It is generally assumed that business intelligence (BI) software is a one-way street – we gather information, synthesize it, and distribute it to the people that need it. However, an important step is missing, which is to then take action on that information. The reality is that traditional BI stacks don’t really do this. I was recently reminded of this point in a weekend twitter-verse conversation with Neil Raden of HiredBrains and Howard Dresner of Dresner Advisory Services – two highly respected analysts and consultants in the BI community.
Why is it that many people assume BI is poor at write-back? If the need is there, why don’t traditional BI vendors meet it? The answer is simple: They can’t, and the reason is found in the architecture of a traditional BI stack.
The first thing you will notice on this picture is that the arrow goes one way – ETL tools are used to copy data from transactional systems into a data warehouse, a semantic layer is put on top to define the metrics and measures that people will want to look at, and apresentation layer is put on top.
Now what happens when an end-user viewing a dashboard or report wants to actually take action – say send an invoice, flag an account to send to collections, or re-segment a territory? In a traditional BI model, they now must go into their original source systems, the transactional apps, make the changes, and wait for those to flow back through the cycle. The action is disconnected, the timing doesn’t match, and the context the user is in when they are viewing the information is lost.
It isn’t that traditional BI vendors don’t want their customers to take action based on their data, it’s that the entire framework that their apps are built on doesn’t allow for it. Why did they design it this way? There are a lot of reasons, but two primary reasons are 1) performance and 2) control.
Performance – transactional systems are designed for write, not reads (well, or at least batch reads) and, in the technology of the time, this meant unacceptable performance. A data warehouse is designed for bulk loading and fast reads, so this would overcome performance limitations.
Control – traditional BI stacks are built on the concept of a single version of the truth. So the idea is once the data is mapped and semantic layer defines the metrics and measures end-users can trust that is correct. These processes and technologies are centrally controlled and managed by sets of individuals inside an IT organization and there are lots of layers of control that go into changing anything. This is great if your requirements never change and your data models are consistent, but for most data uses and most business users this just means IT requests that never get satisfied and long wait times to make any changes.
Today, there are BI vendors that don’t require the strict controls of a traditional BI stack. They offer more choice and flexibility in responding to changing requirements. The ability to take action, write-back, is now a component of BI products (such as ours), that are disrupting the traditional ideas of what BI is, and isn’t.