We all know the importance of data analysis. More data than ever is now available, and the benefits of leveraging this data, through AI, automation, etc., are enormous.
But in practice, at least in digital marketing, there seems to be a huge gap between the possibilities and the realities of data analytics. I will write a series of articles about this topic, but in this article, I will focus on what is probably the biggest obstacle between data availability and ripping off its benefits.
The difference between visualization and analysis
Not so uncommonly, data visualization and data analytics are confused. One is not necessarily more important than the other, but one is normally dependent on the other. The most value is created when both are done correctly.
Data analytics is about finding actionable insights. It is about making data-driven decisions that ultimately lead to benefits, such as revenue growth, cost reductions, higher margins, more competitive advantages, and other relevant outcomes. Without decisions that improve business performance, data analytics is irrelevant. The more relevant the insights, the more value will be added to the decisions we make.
Visualization, however, is about outputting data (i.e., generating graphs and charts) in such a way that analysis may be easier, faster, and more relevant.
Some people are analytical geniuses who can extract those actionable insights very quickly from raw data (think of the Matrix movies). But most people, like myself, are very visual and need order and structure to make the best decisions in a productive way.
If visualization is done correctly, more people can obtain better and more relevant insights and do so faster, with a smaller margin of error.
Reporting and visualizing data the right way
Most reporting is, or should be, done in dashboards. Dashboards automatically display information in a certain way without having to manually generate reports each time. Everyone should aim to have more and better dashboards and fewer spreadsheets and presentations.
When designing dashboards for data visualization and reporting to find actionable insights, I recommend following these principles.
a) Looking at the big picture first
There is so much data, so many campaigns, sources, ads, landing pages, etc., that it is extremely easy to get lost in details. That's why, as the first step, I recommend presenting first the aggregate data, or the big picture.
Looking at the big picture will allow us to answer big questions, such as how we are doing. Are we improving? Are we on track to achieve our goals? Where should we focus our attention? What needs to be fixed?
The most important insights probably won’t come from answering these questions; they will come from answering the next round(s) of questions. But looking at the big picture first will increase the probability that at step two we will ask the most relevant questions and find those key insights.
Zoom in on the important stuff
Even the smallest company has enough data that would require an excessive amount of time to analyze if we tried to do so exhaustively. That’s why establishing priorities and learning to zoom in on the most important stuff will lead us to a more productive and effective analytic process.
b) Identity underperformance
Some of the things we should look out for are the underperformance of sources, campaigns, ads, regions, audiences, etc. By identifying underperformance, we can make resource resignation decisions that ultimately lead to better overall performance.
Resources are always limited, and optimization is about systematically relocating resources to where they will have the highest return.
c) Identify overperformance
Identifying underperformance is not enough. We need to know where to reallocate resources by identifying what is delivering the best results and relocating the resources there. Once again, this could be done by investigating sources, campaigns, ads, regions, audiences, products, promotions, etc.
One important caveat about analyzing overperformance is understanding the decreasing marginal returns we face in digital marketing and the oversaturation of audiences. Our dashboards should help us identify when something that is overperforming might be becoming saturated, and therefore, we should be careful about the over-allocation of resources there.
e) Interpret data: Qualitative thinking vs. algorithmic thinking
When examining underperformance and underperformance data, we should try to identify patterns that are not in the numbers. These could be ad or audience characteristics. Good dashboards should make it easy to identify these patterns, with things so easy, such as putting the creative in the dashboard, so we don’t need to imagine or go somewhere else to see the specific creative that is either underperforming or overperforming.
f) Understanding the importance of statistical significance
A digital analyst does not necessarily require a formal education in statistics. However, there is a technical concept that is critical to understanding (even if we don’t know how to calculate it). In statistics, the size (or amount) of the data matters. Too few data points, too few observations, or too little time could lead to making the wrong decisions. As a rule of thumb, we need to let our efforts collect enough data before deciding. The fewer impressions, clicks, conversions, etc., we have, the greater the probability that the insight we are obtaining is true, not an illegitimate conclusion that is not statistically validated.
g) Don’t forget about seasonality
A few days ago, I saw someone celebrating on LinkedIn how a decision they made on a campaign led to an increase in performance in the first weeks of January versus the last weeks of December.
Don’t get me wrong; maybe the decision they made was the right one. What’s wrong is to jump to that conclusion by comparing the data between the last two weeks of December and the first two weeks of January. For many industries, the last two weeks of December are the worst for most KPIs (CTR, conversion rates, etc.)., and the first two weeks of January are among the best of the year.
Seasonality is present in every single industry that I’ve ever seen. Ignoring it inevitably leads to data misinterpretations and bad decisions.
Data management done right
One of the hardest, or at least more time consuming, parts of visualizing data the right way is the correct management of data. We followed this simple process:
a) Extracting data
Raw digital data is live on platforms such as Google Analytics, Facebook, Google Ads, CRM systems, etc. We need to set up an automatic extraction of data from those sources to make sure everything else is always up to date.
Even though API integration through coding is sometimes inevitable, tools such as Supermetrics reduce the need to involve software developers in the process, in some cases even making it unnecessary to involve software engineers at all.
b) Data centralization
Extracting data and putting it where? Well, it could be as easy as putting it in a series of Google Spreadsheets. If organized and structured correctly, Google Spreadsheets can go a long way as a marketing data warehouse tool. For many companies, this is enough.
For large companies (you would know, by now, if you are one of those), this won’t cut it, and more robust databases would be required, such as Google Big Query.
The good news is that most companies can just start by using Google Spreadsheets, and then, when data starts getting slow and heavy, move to a more powerful tool with the help of a software engineering team.
c) Data visualization tools
I talked a lot about data visualization, but not about tools. There are several amazing visualization tools out there. At JULIUS, we’ve used Power BI, Tableau, and Google Data Studio, the last one our favorite for digital marketing data visualization.
All these visualization tools have the functionality for most of the visualization ideas that we’ve ever had from a digital marketing and e-commerce perspective.
In conclusion, it’s important to understand the distinction between data management, data visualization, and data analytics. These are complementary but disparate functions, and many times, they could or should be done by different people. However, if one is not done correctly, we won't be able to truly be data driven.