It’s a rare genius who can stare at rows and columns of numbers and see the underlying patterns. Most people do better with visual presentations, for example, graphs, that illustrate the patterns directly. But creating effective data visualizations is difficult. In fact, a small army of visualization gurus spends their time telling the rest of us how we could do it better. While this can sometimes be annoying, they are often right.
Much of the problem lies with users themselves. Few people have been trained in visualization techniques. But common tools such as Excel add to the difficulty by offering limited capabilities and being hard to use. Nor do they provide much help in choosing the best approach to a particular problem.
Seattle-based Tableau Softwareáaddresses both software and user skills. Tableau provides a rich set of visualization features, makes them easy to use and offers automated guidance in selecting techniques. The system is intended primarily for data exploration rather than formal presentations, but can handle both.
Users begin a Tableau session by connecting to a data source. Unlike many analysis tools, which export the data into specialized structures, Tableau leaves the data in the original source system. This eliminates the need to run file extracts or create predefined data structures, which may not include all the needed information. But it also means that performance depends on the source data engine.
Technically, Tableau issues queries against the source data and stores the result in memory. It then works with the data in memory, only issuing a new query if the user requests new information. This means that redrawing the currently loaded data is very fast; but adding a new element may take considerable time if the underlying source is large or runs slowly. Users can also save a copy of the loaded data, so they can return to an analysis without re-querying the source.
Tableau issues queries that are beyond the capabilities of standard connectors like ODBC, so it must write its own connectors for each data source.
Available sources are Microsoft Excel, Access and Analysis Services; relational databases Oracle, DB2, SQL Server, MySQL, PostgreSQL and Firebird; multidimensional DB2 OLAP Server (formerly, Hyperion Essbase); and comma-delimited text files. A Netezza connector is under development.
Each analysis can work with only one data connection at a time, so it’s not possible for the system to combine an Excel spreadsheet with an Oracle table, or even two Excel spreadsheets. It does let users join tables within a single source. These could be database tables or several worksheets (tabs) within an Excel spreadsheet.
Once it connects with a data source, Tableau displays the list of available elements. It automatically classifies numeric elements as measures and everything else as dimensions. Users can reassign elements if they are misclassified. Users can also create calculated elements and define filters to control the data used in an analysis.
Preparing an actual analysis simply requires dragging the elements onto slots, called “shelves,” for rows, columns and measures. Tableau does everything else to build the chart, using its own rules to determine the format. Again, users can override its choices if they wish. Additional shelves let users specify how the measure values will be displayed, with options including text, color, size and shape.
Different measures can be shown with different attributes: for example, size might indicate the number of customers in a given category while color indicates their profitability. The system allows any number of row and column dimensions, so the ultimate result of a complex analysis is often a multi-level cross-tabulation where each cell contains a multi-element chart.
Personally, I find these charts-within-crosstabs, or “small multiples,” difficult to read, even though they are much loved by visualization experts. But Tableau is less intended to produce one all-encompassing image of a data set than to support step-by-step exploration through a sequence of much simpler charts. The developers’ stated goal is to create a new graph with one click, which is exactly what happens each time you move a data element or change a display method.
This means users can view an image, formulate the next logical question, and then answer that question by making a simple change. Since some paths will lead to dead ends, the system provides unlimited levels of “undo” to back up in their analysis. It also lets them bookmark a particular configuration to use as a later starting point or display in a presentation.
Tableau shares many features with conventional business intelligence systems, such as drill-down via filters, selection by highlighting cells within a chart; calculated measures or dimensions, annotations; cut-and-paste into Excel, and combining multiple charts on a single dashboard. Even the drag-and-drop method of building multi-dimensional reports is fairly common. Setting Tableau apart is its simple creation of graphical – as opposed to tabular – reports, combined with built-in intelligence to recommend the most effective formats for those reports. These recommendations, available at any point through a “Show Me” button, were consistently effective – and sometimes quite unexpected. Naturally, they follow the best practices defined by the visualization gurus.
The other attribute that sets Tableau apart from most business intelligence software is its price. Many of the major systems like Cognos and Hyperion are aimed at enterprise deployments and charge accordingly.
Without a database of its own, Tableau is a tool for individual users, and it is priced like other personal productivity software: $999 to $1,799 for a single copy (higher prices allow access to relational databases). There are discounts for bulk purchases. The software runs on Windows 2000, XP and Vista.
Tableau was based on data visualization research at Stanford University and the 1.0 version of its product was released in 2005. Now at version 3.0, it has been licensed to about 10,000 users, including copies sold through an arrangement with Hyperion as Visual Explorer. The company says most users are general knowledge workers, such as marketing managers, while about one-third are data analysis specialists.
