Introduction
The second lab of the course served to develop skills in the practical application of data visualization theory and best practices presented in the lectures and readings. The process of information visualization transforms data into representations that can aid in cognition and reveal insights. I elected to use retail performance data based on my knowledge of the statistical analysis used within the industry. The retail industry generates large amounts of data each day which are routinely compiled into statistical/financial reports. These reports are used to analyze current trends, manage inventory, and forecast performance.

Retailers have a preference for looking at as much information as possible on a single page and have sufficed with data often represented in tables. Development of desktop reporting tools has facilitated access to and manipulation of the data improving customization but causing “analysis paralysis”.

• What representation of data supports making quick connections leading to quick decisions?
• Can further insight (intelligence) be revealed beyond the standard performance measures?

Dashboards
Retail analysis is predominantly time-series orientated and is well suited to dashboard visualizations.

“A dashboard is a visual display of the most important information needed to achieve one or more objectives, consolidated and arranged on a single screen so the information can be monitored at a glance.” — Stephen Few

The concept of a retail performance dashboard is appealing but an effective one has been elusive. I’ve been subjected to car dashboards and control panels designed by developers which offer no value or relevance to how retailers analyze their business. The ideal retail dashboard would provide visualizations of the integral measures and dimensions (sales, margin, inventory) with a strong degree of interactivity offering insight into each level over selected time intervals.

 

 

 

This sample dashboard from VizCandy (Fig 1 ) illustrates the concept. Though several visualizations are shown, the neutral palette keeps it from looking busy. I like it most for the functionality of access to information, normally distributed among several reports.

 

 

 

 

 

A bar chart (Fig 2) from the Lunametrics blog samples how to create year over year comparison. Most retail metrics involve current versus last year or budget. This bar chart includes the future months allowing for an easy comparison of trend and forecast.

 

 

 

The area chart from Tableau Public shown below (Fig 3) depicts part-to-whole which can apply to the categories or stores. A line graph from the same dashboard employs multiple lines to reflect locations or merchandise categories over time. Here, % last year is graphed but could be budget or $ sales.

Methods
Tableau Public, the free version of the leading visualization application, was used during the lab. Tableau supports simple visualization, or “viz”, creation using limited technical skills. To learn how to produce more complex vizzes, one could consult their online video library or seek third party training. To create the visualizations and subsequent dashboard I first considered possible datasets. I imported a “reconstituted” table of 12,000 rows containing sales statistics by date(day), location, category, and brand ranging from August 2012 to June 2014. Additional calculated fields could be generated from the existing data. Since the dataset was previously used it was already “clean”. Minor format adjustments were made in Tableau.

I then browsed through Tableau’s online gallery searching for vizzes with time-series representations which included multiple dimensions and comparisons to prior periods and/or budget. Three are highlighted above.

My goal was to produce vizzes that were aesthetically impactful while revealing information about the underlying data. I created multiple line graphs, bar charts, and tables representing day, week, and month time intervals and store and category details. I learned core features of Tableau through adjusting the detail levels and formatting each viz as well as consulting training videos and viz blogs.

Next step was to select and add vizzes to the dashboard by considering the information “story”, layout, colors and other formatting. All of the vizzes were converted to a similar Color Blind palette which is relatively “quiet” but still adds depth and interest.

 

Results
The dashboard below includes visualizations reflecting common retail analysis for day, week, and month. The bar graph provides the clearest representation of the sales data including year over year (YOY) calculations in the tool tip and anticipates future months. The line line graph and area chart illustrate sales patterns over time at the store and brand levels, respectively.
[iframe src=”https://public.tableau.com/views/Lab2_86/RetailDashboard?:showVizHome=no&:embed=true” width=”90%” height=”600″]
Discussion & Future Directions
Producing the dashboard of retail dreams proved a challenging a task despite extensive experience with retail analysis. Though a respectable first attempt using a new tool, revisiting the questions posed at the start of the lab report, the dashboard and its visualizations fall short of offering additional business intelligence beyond the standard tool kit.  Some notable opportunities for improvement:

• The vizzes should be more closely related and integrated through greater interactivity. The time periods and levels appear disjointed. A better evaluation of the dataset and specific objective for the dashboard will make results more impactful.
• The line graph and bar charts need a longer time window to reveal meaningful patterns.
• Retail analysis like any revenue generating business is strongly focused on comparisons to budget. The dataset was missing budget figures so I left them out. For future lab work, it would be beneficial to enhance the sample dataset by extrapolating and adding where useful.

Potential future directions for this dashboard include creating a broader Tableau “story” by:

1. including measures and dimensions for inventory and margin and expanding interactivity to support drill down of multiple views of data
2. adding external data (weather/stock market) to measure possible influences and offer more insight

References
Few, Stephen (2005). “Effectively Communicating Numbers: Selecting the Best Means and Manner of Display” ProClarity
Few, Stephen (2006). “Common Pitfalls in Dashboard Design” ProClarity
Resources (n.d.). Retrieved from https://public.tableau.com