An Initiative to a Sophisticated Marketing Data Analysis Journey
At Glacier Media Digital everyone is constantly going down steps of forming, storming, norming, and performing. With the introduction of Google Tag Manager, we have been exploring countless ways of monitoring user experience on their omni-channel experience on the internet.
We are in the process of learning from the data that we get from various channels in order to outperform previous endeavours. One of these endeavours takes place in programmatic advertising, where we continue to see stellar results over the years, allowing us to create predictions based on trends that we see from data that has been collected.
Isaac Asimov anticipated that in the future we would be able to predict what would happen based on past history. As data comes in more concentrated amounts, it is plausible to say that maybe one day we can accurately predict future interactions with a very high confidence interval. The fact is that data has to provide meaning in what our interactions are, which leads to classifying interactions based on a set of principles; as it shall be understood, it is our key to learning aggressively to match complexities that arise from different sets of outcomes. Technology allows us to to do this a lot faster. Asimov anticipated that education would be forever changed by technology, and it would be a fundamental part of teaching and learning. That is why, when learning about programming, cryptography, mathematics, statistics, physics, biology, and others, creating something beautiful comes as unexpected but soon it is realized that creation itself is beautiful but definitely not unexpected as it is only created after something that already exists. With respect to Isaac Asimov and his book series, “Foundation”, it is relieving to be able to imagine the future. It is always best to use our imagination rather than to succumb to only rational outcomes that may be a possibility of occurring in the present because of the fact that there are some events that shake the whole world that are extremely hard to predict. Apple TV+ created a TV series from Isaac Asimov’s book, “Foundation”, because the concepts inside the book are getting more prevalent in this day and age.
This article contains data collected from programmatic sources from January 1st, 2020 – September 8th, 2021. Not all sources will be analyzed and only data that has been streamlined from two feed sources will be analyzed. For the sake of this exercise, data collected will be anonymous and names will not be revealed. Let’s take a journey and dig deeper into patterns that emerge from close to 2 years of progress and let us keep in mind economics as a factor being affected by the black swan event, severe acute respiratory syndrome coronavirus 2 or better known as COVID-19, which got world wide recognition on December 31st, 2019 when WHO (the World Health Organization) was informed of cases of pneumonia of unknown cause.
Summarization of the main components into 3d graphs is done to showcase trends. The resulting 3-dimensional structures allow us to understand how different variables interact with each other.
Programmatic Development (Impressions vs CTR vs Date)
The confounding variables as a result from COVID-19 can be seen to create dispersions in the data. Seasonal trends can be seen to create date clusters. Impressions served have been growing achieving great results in CTR.
Programmatic Ads Comparisons (Impressions vs Clicks vs CTR)
As impressions increase the rate of an ad receiving a click increases and it shows that CTR falls due to the same reason.
Programmatic Geography Comparisons (Impressions vs CTR vs View Visits)
View-through visits (visits as a result of delayed click-throughs) can be as powerful as CTRs, and should be taken in account in order to monitor the behaviour of ad viewers through different geographical locations.
Programmatic City Comparisons (Impressions vs Clicks vs CTR)
CTRs normalize as impressions and clicks increase. Impressions and clicks illustrates a noticeable linear relationship.
Future Developments in 3D Predictive Analysis
The next steps to visualizing data is creating automatic predictive analyses, which then takes the current 3D visualizations and illustrates the predicted outcome based on a set of patterns (how variables can change in relation to impressions, clicks, view visits, CTR, date, and whatever else shows a positive or negative correlation). For example, these future prediction can be layered over the current 3D illustration in a different color to not be confused, or they can be illustrated totally differently. It all depends on how one may like to visualize data, what it is influencing, and the nature of future predictions.
As a requirement for complex automation, algorithms need to be used to help classify decisions on modelling the data. For example, using a popular method for a 3D predictive visualization model similar to the 1st illustration, “Programmatic Development (Impressions vs CTR vs Date)” such as the Random Forest algorithm – would require a breakdown if there are patterns that emerge from confounding variables. These sequences must then be correctly classified to be used to create regressions which can be done through the popular K-Means algorithm to create data cluster (in this case it would be clustered based on the z-axis, “Date”). The following points as a result from the regression must then be allocated according to the other 2 variables on a 3D illustration. This would result in a complex 2D plane that showcases a probable future of where events would lie. It would be easier to create a plane for every cluster formed, these planes can be used to compare where current data points are in relation to where they have been predicted to be to train the program to predict with greater accuracy and to spot outliers and determine the factors affecting them.