​

Better BI
What Really Leads to Better Business Decisions?   

 

Let’s define business intelligence (BI) as a technology-driven process for presenting information that enables managers to make more informed business decisions.  The technologies include data mining, data integration, online analytic processing (OLAP) and data visualization.  Data visualization includes key performance indicators (KPIs), reports, scorecards and dashboards — all with interactivity.  The business decisions focus on increasing revenue by identifying trends or reducing costs by optimizing internal business processes.  We also assert that BI is more effective for organic growth than growth through globalization, and note that the era of organic growth may be upon us.

​

Alternate definitions may be found.  Wikipedia defines BI as a philosophy (October 2016). Altruism is a philosophy; BI is a process.  The Gartner BI ratings seem to focus heavily on data visualization.  Data warehousing and other technologies are rated in different Magic Quadrants. Our definition covers the entire process and emphasizes the end result, business decisions, implying that better BI means better business decisions.  Within this framework, let’s explore some technological developments to see if they result in better BI.

​

The first technological development is real-time BI.  Real-time means raw data is transformed into business information fast.  Stricter definitions require this to happen with near zero latency (stimulation to response).  Looser definitions describe this as up-to-the-minute or even right-time availability.  Tools that support real-time BI are memory-resident data warehouses and query engines.  Examples of real-time BI applications are stock market transactions and internet advertising.  Faster stock market analyses enable buy and sell decisions that preempt the competition.  Real-time internet ad analysis allows for personalization, meaning ads that you have a tendency to respond to are inserted on web pages as you view them.  However, this is not BI as defined above.  We feel it is important to distinguish between real-time BI and applications that fit the BI definition provided above.  Real-time BI consists of custom applications that help automated programs make transaction-level selections.  They do not describe the end result of a business process and do not help business managers (people) make better decisions.

 

The second technological development is data discovery.  This is a subset of BI that allows users to derive the answers to business questions interactively from source data without pre-processing the data.  The important words are “interactively” and “without pre-processing.”  Data discovery is enabled by powerful memory-resident query tools, some of which can even read the source data schema, and can be relatively inexpensive when purchased as a subscription.  Data discovery is useful as an entry to BI but it does not scale.  You’re asking the memory-resident query tool to link non-indexed tables and aggregate the resulting data, both of which are processing-intensive.  This works for single users for small to medium data sets.  For complex data sets with many users, the processing is overwhelming.  The math is simple.  Memory may be 1000 times faster than disk, but if you ask it to do 10,000 times the work, you’ll get the answer 10 times slower.  In addition, you can’t really integrate data from different sources (different databases) because the link values (key values) don’t match.  Translation tables that are not in either source are needed.  Lastly, not all needed data is available in source systems.  We provide two examples.

​

Example 1:  Salesperson Performance — A key metric in evaluating salesperson performance is how much they discount the product.  Good salespeople communicate the value proposition while less skilled salespeople offer a discount.  Curiously enough, all (or most) enterprise resource planning (ERP) systems only store a product’s current list price.  This list price changes over time.  The list price at the time of the sale is not stored with the sales transaction, only the actual sale price.  It is thus not possible to see which sales people are discounting products and by how much.  A BI process that includes data modeling can add the current list price to the sales transaction as part of the daily extraction from the source system.

​

Example 2:  Sales Orders Over Time — All (or most) ERP systems store only the latest version of a sales order.  If a client orders 1000 widgets in May and adds 500 more in July, the ERP sales orders only shows that 1500 widgets were ordered and the order date is July.  Some companies want to track these changes.  A BI process that includes data modeling can detect these changes as part of the daily extraction and maintain the fact that 1000 widgets were ordered in May and 500 more were ordered in July.

​

The third technological development is advanced visualizations.  In September 2016, Dresner Advisory Services indicated that their audience rated advanced visualizations as the 3rd most important BI feature, out of 30 features.  We would like to offer an additional perspective.  BI visualizations are supposed to facilitate business decisions.  The most valuable visualization is a single number in green or red and a percentage.  The number indicates the metric value, like revenue, and the percentage indicates how the metric compares to yesterday, last year, or some plan.  The stock market is reported daily using this method.  This is not a very advanced visualization.  The next point of interest is how this metric breaks down based on some category: time, product, customer, candidate, location, etc.  This is best shown by column charts or geomaps.  After this, visualizations can get a little creative.  Funnel charts show progression through stages.  Bubble charts show multiple metrics at the same time.  Scatter plots look complicated but are good at identifying outliers.  These may be advanced for Excel but are pretty standard among BI products, so we won’t call them advanced visualizations.

​

Then there’s another level.  Please Google “D3 Visualizations” or go to https://github.com/d3/d3/wiki/Gallery for examples.  Consider one of these advanced visualizations, the Sankey diagram shown below.  Energy supplies are on the left and demands are on the right.  Intermediate nodes show how energy is converted before it is consumed.  The diagram is captivating and provides context, but what can you conclude about energy bottlenecks, climate change, and so forth?  The point is that advanced visualizations don’t necessarily drive business decisions.