3 Questionable Gaps in Visual Analysis and a Potent Way to Overcome Them

This article is the follow-up to my previous article “Visual Dashboards Aren’t Optimal for Analysis.” I recommend that you read that article if you have not already read it. 

Before we go into solutions for visual analysis, let us try to look at analysis from first principles. While analysis is different for different user personas, there are common themes of analysis that are generic for any analysis use case. We will review the common analysis challenges and offer solutions for them in this article.

1. How can the user easily observe all the required performance data? 

First off, the starting point of visual analysis involves being able to view all the performance measures that the user is likely to be interested in for the entity he/she would like to analyze. Performance measures could be for any entity that you are analyzing such as a company, business unit, product, region, etc.

Performance measures that are relevant to the analysis should be easily accessible and observable. In this data-driven world, the number of measures is increasing. As I discussed earlier, dashboards offer limited visual observability because they can only visualize limited measures, while semantic layers only offer a clunky querying experience that is tedious for end-users.

So, what is a better visual analytics solution that makes performance easily accessible and observable for any end-user?

Solution: 

I am going to refer to the solution for easy accessibility and observability as “Scorecards.” So, what is a Scorecard? There is no agreed definition for Scorecards. I define it as a place where you get to view all the performance metrics for the entity you are trying to analyze.

A holistic scorecard would include measures from even different functions such as sales, product, finance, operations, and human resources.

Viewing a Scorecard with all the performance measures relevant to analysis enables users to get a holistic view of the performance of the entity that is being analyzed. So, what does a Scorecard look like?

Below is an example of a company-level scorecard for a B2B SaaS firm. Scorecards will be different for different industries, functions, or personas, but the overall form remains the same. I.e. key measures required for analysis are categorized and organized vertically.

Although the example below shows the time-series scorecard at quarterly time points, scorecards can be for daily, weekly, or monthly time points. Similarly, functional scorecards like a finance scorecard can include more financial metrics and fewer non-financial metrics, but the overall structure of the scorecards essentially remains the same.

The concept of Scorecards is generic and can be applied to any use case. They can be created for any entity within an organization including but not limited to company, business unit, function, product, etc. Visual analytics solutions can be built such that users can easily pull the scorecard of any of the entities on demand.

One advantage of Scorecards is that they can scale vertically to include as many measures as desired. In this big data world, where more performance measures are being generated, they give a holistic view of all measures required for analysis in one place. Visual dashboards cannot scale to hold a large number of measures like Scorecards because of screen real-estate reasons highlighted in the previous article.

While using semantic layers for analysis can give users the ability to query measures, it does give a user experience for analysis. While a Scorecard seems like a static view, the measures in it can be easily added or removed, and deep interactivity can be layered on top of it.

One key aspect that strikes as soon as you see a Scorecard is that it is essentially a tabular view. In the world of data visualization vendors, tabular visualizations are generally looked down upon and graphical visualizations are preferred. This is true when you are doing a simplistic analysis or communicating insights to an audience.

But, if you are doing somewhat complex analysis, tabular views enriched with heatmaps and sparklines are powerful. Users love Excel and pivot tables for analysis for a reason. In a Scorecard, the limitations of tabular views can be offset by enriching them with heatmaps and sparklines.

So, why do tabular visualizations beat graphical visualizations when doing advanced analysis? One reason is that tabular views simply take less screen real estate. So, tabular views give you more context for analysis compared to graphical visualization for the same screen real estate.

This difference is illustrated in the 2 visualizations below. In the same screen real estate, a tabular scorecard can show 10 measures while graphical visualization shows only one. The second reason is that Scorecards are organized and structured. Visual dashboards, in contrast, tend to be disorganized. Deep interactivity can be built on top of scorecards because they are structured.

But, what about graphical visualizations? A user experience hack that I refer to as “Visualization-on-demand” involves building interactivity on top of Scorecards to generate visualizations on demand, i.e., users can select the measure that he/she wants to visualize and click. And voila! any visualization can be generated!

A similar user experience hack that I refer to as “Dashboard-on-demand” involves building interactivity patterns where a user can select a group of measures that he/she wants to visualize and be able to generate a dynamic dashboard. 

While visualization-on-demand is achievable with the features provided by the data-visualization vendors. Dashboard-on-demand can be a challenge unless data visualization vendors improve their feature set over time, as most vendors offer only dashboards with fixed layouts.

Scorecards enriched with interactivity to generate visualizations and dashboards on demand solve the pervasive problem of visual observability, i.e., view all performance measures in a single place and generate any custom visualization on demand with a single click.

Similar to visualization-on-demand or dashboard-on-demand, more advanced interactivity patterns can be layered on top of scorecards that we will review in the section below.

2. How can a user see any performance in different contexts and dive deeper when needed? 

Data by itself is meaningless. To analyze data, the user needs to view the data in context. Notice that the scorecard always gives more context in the form of more performance measures. Context is key for analysis.

