A Miracle of Data Visualization

Opinion

The correct way to improve the quality of business analytics

Choosing the right representation of your analytical work results directly influences the quality of management and decision-making. In this article by Andrew Bush, the founder and CEO of A17 Technologies company, find out why data visualization mistakes can cost your company a pretty penny and how to avoid significant losses.

No matter how skilled your company’s analytical team is, or how complete their data and algorithms are, the results can turn out to be dissatisfying. Often, this happens due to incorrect visualization of the work, on which directors, investors and shareholders base their decisions.

Visualization is one of the stages of data analysis that helps to demonstrate the main trends, dependencies and deviations of indicators in an informative and interactive way. To this end, various graphical elements are used. For example, tabs, diagrams, informational panels (dashboards), etc.

It is important to understand that visualization is not just a fancy report, but is rather a chiefly comprehensible business logic which helps to make correct management decisions. The truth is, that those who are new to implementing and regularly using BI tools may conflate these ideas. The entire analysis value could be erased in minutes due to a bad presentation of data to your senior leadership team. Having spent an enormous amount of time studying a summary tab, a decision-maker may fail to notice dependencies because of imperfect or inconvenient presentation formatting.

Let’s consider the situation of presenting the results of your analysis to your director in order to make changes in the company. If your data is not presented well, your boss is unable to comprehend information in the first three seconds. They have no time to look into it, thus doing only a cursory review. The whole work turns out to be ineffective. That’s why you need to learn how to visualize data.

Business analysts who not only possess the skill but also know the correct approach to present data findings and insights, plus to provide a satisfying solution to their client’s demands are always highly valued… and desired in the job market. There is also quite a strong demand simply for specialists in data visualization.

There are several important things that should be taken into account. First of all, consider the audience — those who will review the data. Most often these are company directors, department and subdivision supervisors, stakeholders, investors, and so on. Good visualizations are made in much the same way that presentations for corporate events. E.g. depending on the target audience present at the meeting, the presentation will look different.

A visualization method also depends on its purpose. When you have to prepare a presentation on the dynamics of the quarterly performance stats for your supervisor to quickly look through is one thing. In this case, you need a simple diagram, which can be understood in 1–2 seconds with no further explanation and be instantly interpreted. However, when visualization is required for digesting complex corporate processes for transforming the business is an absolutely different story. In this case, it should be finely detailed so the decision-makers can have the opportunity to see the whole picture of the processes at hand. Then, the interpretation can take hours, days, or even months.

Besides, visualization can be a one-time requirement when the status of a business’s margins needs to be presented, e.g., in a quarterly statement. Or when visualization is required for preparing constantly updated dashboards. These are completely different situations and they need different approaches. For instance, dashboard visualization is suited for quick comprehension, and as such, it should have minimal information so a supervisor can quickly understand the data.

Moreover, there are several basic rules that should be followed regardless of the type of visualization. All of them are connected to comprehensibility and represent the correct meaning of the data.

First and foremost, keep in mind the «less is more» principle. That is, the fewer excessive details there are the better. For instance, avoid placing various indicators on the same axis, using 3D diagrams, shades and other fancy things. Removing a diagram grid if it is not necessary as well. The more lines there are in a graph, the more difficult it is to read the information correctly. For example, if we have a diagram with several lines showing profitability, available stock, etc, then they should be broken down into separate graphs. A notable exception is when we need to compare data. This may occur when you compare the profitability of your company’s branches according to their regions or federal districts.

Here you can find a few more examples of the worst diagrams of all time.

One way or another, all visualization is built on comparing: with zero level, with similar figures for the past periods, in colour. Should your diagram have several lines, you have to provide a clear description: which values mean what. Apprehensible legend in such cases is a must. Besides, do not use several Y axes: this may lead to confusion of the reader and prevent a better interpretation of the data.

One more mistake that beginners in analytics often make is axis scaling (starting from non-zero points). This leads to incorrect data interpretation. As a result, the quality of the whole analytical work suffers severely. Most often this situation occurs when using automated tools. For instance, if you take the graph about the workload of a company’s production line for a year in percentages, it may only change from 80% to 100%. Meanwhile, the figures themselves are in a range from 0% to 100%. By default, some automated visualization tools can conceal values from 0 to 80% by showing only figures from 80% to 100%. And so, readers of this graph will have an impression that these values vary significantly, while in reality, they do not.

