There’s no doubt about it – Tableau is synonymous with best practice data visualization. Explain a problem to any consultant about visualization and tableau’s name will enter the conversation (well in most independent cases). Tableau’s approach is targeted at visual investigation – allowing the user to work with with data in real time and draw conclusions from it. An approach that was the original intentions of OLAP technology and like OLAP tableau allows the developer/user to define a model within the (tableau) workbook and create visualizations from that model. This is an important call out because it is not dependent on a single table import and can combine data from several disparate sources (I’ve heard some consultants say that you can only display data from a single table).
So with this in mind, what do I like in the next release (V9)? I was originally going to publish this after the Singapore conference however, a recent brief at the BBBT revealed some very nice features that are definitely worth a call out. Surprisingly, the features are targeted at the user experience and not so much at visualization improvements
Importing/Connecting to Data
There’s a new connector for SAS and R data files which may be great for passing data along a work stream. But from my point of view the most useful improvement is the concept of a data interpreter. This investigates a source (Excel Sheet) to look for and interpret tabular data (or a data set) that is the true intention of the import. This has several nice consequences of the feature. You don’t have to have data start cell A1, infact it could be a ‘report’ somewhere in the sheet. Headings (non tabular cells) are striped as are columns, so you could import source systems exports that are not in a strict tabular format. The import can manage formatting and cell merging. For example column headings that appear over 2 lines are imported without issue. Finally, the interpreter applies its own delimiting function. Fields that are delimited (or even the import of a delimited file) can be identified and broken down into multiple fields.
Is this a mature ETL feature? Certainly not. There’s practically no way to control how the the feature works but that’s not the point. The benefit from my point of view is that there is a somewhat intelligent way to use other sources and the ‘month end’ report pack just became a lot more functional.
LOD Expressions & Windowing Functions
Proportions (and trends) are the bread and butter of analysis. We naturally classify items into groups and show significance of items to the group by its proportion to the group total. Of course the calculation of the ratio requires 2 values (detail and group total) and the group total is an implicit windowing problem. In order to derive the total, we have to define a range of data to perform a calculation over (that is window) and this is a usually a problem because there is an associated hierarchy between an item and the group.
In MDX we can define the calculation by reference to the current members parent (provided that a hierarchy exists), however, in reports its not so easy and the most common implementation is found in control which allow this as part of there feature (pie charts are a natural fit for this). Unfortunately, the values (actually only the ratios) are artifacts of the control and unavailable outside its scope.
Tableau solves this issue in 9 with a scoped calculation called a Level Of Detail (LOD). Its a really impressive method to define the scope of an aggregation. This is a really cool addition because (in addition to ratios), those measures can be reused through out the workbook. You can define a calculation with respect to the scope of data that’s shown in a control.
The only consideration i have not tested is solve order …
While we are on the subject of measures, there’s also a new in-line editing feature that allows you to define calculations in the worksheet. You just start typing the (intelli-sense enabled) formula in rows or columns and the calculation is added to the sheet. Then, If you want to create a measure for the workbook, you can simply drag it to the measures tab.
There have previously been two options for sharing workbooks in Tableau. Firstly, you can run up your own instance of Tableau Server which was the enterprise web server. For those not familiar with it, I would liken it to Share Point or Report Server for Tableau workbooks. Something that you need to engage the IT department on to get up and running. Secondly there was Tableau Public – a no security implementation of Tableau Server that is open to everyone – not really a reporting solution for a department.
This leaves some organisations that I speak to in an interesting predicament. They like the desktop tool but can’t share the workbook because of challenges with their IT department (the direct sale of tools to business users seems to exasperate any ill feelings between IT and business doesn’t it?). Tableau Public is just not an option and a Server wont get past IT or the business.
Enter Tableau Online. A secure pay by the user hosted service of Tableau Server. Actually, thinking about the offering, I’m surprised that it has taken this long to eventuate because (in hindsight) it seems like such a sweet spot for the sales model (well at least the clients I speak to). Its a secure, private implementation of Tableau Server that’s hosted by Tableau – a very nice offering indeed.
Another very nice feature included in the new server (all versions I believe) is search functionality. Workbooks can be tagged and searched within the site. This type of functionality seems to be the new norm for finding what you want.
If your not interested in the server, you could try tableau reader for some method of distribution and collaboration. This is equivalent to a PDF reader (for tableau files).
So not much new (well perhaps advertised) in the visualization space. There are some nice features for grouping (lassoing) data points in maps but for the most part, I see the improvements relating to how the user interacts (and perhaps what they can define) with data.