May 05, 2021 Marketing Analytics 101: Key Terms to Understand and Why It Matters
Marketing Analytics Key Terms
In my last blog post, I discussed the four phases of the analytics journey and the types of business questions each phase is meant to address. In this post, I will introduce some of the key words and phrases used in the world of analytics we see today. The intent is not so much to create a data scientist from scratch but to build a basic set of vocabulary that allows one to have intelligent conversations with others about solving business problems with data, even if you are not an “expert”. In addition, more and more of the analytics software being offered these days is intended for “citizen” data scientists…those who are not trained statisticians but who need to provide internal/external clients with answers to business questions nimbly and accurately. If you are someone who is in this role, you’ll need to read on carefully!
Data Science
Let’s start by talking about the concepts of data science vs. machine learning. Data science is a field of science whereby data that is collected from various sources are analyzed to detect patterns that can generate insights that lead to better-run business processes and better decision making for businesses of almost any size. The data used can come from anywhere from Excel or Google sheets to large data warehouses. The data are massaged in ways so that they can be “joined” together to create a single data set for analysis. Data “mining” then occurs to find correlations in the data that can be used to inform more advanced analytics techniques to answer a variety of business questions. (We will discuss some of the techniques as well as how they are used in future blog posts.)
Machine Learning
Machine learning (ML) refers to the specific set of techniques used to develop complex algorithms for processing data and delivers results to its users. The data are partitioned in to a “training” set, where models are developed to find the best one that fits the data, then the model is “tested” on data not used to develop the model to see how well it did at predicting outcomes. This involves some manual intervention on the part of the analyst to review the results and, through an iterative process, refine the estimates to improve its predictive quality. This type of analysis is referred to in the literature as supervised learning.
By contrast, unsupervised learning does not rely on trained data sets to predict the outcomes, but it uses direct techniques such as clustering and association rules (more on these later) in order to develop the predictions. By doing so, they involve little or no human intervention and can be run continuously to improve model results, stopping only when a pre-set error threshold has been met.
Artificial Intelligence (AI)
There is one more term you need to know about…artificial intelligence (AI). Artificial Intelligence is the study to create intelligent machines which can work like humans. Machine learning is key to AI in that it uses complex algorithms to learn through repeated iterations of a task (i.e. a model). This process continues until, once again, the pre-set error threshold is attained and the model results are optimized.
Why does knowing these terms matter? Because the analytics process is moving from one where trained data scientists with extensive backgrounds in statistics and computer science do all of the work to one where marketing analysts with little formal training can quickly run a model (or models) and respond to clients’ questions in hours, rather than in days or weeks, which is what has been the case for the most part currently. This is happening because most of the analytics products being offered today have an “automated” ML component to them. This means that the tool looks at the data and automatically builds a set of models, then chooses the best one and recommends it as the solution. This sounds wonderful on its face but is actually both a blessing and a curse. One problem is that the automated process is only as good as the tool’s ability to identify the right model. Usually, the tool is programmed to only use a subset of the total number of techniques available to solve a particular business problem. However, because the automated ML process is limited in the types of models available, it doesn’t always provide the best technique for the job, leaving the analyst to generate additional programming to add the technique(s) needed to model the data correctly. The second problem is that, even after the models are built, the analyst then has to be able to explain the results to clients (internal or external) as well as how they can be used to address the business issue at hand.
The bottom line of all of this is that, if you are thinking about adding more analytics to your business to develop data-driven insights that can help your business, you need to understand what the trade-offs are between having a well-trained, in-house analytics staff vs. the trend toward “democratized” analytics where everyone can participate in the process. Apollo Consulting can help you in this process, providing guidance and support at every stage of the journey to ensure that you make choices that will lead to success for your business.
Ed Holmquist
Apollo Consulting Group, Providence, RI
No Comments