A similar issue is logarithmic graphs. It’s better to never use a log scale, as it is non-linear. If for some reason such a representation format is used, then it should be expressly noted: readers should understand the fact that it is a log scale, and why it is even being used. By no means you should indicate this by using an asterisk below and in small text size. That is a red flag.

Special care should also be taken when using histogram graphs. Histogram values should always be sorted by measure, not in alphabetical order or randomly distributed, as is usual. The result of such a visualization should be triangularly shaped, showing figures from the smallest to the largest or vice versa. Such a representational format allows you to easily recognize maximal and minimal values. It will be easier to work with these data in the future, having a clear picture of the situation in mind.

Integrating Corporate style and identity into your visualization is significant, which means including the company’s colour scheme and fonts among other things. It is good practice to utilize a colour pattern from a company’s brand book. For this reason, analysts should inquire if there is a separate guideline for data visualization in the corporate brand book. In larger companies, it definitely should exist and will also probably contain a certain font, line weight, and text layout.

You should also take into account psychological associations between colours and ideas. For instance, by default the colour red means “stop, danger, or caution”; green means “go, safety, OK”; blue is usually something cold; yellow often marks non-important things. These colours also affect each other. Consider the previous example of figures in a comparison diagram of branch profitability in various regions. If they are coloured differently, red and green must not be used to differentiate the data from one region to those in the other region. Due to psychological biases, data values can be incorrectly interpreted. Red-coloured figures may be perceived negatively, and green, positively. It’s best to colour all values into one neutral colour and tweak its saturation (brightness) or use shades of close colours.

The most commonly used BI tools already have default settings which help to prepare diagrams in a unified style. Analysts just need to make correct diagram legends and adjust colours. But sometimes custom visualizations for an annual meeting of shareholders are required. After business analysts complete their work on the prepared data and finished dashboards, these should be forwarded to designers who redraw them at the same scale but with correct indentations, typography, colours, fonts and line weight.

One more principle of high-quality visualization is knowing its use case, i.e. how a user will utilize diagrams, dashboards and report sets. And for this purpose, it is necessary to provide easy navigation from general to particular things.

Navigation is the scope of certain elements in diagrams which help to switch to various types of diagrams (pages) and perform actions like sorting values by years, quarters, months, branches, and so on. As a rule, widespread BI tools generally have inbuilt settings for navigation, such as button filters, or hyperlink compatibility to other pages. But a developer has to calculate how specific infographics will be used by the company’s employees. It’s insufficient to rely only on one’s own opinion. The analyst has to observe how the supervisor uses analytics at the present moment. This is done by tracing their clickstream between Excel files, evaluating sequences of questions on functionality via correspondence or messengers, etc.

While working on visualization it is important to understand how interactive it is. An example of static visualization is screenshots from a corporate BI system based on which a graph, diagram or picture is created. In the end, all of them are used in presentations or PDF files. They cannot be changed over time.

Interactive visualization requires a different approach. For instance, most analytical tools allow for the creation of dashboards or reports that support further analysis. They provide options for all kinds of cross-sectional reports and filters. It is intended that a user can continue working with all the data, and change graphs and their contents at the person’s discretion. In this case, it is important to divide graphs as much as possible into separate reports and report pages, sorting them by purpose, cross-sections or degree of detail. Here the principle of “less is more” is again relevant.

Apart from that, when preparing a visualization it is worth remembering to adopt it for various types of devices: PCs, laptops, tablets, smartphones, etc. The solution for this can be both prepared PDF files and dashboards. At the preparation stage, it is vital to take into account all the peculiarities of your content visualization on large and small screens, so that developers will not have to check the formatting after each data refresh in the future. Alas, many, unfortunately, a lot of BI systems are not adapted for working on mobile devices. As a rule, compatibility in this area is achieved by utilizing a programming specialist from the company.

Also, the data have to be correctly prepared for visualization, so that the amount of data loaded into a report is minimal. This issue is usually addressed by a system analyst. If there is too much data, a report will take a lot of time to load. If the report is not drawn up for a couple of seconds, nobody will want to work with it.

To sum up, as a matter of fact, specialists responsible for data visualization for business analysis deal with improving the user experience for decision-makers. Their task is to ensure optimal user interaction with an analytical report or a dashboard. The visualization system itself tends to solve issues pertaining to UI.

All images unless otherwise noted were created by the author.

A Miracle of Data Visualization Republished from Source https://towardsdatascience.com/a-miracle-of-data-visualization-5e966c2e1aa0?source=rss—-7f60cf5620c9—4 via https://towardsdatascience.com/feed

<!–

–>