Saturday, November 21, 2015


I attended the recent Tableau Conference in Vegas where the Tableau team debuted new features and functionalities in Tableau 9. Since I was still using Tableau 8 at that time, the new features warranted a download of the new version 9. That said, I'm currently using both versions since the company still hasn't integrated Tableau 9 server functionality and the only way to access dashboards built using the new version would be through Tableau reader/desktop.

Recently, I had a request from a client to build a dashboard that they are currently receiving from one their partners - a Demand Side Platform (DSP) and the client loves the functionality there. I set out to build the dashboard on my end and created a look-a-like in a few hours.

Here's how it looks (click to enlarge):

The main functionality of the dashboard would be to provide rough estimates at where the optimal spend might be for the account. This rough estimation is achieved through displaying a time-series line chart with 2 lines (commission and spend). There are a few filters in this dashboard to select one or a few campaigns, date range and most importantly an attribution methodology of fixed percentages on commissions through either post-click or post-view.

Overall, I would honestly say that this isn't the best dashboard used for the sole business purpose of determining optimal spend. However, an added benefit is that this dashboard might double as a tool to investigate outliers in the data, for instance the huge spike in the commissions chart which might be a result of the client running a promotional event.

Moving away from the above dashboard, I was working on another 3 dashboards today as part of an interview exercise and I found the experience quite fulfilling - particularly around the insights that could be derived from the given dataset.

Here are the dashboards (click to enlarge):

I was provided with a csv file and with that I ingested the data into Tableau to create the above dashboards. But prior to this there was another exercise where I had to script a SQL query given some information about the database schema in order to generate the same dataset.

The first dashboard (Bike Usage) was mainly created to look at the granularity of a bike level the number of trips that were taken and the number of days used. I had to do a simple count for the former and a simple calculated formula for the latter to convert data from seconds into days. The premise of the 3 views in this single dashboard is to inform when a bike needs to be replaced because of being worn out either from the number of trips taken or the actual usage of the bike. I left it as-is and did not go ahead with blending the two sets of data in trying to create a 'bike-replacement' prediction since there really isn't an accurate way to model this out without data for the dependent variable.

The second dashboard (Bike Movement) is probably the most informative dashboard out of the three dashboards created. This dashboard would inform where there are bike shortages and excesses at respective stations shown in the top two packed bubble charts and the view in the bottom half. The views are quite self-explanatory, which in essence is the crux of creating a good dashboard. Top-right of this dashboard is a view of a geographic map of North America showing the number of bike sharers in a specific zip code, and allows filtering by subscribers/customers. This geographic view in my opinion serves no real purpose other than to show where users have come from and I have only included it as an accessory to the dashboard.

The third dashboard (Platform Popularity) shows the growth trend in terms of the number of trips that were taken over time. One thing that stands out is the more or less consistent drop off in trips taken during weekends (troughs in the area chart) which would mean downtime and lower revenue generated. Another interesting finding is through the filter by bike station where you will be able to dive into which stations are heavily utilized and which are not. From there you will then be able to decide whether to expand the bike station to accommodate more bikes or to demolish under-utilized stations if there is focus on improving the bottom-line.

Overall a great exercise to work although it did take up quite a fair amount of time.