Coy Yonce's Blog

Every employee within an organization requires information to do their job effectively.   So, when considering the types of users for a Business Intelligence deployment, everyone must be considered.  This includes the business analysts, power users, casual users, management, and the executive suite.  This wide span of users also covers all departments within an organization (e.g., Human Resources, Operations, Marketing, Legal, IT).  It should also be considered that the role for each of your users can fluctuate as they perform their day-to-day jobs – especially within smaller or more flexible organizations.  John Brookmyre presents this very well in his post from December of 2008. The key point from John’s post that really interested me was this:  “…BI interfaces could be improved if we designed them with the aim to empower users to move freely and use all of the functionality available”.  This is one of the critical issues regarding how Business Intelligence vendors can make our products more pervasive.  Our users are intelligent.  They understand the data of their business and most of them are naturally curious to follow an insight to its conclusion.  In order for BI vendors to help organizations make BI more ubiquitous, which would allow insights to be made and stories to be told by all users, our products and workflows need to be simpler and we need to provide interoperability.  In addition, we should expose functionality only when it is necessary.  This means presenting the novice user with basic functionality and gradually exposing more functionality.

Simplicity

Simple, but functional is key to any software interface.  When thinking about the design of Business Intelligence software, it is important to consider the business scenarios in which the software will be used.  Ultimately, BI software will be used in an effort to ease decision making.  As such, if we can make the interface simpler for the end user and content creator, then we can make the decision making process faster.  For example, SAP BusinessObjects Explorer allows end users and analysts to perform Google-like keyword searches for data.  This data is then returned in an iTunes-like interface.  Users can then take the data that they find interesting and analyze it within the Explorer interface.  This is a simple tool that does not get in the way of the business process.  QlikView takes a great step towards simplicity by allowing a user to download a full copy of their product within minutes.  This, combined with the ability to quickly create content within QlikView via immediately available demos, makes it easier for BI software to proliferate. Tableau takes this one step further by allowing a user to download their professional edition without even registering.  How does this make the decision making process easier?  The software is easier to install, easier to learn, and provides a quicker means to analyzing information.

Interoperability

Allowing the products within our suite to interoperate is also important to the proliferation of BI.  Currently the more established BI vendors have products that fit specific purposes: report writers for corporate reporting, dashboard builders for consolidating information views, analysis tools for slicing and drilling into information, and the list goes on…  Allowing content to be transferred throughout these tools not only makes it easier for the user to take advantage of the features of the entire product suite, but also allows more users to put together a story from the information they are viewing.  For example, if I were to build a document within SAP BusinessObjects Web Intelligence, I am pretty much stuck in that format – unless I want to rewrite the report again in a different product.  I should be able to perform my query and analysis in Web Intelligence, push part of the information out to Crystal Reports as static content, and then push other parts of the document out to Xcelsius to start building a consolidated visualization to tell a story.  The content created within each product in a BI suite should be portable and easy to push into other products.

QlikView has taken an interesting approach to this problem by creating a single content type.  For a new vendor in the BI market, this is a great step towards presenting themselves as easy to adopt and easy to use.  There is no struggle to determine which content type is best for a particular need and no need to give users access to multiple tools.  It is all the same tool.

Gradual Release of Functionality

Also important to the pervasiveness of BI is how we expose functionality within the product.  Imagine if you were to open up a product that you had never used before and were presented with all of these cryptic menu items and buttons.  How quickly would you move on to find another product that looks easier to use?  A BI product that is meant to be used by all within an organization should present its basic functionality within the immediate user interface and allow more advanced functionality to be discovered.  For example, when you open Excel, you can immediately see the purpose of the menus related to specific high-level functionality: insert, page layout, formulas, data, review, and view.  It is clear from these labels what I will find as sub-menu items.

In addition, the buttons with which the user is immediately presented are pretty clear.  For example, the font, alignment, and number formatting controls are intuitive.   As another example, we can consider SAP BusinessObjects Web Intelligence Rich Client.  When you open this tool, the first thing you are presented with is a menu allowing you to choose to create a new document or open an existing document.

If I choose to create a new document, then I’m ready to start within two additional clicks.  If I chose to create a new document against an SAP BusinessObjects Universe, then I am presented with an interface that allows me to drag objects exposed within the Universe into my query in order to create my document.

This interface is fairly intuitive and the text helps the user figure out what to do next.  There are some elements that are unclear. The text at the bottom that states “Display by objects” really means nothing to a user that has no understanding of what an object is.  I like that the “Scope of Analysis” section of the interface is not displayed by default.  This should be considered an advanced type of functionality that the user can expose once they understand what it does.

QlikView also provides a fairly intuitive interface.  The first screen with which a user is presented is a highly intuitive set of  examples and demos.

