The essence of every marketing effort is to maximise ROI. It is the driving force for every marketer; the single thing making them jump from bed in the morning and go to work. But in order to maximise ROI, it is necessary to understand all the factors which play a role in the overall success of advertising efforts.
Thankfully, tools for online marketing provide many indicators that allow drilling deep into various phenomena related to an audience, their structure, behaviour, and engagement. With relevant information, you can understand causes and keep improving your marketing activities.
While watching at charts and data tables might be insightful at the intuitive level, there is always a risk that you simply do not see the important data, or misinterpret it. For that reason, instead of relying on pure intuition, sometimes it is good to build and engage a cognitive framework, which can help to efficiently direct thoughts when exploring data. Such framework could provide strong guidelines to more efficient understanding of analytic data without repeating common mistakes in reasoning. Exploratory data analysis is a part of analytical thinking and deserves to be structured in a cognitive framework.
One of the most important influencers and experts in the presentation of informational graphics such as charts and diagrams is Edward Tufte, a professor of political science, statistics, and computer science. Among other prominent literature, he published a book called Beautiful Evidence. The text explains basic principles of analytic graphics, which can be applied to various applications of data presentation and exploratory analysis. Those principles are:
1. Show comparisons, contrasts, differences.
2. Show causality, mechanism, explanation, systematic structure.
3. Show multivariate data; that is, to show more than one or two variables.
4. Completely integrate words, numbers, images, diagrams.
5. Thoroughly describe the evidence. Provide a detailed title, indicate the authors and sponsors, document the data sources, show complete measurement scales, and point out relevant issues.
When I discovered Beautiful Evidence, many questions opened in my mind. How can this framework be applied to online advertising? Are there any specifics in advertising industry in comparison to general principles? What can we do at our company to make analytical tools and reports more appropriate to support these principles?
So, I decided to share some of my reflections in this post. I focused on three principles at first, because they are particularly important for anyone who works in online advertising or affiliate marketing.
Show comparisons
Tools for online advertising are capable of tracking a wide variety of data about your audience and your campaigns. Aggregating this data among various dimensions yields valuable measures of relevant user actions such as clicks, purchases, leads, or app installs. If interpreted in the right way, these measures can serve as indicators of audience building, awareness, user engagement, and other parameters you define as goals of your marketing efforts.
However, each of those indicators is always relative to something else. It cannot stand by itself. In order to give meaning to absolute numbers, you must compare them to something relevant. Even when you rely on your intuition while quickly browsing through preliminary indicators, you are actually comparing them with your expectations learned from experience. So, instead of risking possible misleading subconscious comparisons, it is always better to raise awareness and learn how to compare indicators reasonably, systematically, and purposefully.
The first big decision is to choose the comparable and relevant object for comparison. In scientific research, this is usually a control group – a heterogeneous data not subjected to the change of key variables. For example, if scientists were testing a new medicine, the control group would not receive any treatment, so they will be able to compare whether the medicine works better than not taking it at all and whether it has any negative side effects. This principle can be loosely applied to online advertising.
Let’s say that you have a new landing page and you would like to understand how it is performing against user engagement and conversions. In that case, you might compare it to the old landing page and measure the number of conversions or time on site. Such a comparison will clearly show whether changes in the new page made the page better for visitors, providing you did not change any other aspects (such as marketing efforts to acquire better converting traffic).
It is always important to make data comparable. If the new page has more visitors for the same period of time as the old page, you need to normalise some indicators. So, instead of comparing absolute differences between visits and conversions, you should compare ratios, i.e. conversion rates. There are a lot more situations when data needs to be normalised, so do not forget to do this.
In online advertising, there are many things to compare to – various campaigns, different ad types, design and text on ads, other target groups, other time periods, and much more. So, be ambitious – think of various possibilities and do not forget to consciously compare indicators.
Show causality
It’s a habit of indicators to primarily disclose phenomena and not their causes. If you want to act and improve your marketing performance, understanding the underlying causes is essential. Instead of just observing a single indicator, try to put it in the context and build a causal framework for thinking about it. The context will help you to understand why the indicator has certain values in comparison to another one. More importantly, considering the causal framework can lead you to actionable ideas.
For example, if you notice that your conversion rate is dropping, try to think of possible causes. Has the quality of traffic changed? Have you tuned some campaign parameters? Are there any changes on the landing page? Is there a new competitor on the market? Keep in mind that the causal framework might include factors which are not under your control or which cannot be accurately measured by tools that you use. However, if you become aware of the real cause, then you will have the power to do something. You could reduce bad traffic with better targeting, try different ads, optimize your landing page, or improve your offers if a new competitor has better ones than you do. The beautiful thing about online advertising is that it can easily point out problems that would otherwise require a constant and thorough analysis.
Again, instead of ignoring the possibility to visualize causalities and underlying mechanisms, we’ve implemented features that allow showing multiple dimensions in a defined scope into all of our products. Those features are fundamental for showing causal relations between indicators.
When building a casual framework, it is important to keep away from a common logical mistake that observed correlation in data always implies causation. This fallacy is also known as cum hoc ergo propter hoc, Latin for “with this, therefore because of this”. Sometimes even experienced scientists make this mistake.
As an example from online advertising, let’s say that you noticed an increased amount of clicks form Germany together with an increased number of overall conversions. You could jump to conclusion that German visitors are performing better and decide to invest more in German traffic. But this may not be a good decision. There could be a number of possible reasons for the better rates of conversion.
Maybe there is another country, which has even more clicks and even better conversion rates. Maybe you used different ads for German traffic. Take time, think of possibilities, and compare. Try to find the right cause of phenomena you observed and this could save you from making mistakes. As Tufte suggested, correlation is not causation, but it sure is a hint. In order to better understand relations between indicators, show them together as multivariate data.
Show multivariate data
The real world is multivariate. There are many dimensions within which you can observe data. It is the same with online advertising. You can observe the same indicators for different countries, campaigns, ads, target groups, device types, operating systems, browsers, traffic sources, internet connection types, bandwidth, keywords, and many other dimensions. In order to compare data and to understand a causal context, it is necessary to show indicators against each other. Tufte calls this “escaping flatland”.
In our previous example with increased clicks from Germany, you could easily see whether there is a country with better conversion rates just by showing the same data for all other countries. Always think of possible dimensions and build new perspectives on your data, otherwise indicators may mislead you.
Analytical tools and reports can significantly affect your way of thinking. Their limitations and interface designs may close some perspectives. It is important to raise awareness and always try to find a way to explore data from different angles. If your tools do not allow you to show multivariate data, maybe you can export raw data and continue the analysis in a different tool.
Conclusion
When making important decisions for either our companies or our private life, we rely on various cognitive frameworks. We use benchmarks, strategic planning, SWOT analysis, the Six Forces Model, and many other principles, some of which we internalised during our education and some of which we learned in practice.
Analytical data exploration is very demanding and deserves a specific framework, which can help us avoid common mistakes and focus on real causes. The set of principles explained in this post can help you understand data better and therefore make better decisions.
Next time when you open a dashboard in your favorite ad server or affiliate tracking software, do not forget to compare indicators, think within a causal framework, and show multivariate data. Just being aware of those important principles may lead you to better actionable conclusions.