BeyeBLOGS | BeyeBLOGS Home | Get Your Own Blog

« November 2010 | Main

December 13, 2010

Bime Tip: Decompose, Drill Through and Focus features

In the new version of Bime, you can use the Decompose, the Drill-Through & the Focus features. Here's a quick tutorial on how each one works.


If you choose to decompose then you can analyze data using a different axis (i.e. a different attribute).

The cartesian and pie charts allow you to 'decompose' your visualization, both on queries and on dashboards (if allowed in the dashboard preferences). Click on a data range and a pop-up appears, offering you this option. Click, and you can select to which attribute you want to dig down into.

For example, this pie chart shows results based on one attribute (here, 'Year(Ship Date)') - but you can focus in on one element of that (here '2005') by using decompose.

Having opted to decompose using 'Sales Channel', the chart changes to show results only by the chosen Year, which has been moved to be a filter.

You can continue to 'pivot' your data by altering the filter. Alternatively, decompose another segment of the pie and each new view adds that segment's attribute to 'filters', enabling it to focus in on that segment, and uses the new chosen attribute in 'columns'.

Thus, decompose allows you either to drill down into the detail, or 'flip' the visualization from being defined by one attribute to another.

Drill through

If you choose to drill through then you can visualize the underlying data in each query.

In the pivot table, you can use the Drill Through feature on cartesian and pie charts to see underlying data for a selection of data points, displayed as detailed lines. For instance, the following figure illustrates the result of a query for which we visualize the Shipping Cost per Region. Let us imagine you want to see the underlying data for the Region Central. To do this, click on the Central horizontal bar and then click on the Drill Through item.

Once you click on Drill-Through for selected data points, Bime retrieves & computes all the underlying data and displays them in the Drill-Though pop-up illustrated by the image just below. In our current example, the Drill-Through operation has retrieved 2100 underlying lines (i.e. 21 pages of 100 lines). Consequently, keep in mind that the more numerous the underlying data are, the longer the Drill-Though computation time.

In the Drill-Through pop-up, you can navigate through the underlying data with the 'Next' & 'Previous' page buttons, visualize a specific page directly through setting the index of the wanted page in the numeric stepper, and exporting the underlying data as an CSV file through clicking on 'CSV'. In addition, you can modify the number of lines which are displayed per page. Then, click on 'OK' to close the drill-though pop-up.

To conclude, please note that the Drill-Through feature is available only on in-memory & RDBMS connections.


If you choose to use focus then you can restrict and execute an original query of the dashboard only for elements you have selected.

In the pivot table, the focus feature is a shortcut to filter elements visually. As illustrated in the following figure, let us imagine we are analyzing the profits we realize per customer segments.

Let us imagine now that we want to restrict our analysis to the Corporate & the Home Office customer segments. For this, we have only to multi-select Corporate & Home Office, and then click 'Focus on selection'.

As a result, the current result is restrict for Corporate and Home Office, as illustrated by the following figure.

Posted by We Are Cloud at 8:15 AM | Comments (24)

December 7, 2010

Pre-packaged dashboards or ad-hoc analysis: what's the best ?

When you provide pre-packaged dashboards, people want to be able to do ad-hoc analysis. When you give them a powerful BI tool for great ad-hoc analysis..they want pre-packaged dashboards. So what is the best solution, to provide one or the other, or both? We gathered some inspiration from LinkedIn members.

First of all it is important to note that pre-packaged dashboards and ad-hoc analysis tools have different use cases. Dashboards with already in-built queries are often used for a quick overall glance at data (what and how), whereas ad-hoc analysis is used to drill down deeper into the data (why). Pre-packaged dashboards are likely to be more useful to top level management. Ad-hoc analysis may be preferred by analysts and power users.


Why pre-packaged dashboards?


Pre-packaged can bring the quickest ROI, and if this is what an analyst is looking for then there is no reason why a pre-packaged solution could not sufficiently meet the needs of the organization. Getting results to users quickly can be a top priority for some people and pre-packaged functionality can bring speed to the process. Good for general trends and KPI measurement, dashboards should contain a mix of standardized and custom KPI's, as every organization is different, and an out-of-the-box solution is not likely to suit them all.


Why ad-hoc analysis?


Members of the LinkedIn discussion felt strongly that ad-hoc analysis was essential at some point, as pre-packaged dashboards are no replacement for custom analysis. With a pre-packaged solution there is the danger of having a "false sense of security" regarding your results, because of the level of detail a pre-packaged dashboard provides, which can lead to viewers jumping to the wrong conclusions. The order in which they should be used also cropped up in the discussion - and it was established that a pre-packaged dashboard should always be used before ad-hoc analysis. Why? Since it gives everyone a common set of metrics of which to base performance on. If you supply ad-hoc first, different people will begin basing their analyses on different criteria, which makes a useful dashboard more difficult to create. The advantage of ad-hoc reporting is the flexibility it gives.