Also, the high-level menus are easy to understand – with the exception of “Selections”, “Bookmarks”, and “Object”.   The buttons presented are similar enough to Microsoft Office that any user should be able to figure them out quickly.

QlikView does a great job of allowing their users to immediately build content by making the interface easy to follow and presenting embedded demos.

Tableau is another great example of gradually exposing functionality.  The first page within the Tableau interface presents with the options of opening existing data sources, connecting to a new data source, viewing samples, or viewing training videos.

After opening one of the sample data sources (Coffee Chain), the user is presented with a fairly intuitive interface for creating an information view (in this case, a report).    I can quickly create a report by dragging a dimension (e.g., Product Type) into the “Columns” section and a measure (e.g., Sales) into the “Rows” section.  While building the report I get an immediate visualization and the application even detects the best chart for displaying my information.

Not only is it simple to build a report for a novice user, it is also easy to add filters and change the properties of the report.  It is quite simple to change their suggested information view into another format by using the “Show Me!” button at the top and selecting a different visualization type.  This is made even easier by the fact that the application hides visualizations which are not relevant to the data being viewed (notice the greyed out visualizations).

As can be seen from the examples above, each of these products makes it simpler to get started with information viewing and analysis, but only exposing a basic level of functionality at within the immediate interface.  The more complex functionality is accessed within sub-menus or via right-clicks.  Over time users will become more experience with the products and will start to investigate other functionality via exploration or reading documentation.

So, what’s the point?

The objective of this post was to higlight specific examples of simplicity, interoperability, and a tiered exposure of functionality.  The end goal of building our BI interfaces with these ideas in mind is to allow more users to take advantage of the software without being overwhelmed and without having to ask for help.  In the end, this will allow BI to become more pervasive and will enable more stories to be told based on the data within an orgnization.  This will allow those organizations to really see what their data is telling them so that they can improve their processes and become more efficient at doing what they do best.

Being able to tell a story with Business Intelligence content is a must for any organization.  It is this story telling ability that allows the organization to realize new insights and then put the appropriate context around the information to improve processes or decision making abilities.  It is also an iterative process.  I may gain an insight from an analysis session and present this as a story to several departments within my organization.  Those departments will then take my story, do their own analysis on the information within their own contexts, and tell their own story.  The end result is that the organization is unifying around stories that attempt to understand what their data is telling them and then collaborating on improvements to achieve more efficient business processes.

I would argue that this level of collaboration can only be achieved if Business Intelligence is provided to everyone within the organization.  Only then will employees have the power to determine these insights on their own without waiting for the problem to get bigger or for someone else to realize the problem exists.  As Business Intelligence vendors we have been saying for years, we want to make our products ubiquitous…but how?

Cindi Howson posted an article in February of 2008 talking about how we should achieve ubiquitous, or pervasive, BI. She found that, at the time, only 25% of workers were using BI.  I would venture to say that this number is not much higher today.  I’m sure that there has been some increase due to the focus on government audits and a larger understanding of the importance of data, but overall, the number probably has not moved much.  The reasons are ultimately still the same as Cindi points out in her article: companies need to realize the value of the data they have, they need to deploy the tools to more than just analysts and power users, and they need to provide the rights tools to the right workers.  So, why haven’t companies made more progress towards these goals already?  Largely because the products they are using do not make it easy.  As BI vendors we need to do more in order to make BI easier to adopt for all users within an organization.

Within the next three posts, I will focus on the following ideas with regards to ubiquitous BI:

• Creating simpler, interoperable products that start with basic functionality and expose advanced functionality when needed
• Ensuring that our products are enjoyable to use by taking advantage of collaborative techniques and our community of users
• Focusing on making our products embeddable

The goal of these posts will be to explain how BI vendors can further enhance their existing features along these lines and learn from other software that utilize these ideas to create easy to use and easy to deploy products.

The objective of Business Intelligence is to improve the decision making process. This is true whether it is a large organization or government institutions being transparent with the public. The many different vendors that provide Business Intelligence software assist in this effort by making it easier to discover, analyze, consolidate, and share information.

For example, SAP BusinessObjects, my employer, has recently launched a product called SAP BusinessObjects Explorer which is used to discover information.   We also have Web Intelligence which is a product used to analyze information and Xcelsius which is used to consolidate information.  IBM Cognos also has the ability to do analysis and QlikTech provides information consolidation.

When organizations utilize Business Intelligence to gain insight from information, they are typically doing so via reports, visualizations, and dashboards.  Unfortunately, the developers of these types of information displays typically do not design them with a story in mind.  Sure, they present information that can be monitored and through which insights can be derived, but it doesn’t truly tell a story.

For example, take a look at this dashboard.  Aside from utilizing unnecessary three dimensional graphics and lacking information necessary to make decisions, the information that does exist is really just there to be monitored.  You could derive minimal insights from the display; however, you can’t really tell a story.

