk and walking the walk.