By Craig Larmer, Lead Consultant at Stellar
There are a lot of things that need to be in place to provide world-class business intelligence and analytics capability to your business.
One of the key elements is the logical interface between people and the data. This comes in the form of a data model – sometimes called the semantic layer.
I’m not talking about the user interface for a BI tool or any particular technology for that matter. Rather, the data model is the logical layout of information in the organisation.
It is critically important that the data model is optimised for your analysts to use. One of the fundamental activities in a BI system is translating data that has been collected by a bunch of applications spread around the organization and bringing it all together and into that data model. This gives the analyst a view of what’s happening in the “Real World” that they can understand, and that can be used as a basis for making informed decisions.
The difficulty in doing this stems from the fact that applications “talk” to each other in ways that are optimised for machines, and which are quite different from the way humans talk to each other. So it’s quite an undertaking to build a system that does this translation automatically, repeatedly, reliably and quickly.
To make matters worse, those applications are generally not quite up to date with what’s happening in the business. So to get a more representative view of the world analysts manually add another layer of adjustments and merge in datasets not held in applications or gathered from outside the organisation.
It all gets pretty tricky pretty quickly.
Worth the effort
Fortunately, creating an appropriate data model for your business pays dividends.
A data model that is shared and has an enterprise view of the organisation quickly becomes the most valuable asset in the BI solution.
As well as making sense of that machine-speak, it provides a layer of insulation between the applications that come and go over time and provides a stable platform for analysts and report developers to easily work with.
BI practitioners over the decades have come up with a variety of approaches to develop the model and a variety of structures to handle the types of data available to the organisation. Be it third normal form, dimensional modelling, data vault or any number of others, they all have their own pros and cons.
BI tools vendors have come up with a variety of technologies to help the process too, although most of them are trying to lock you in to their toolset! It’s a delicate balancing act to get the most out of the tools while being able to share the model over a wide set of tools.
I’m sorry to report there is no one-size-fits-all approach and generally an organisation needs a variety of structures and models for different areas of the business and types of analysis. To keep this under control you need great governance and processes to make sure that point solution data models don’t crowd out the enterprise view.
What I can’t emphasise enough though is that the primary goal of any BI solution should be establishing and building the data model to cover as much of the organisation as possible, in a way that the greatest number of analysts and people in the business can use it.
It’s an investment that you won’t regret.