Image First Seen on Stephen Few's Blog

The ways that people usually tell stories with data is either by presenting it in the form of an interactive visualization, a presentation, via documents, or on websites.  In this way, they are able to mix the graphics with explanatory text to add context to the information being display. For example, using an example from earlier post, why is it bad that customer satisfaction is down to 82.31%? What does this mean to the organization? There is no way to tell from the speedometers presented below.

It is important to ensure that your Business Intelligence content not only benefits the organization in how they monitor business processes and performance, but also provides the ability to tell a story when an individual or multiple individuals derive previously unrealized insights.

Why is storytelling important?

If I work in HR for a company and I notice a trend that our turnover rate has increased over the past three quarters, then I have discovered an insight.  I would obviously want to take this to the next step and analyze the information to determine the root cause behind the turnover.  I could discover through my analysis that the turnover started increasing when a new training program launched within the company.  Now, I need to perform further analysis to see if there is really a relationship or if this is just a coincidence.  If there is a relationship, then I need the ability to tell my story about how the modifications in the training program are causing employee satisfaction to decrease and thus impacting company productivity through turnover.

Based on the example above, I have followed, what I feel to be, the typical scenario for an end user within an organization.

1. I discovered something interesting – an insight
2. I analyzed information available to me
3. I came to a conclusion
4. I wanted to share this with those within the comany that could change a broken process

In addition, I think it is important to include another important aspect to this scenario – collaboration.  Once a user has derived insight and they start to analyze more information, they may need help from another user that has access to different information or they may want to run their hypothesis by other members of their department.  In this way, they are collaborating with others to build a story.

Now, can you imagine an employee in your HR department performing this analysis and storytelling through your currently implemented BI software? It is probably difficult to envision as it is not easy to do with most of the BI suites available.

Those of us that work in BI need to start thinking about how we provide our end users the ability to tell a story.  These stories are the real way to shine a light on ineffective processes and complicated cause and effect issues.

Over the past few years, I have been guiding customers on designing pre-defined and, in some cases, dynamic drill paths for their BI Deployment. When it comes to pre-defined drill paths, most deployments opt for either providing content to content links, establishing a pre-defined hiearchy of data, or just allowing the user to drill through the actual data in the datasource (think OLAP). For dynamic drilling most deployments implement a way to drill outside of a pre-defined drill hierarchy or, again, allow the user to drill through the actual data in the data source.

After seeing the following image, I thought it would be interesting to learn more about the analysis being presented in an effort to see how it can be applied to Business Intelligence.

This image presents a way of looking at the structure of a piece of music based on Schenkerian analysis. A goal of this type of analysis is to reveal different hierarchical levels of a musical work.  It is a reductive form of analysis that attempts to show musical interactions on the surface of a piece (i.e., the piece itself, represented by the notes that appear on the sheet music).  Music theorists calls this the foreground, or vordergrund.  If one then removes certain notes or figurations that seem to be more decorative than structural, a middleground, or mittelgrund, analysis is created.  Depending upon how much musical material an analyst considers to be decorative, one can have an active middleground or a more sparse middleground.  The ultimate goal of the analyst is to reveal the background, or hintergrund.  This is the most reduced, skeletal layer and shows the basic structure of the piece.  By identifying these layers, an analyst has revealed how the composer has embellished a basic background structure to create a distinctive, original piece.

How does this apply to Business Intelligence?

This type of analysis is also important within a business environment.  Based on my role, I may want to see basic information about how my company has been performing year over year in terms of revenue.  If I see a trend or stagnation, then I may want to look further to determine the cause of this trend or stagnation.  This then becomes another level of analysis that needs to be performed.  This is demonstrated below by showing three views of the same data that get more detailed as you go from the first view to the third view.  When comparing this to the example of Schenkerian analysis, we are actually going in the opposite direction.  Our analysis is getting more granular as we drill from the first view to the third view.  In Schenkerian analysis, it is important to reduce the amount of detail as you go further along in your analysis.

As we look at the initial view displayed in the presentation above, we can see that our revenue drops from 2001 to 2002, 2002 to 2003 and 2007 to 2008.  For our session, we want to present an analysis of the drop in revenue from 2007 to 2008.  As such, we have provided another view showing the performance by line of business in 2008 and a third view showing the performance of a specific product line, Music / Movies.   This presentation of information is providing a guided analysis, or pre-defined drill path, for the user.  In addition, the staggered display of each tier of data along with the headings at the top allow a user to pick out the view that is most important to them based on their needs.