To achieve the best of both worlds, our team is working hard to build a marketplace with dashboards for your department. Regarding the issue of quick ROI, we are thinking about the pricing of each: depending on number of stars, etc.. so who better to ask than our readers and customers? For a pre-packaged dashboard, with useful KPIs, which is ready within one hour, what would you be willing to pay?

We'd also like to thank those who contributed to the discussions we set up:

From LinkedIn: Christophe Dejoie, Shawn Hayes, Dmitry Gudkov, Ralph Winters, Hrvoje Smolić, Pedro Perfeito.

From Darren Bond, Kevin, Benjamin Breeland, Robert A. Marchello, Gil Hakami.

If you have an opinion on the topic we'd love to hear it too. You can comment on this post, or access the LinkedIn discussion or the discussion.

Posted by We Are Cloud at 9:15 AM | Comments (690)

December 3, 2010

Which chart should I use and when? A guide to dashboard charts

With the advanced offerings of many BI solutions nowadays, you are often met with more types of chart than you know what to do with. But, remember, they are there for a reason : each one is useful in its own way. The trick is just knowing which one to use, and when. Hopefully this article will give you a better understanding of some of the more common charts you might be using.

To the casual user it may appear that different charts can simply be used interchangeably, and this is often the case, but not always the best solution. Experts have come up with guiding principles that can be applied to achieve maximum data visualization effectiveness. This also comes at a time when viewing data on mobile devices is becoming more and more popular, and this should be taken into account too.

Basic charts consist of:

Why are they considered basic? Because in their most basic form they measure one element against one or more quantitative metrics associated with that element.

Take the bar chart. This should be used to display one element (product), against one quantitative metric (profit) or several quantitative metrics (profit and number of items sold). Because bar charts can become difficult to read if you add too many metrics, they work better if you use efficiently the space you have available.

Time or date dimensions are likely to be better displayed using either a line or a point/bubble chart. Just as with bar charts, multiple quantitative metrics can be displayed in a single chart, assuming that the chart does not get too crowded. Adding a mean statistic line onto a line chart can also be easier to read than for example adding one to a bar chart, and can show how far a measure has deviated from the average. However it is important to remember that these types of statistics should only be added in cases where they contribute positively to the understanding of the data, and only where they are justified - there is not much point adding a trend line to 3 data points or less because it implies a trend that probably does not exist!

The point/bubble chart is in it's simplest form basically a line chart but without the lines. It plots each data point as the line chart does and just does not connect them with a line. They are useful for displaying a single quantitative metric, but can be confusing if trying to display multiple metrics, unless the software you are using is designed particularly well. Ways of doing this could be encoding by color, or size for example. Joining the points on a point chart using a curved line can be visually appealing but also misleading in terms of accurate data analysis. This is because a curved line suggests that a polynomial relationship exists - which may or may not be true. Equally turning the lines of a line graph into curves can have the same effect. Avoid it if it changes the meaning of your data!

Last of the basic charts is the pie chart. Pie charts are ideal for showing data that has a number of components and only one quantitative metric. They require that the sum of the component values add up to 100% and that there are no negative components. For data points that are very small, grouping them into a category together ("other") is a good idea, to avoid crowding and confusion. Visualization gurus suggest that the maximum number of data points to display on a pie chart is between 7 and 10, although physically you could display any number of segments.

There are other types of chart which are perhaps less basic, but equally as common in dashboard apps.

These can all be used to measure actual vs anticipated performance, current vs previous performance, and deviation from expected trends or benchmarks. They are also useful for giving a top-level overview for those that don't have the time or desire to examine data in great detail.

Sparklines are designed to show overall trends, as well as highlight the most recent data point in the context of the history of the data. Sparklines should be used to provide immediate and unambiguous trend comparisons among related data entities.

The bullet chart is common for displaying a quantitative value and a horizontal scale. This version of a bullet chart adds additional context by displaying a threshold via the black vertical line, and also encodes a qualitative measure using color. When should you use this type of chart? Comparing related metrics with a bullet chart can give you a quick snapshot of your data, so when you are looking for an instant comparison this is a good choice.

The guage uses a needle to indicate the current value of a metric. As with bullet graphs, guages make visual comparisons quick and easy. Although often visually appealing, these visualizations do not tend to use dashboard space effectively, so try to use them when space permits, and not at the expense of, say, bullet charts.

Final note

A problem you may come across in your dashboard once you have chosen all your charts carefully is that of data update. When you update data, this could skew your results considerably, and render some of your visualizations less effective. Unfortunately there is not much that dashboard designers can do about this, it is just a matter of you as the creator to test the extremes of your data set, if possible, and choose your visualizations with the results in mind.

Of course Bime incorporates all the types of chart I have discussed above, plus more. So next time you use Bime to analyze your data, I hope you are able to make the best visualization choices and use Bime to its full potential!

Posted by We Are Cloud at 10:15 AM | Comments (118)