In the scorecard example above, each measure is seen w.r.t. to other time points. So, time is the other context for the analysis in the scorecard example. To analyze performance, users should be able to see performance in multiple different contexts. That means users should be able to:

• View performance w.r.t. time

• View performance w.r.t. other comparable entities (business units, products, etc.)

• View performance w.r.t. plan

• View performance w.r.t. other performance measures

• View performance w.r.t. competitors 

• View performance w.r.t. industry benchmarks 

Viewing performance measures in different contexts helps analyze performance holistically. In addition, users should be able to break down any of the performance measures that they desire. Let us see how these problems can be solved with Scorecards. 

Solution:

The Scorecard that we have seen previously where each performance metric is seen w.r.t. other time points is called a Time Series Scorecard. We can create other kinds of scorecards where the scorecard template remains the same i.e., how the measures are organized remains the same, but the context of analysis changes.

For example, the user can see the same scorecard measures w.r.t. other entities you want to compare against or performance benchmarks or w.r.t. plan. Interactivity can be built to allow users to switch from a time series scorecard to other kinds of scorecard views seamlessly. Such interactivity patterns are powerful for analysis.  

In addition to viewing performance measures in different contexts, a user should be able to pick any measure in the scorecard and be able to dive deeper by breaking the performance down along desired dimensions. Like visualization-on-demand, the user should be able to select any measure, right-click, and then select how he/she wants to break down the performance. This type of interactivity is illustrated in the diagram below. I called this feature “Measure-breakdown-on-demand.”

3. How can a user do deeper analysis with ease?

Till now, we have seen how to see all relevant performance measures, see any performance measure in different contexts, and break down any performance measure when needed.

The last challenge is that an end-user should be able to do different kinds of analysis with ease. The analysis could include basic analyses such as growth analysis, variance analysis, margin analysis, etc. or it can include more advanced analyses such as outlier analysis, correlation analysis, forecasting, etc. How can user experience for such advanced analysis be achieved? 

Solution: 

The scorecard would still be the foundation for even this kind of advanced analysis. Let us see how we can achieve it. Analysis capabilities such as growth analysis, variance, analysis, and margin analysis can be layered on top of a scorecard. Users should be able to generate the analysis for the selected measures or the whole scorecard just by clicking on the type of analysis needed.

I call this feature “Analysis-on-demand.” I am illustrating how growth analysis applied to the whole time series scorecard would look like:

If the user then wants to drill down the growth rate based on specific dimensions, that can be accomplished by “Measure-breakdown-on-demand” which I described in the previous section. Such capabilities are extremely powerful for analysis!

More advanced analysis functionalities can be layered on Scorecards. Similar to the dashboard-on-demand that we reviewed earlier, users can select specific measures in the scorecard and be able to generate correlation analysis or outlier analysis on demand. Moreover, the whole scorecard can be forecasted with a forecasting algorithm.

I call this “general capability” when an end-user can do data scientist-like analysis as “Data-science-on-demand.” The illustration below shows how a correlation matrix can be generated simply by selecting the measures in the scorecard and choosing the right analysis option. Many analysis possibilities can be layered on top of the scorecard when organizations leverage the scorecard as the foundation for visual analytics.

Advanced analysis, which is typically only accessible to technical users who can write scripts, can be accessible to end-users by having scorecards as the foundation for analysis and layering analytical capabilities layered on the scorecard. My vision is to see even business users have deep analytical capabilities and anybody in the organization can become a data scientist!

Conclusion:

Making advanced analysis capabilities accessible to end-users is what makes any organization truly data-driven. This requires a change in how we approach visual analytics and move away from dashboards and semantic layers for analysis use cases to a scorecard-based analysis with advanced interactivity built into it.

My vision is that analysis should not be the responsibility of a specialist analyst function, but anybody in the organization should be empowered to analyze with ease. An organization’s best path to be able to generate data-driven Insights is when users with high business acumen can do advanced analysis. 

I have had reasonable success in the organizations that I worked with; with the approach I proposed in this article. While I obviously could not accomplish all the capabilities outlined above because of the feature limitations of the data visualization vendors, the outcome is still very impressive and transformative to business users’ analysis capabilities.

Although such requirements might not come directly from the users, building a prototype would help get buy-in to embark on such a kind of visual analytics transformation. My experience is that once you build a prototype and show a scorecard-based visual analytics solution to business users, they will love it and say this is exactly what I want! I would love to hear your feedback.

Please feel free to reach out either to give feedback or if you need help in embarking on such a journey.

*Note: Indukuri wrote this article as an independent contributor and not as a company spokesperson.

About the Author:

Chaitanya Indukuri is a product manager in analytics and insights at Deutsche Bank. He has over 2 decades of diverse expertise across data engineering, business analytics, and data visualization.

Indukuri holds an MBA from the Indian School of Business and a Finance certificate from the Wharton School of Business. He is also a certified Financial Risk Manager. He started his life as a data engineer building enterprise data warehouses and then moved into highly analytical fields such as Equity Research and Finance, Planning, and Analysis where he consulted for leading Fortune 500 organizations.

Indukuri is deeply passionate about visual analytics and has had success in driving visual analytics transformations in many Fortune 500 organizations.