An analysis session

If I were to build this as an actual BI tool for information analysis and presentation, then I would provide the capability to combine these views into a single “package” and then publish them on a portal for others to see.  When those users opened the “package”, they would see the information that I thought was important to display for my analysis of why there was a drop in revenue from 2007 to 2008.  They navigate from view 1 to view 2 to view 3 (and back up) by using the arrows at the bottom right and top right of each view.  Alternatively, I would also build in the functionality for a user to change what was contained in view 2 and view 3, by selecting a data element in the prior view.  In this way, a user could perform their own analysis without having to build the entire “package” from the beginning.

For example, say I wanted view 2 to actually display the performance by line of business for the year 2007.  To do this, I would simply click on the cell that contains 2007 in view 1. If I wanted the performance by line of business view to show a span of multiple years, then I would select multiple years. The view would automatically update to show this new information.

Reversing the analysis

We could also perform this analysis in reverse, more in-line with the purpose of Schenkerian analysis.  In this manner, we would provide the detailed information in view 1 and then allow the user to reduce the amount of detail that they see as they “drill” from view 1 to view 2 to view 3.  This concept of a reverse data “drill” would allow me to clear out the data clutter in my information display to see the larger impact of what appears to be a small problem.   For example, in view 1 above, I see that my revenue from CDs dropped significantly from Q3 to Q4 of 2008.   If I “drill” up from view 1 to view 2, I can see that the larger impact is that my revenue from “Music / Movies” dropped by 16,500.o0.  If I “drill” up to view 3, I then see that my overall revenue for the year 2008 was down by 69,700.00.  As “Music/Movies” was down by 16,5000, this means that this one category was roughly 25% of my total revenue drop in 2008.  In the end, this means that I need to focus more on my “Music/Movies” sales or find a way to compensate for the lack of sales in 2009.  Either way, it shows that the “Music/Movies” line of business is a critical one that needs to be tracked in 2009.

Benefits for this type of information design and analysis

• It is easy to see which view is applicable to me as the view headers and table headers are visible across all views.  This means that I can quickly make decisions on which view to use for my analysis.  This ultimately leads to faster decision making.
• It is easy to switch between views as I only use the arrows at the bottom and top right corners of the screen to see more detailed information.  There is no need to click on a hyperlink to take me to another content type that has to load.  Each view is already populated.
• It is easy to change the context of my detailed analysis, but clicking on a different data element in the prior view.  This allows me to populate the detailed view with data that is more relevant to me without having to enter a design panel.

I’m sure that there are some drawbacks to this type of information design and display. I am also sure that I have left out many benefits.  I would love to read some other views and opinions as this is just a rough sketch of an idea.  This type of information display or BI tool may already exist.  If so, I would also like to hear about it so that I can review how it works.

NOTE:  Thanks to my wife, Tammy Evans Yonce, for providing me with a quick tutorial on Schenkerian analysis and for showing me the analysis presented in the first image of this post.  I may have not done Schenkerian analysis justice in my interpretation. For that, I apologize. I will make every attempt to correct where I have mis-stated.

I worked with a  customer recently on some stability problems with a software package. They presented me with some charts showing the instability of the system from January to March of 2009. Some examples of those charts are presented.

After looking at the three charts above, I could not help but wonder how I would present the data so that it would be easier to understand and provide more detail. My representation of a dashboard providing the same information – plus a bit more – is below.

This new representation of the information not only provides the previous insight of number of outages per month per server, but also provides the reader with the number of outages per software service per server and a yearly trend line beneath each month showing the trend for the year to date. The blue dot at the end of the yearly trend line is the current value. The reference line would be a number that indicates an acceptable number of outages so that the reader could understand the level of deviation.

Here I have provided an extraction of just March to show the details displayed for each month.

You could easily expand this dashboard out to include every month for the year, print it out on A3 paper and take it into meeting to demonstrate the failures of a software suite compared to other comparable suites deployed in the same environment. This would make quite the point.

The image the right provides an example of a dashboard that uses both text and speedometers to provide updates to an audience on quarter-to-date metrics centered on customer support. This particular dashboard is quite redundant as the text at the top says the same thing as the speedometer series underneath. The only differences are that the text at the top does not show the target value and the speedometers do not show the change from the last update. In addition to being redundant, there is no indication of when the last update occurred. So, we have nothing to which we can compare our new quarter-to-date numbers.

The image below presents a redesigned version of this dashboard using a table structure. The display is much cleaner and easier to understand. The sparklines to the left display the trend of associated metric. This gives us an idea of how we are doing compared to the last few times the number was tracked. In addition, the table format is easier to read as it consolidates all of the necessary data into one place rather than having to look at both the text and the speedometers in the previous dashboard. The sidenotes provide the audience with an understanding of what they need to do in order to improve the numbers so that the targets can be met. This is a great improvement over the previous dashboard.

The only thing that would make this version of the dashboard even better is to display the amount of time that is covered in the sparklines. For the purposes of this dashboard, the appropriate time period should from beginning of the quarter to current date with a red dot at the end of the sparkline for the current data value.