There is a potential fallacy that we hardly ever address: Isn’t there a risk that business intelligence systems – seemingly – become just too good? That they describe the market, the customer etc. so accurately that business users will consider this information as reliable as first-hand experience and forget that they can only see what they are being shown? Maybe these users should receive frequent warnings that say something like “Ceci n’est pas un pipe.” These are the famous words that the Belgian surrealist artist René Magritte wrote under his painting La Trahison des Images, which shows – a pipe.
The comment is intended to remind viewers that they are looking at a piece of art, not the actual objet. They can’t smoke this pipe, no matter how well it’s depicted. They can’t even tell what the actual pipe really looks like or whether it has existed in the first place. This is an essential idea to keep in mind when dealing with information systems. It’s tempting to believe that you have full and valid information at your hand, especially in highly endemic systems. But is it really true?
During the recent credit crunch, some people in the financial markets found out that it wasn’t. They thought they were Masters of the Universe when they were, in fact, Boys in the Bubble. Business users, therefore, must rate the informative value of their various indicators. At the same time, they absolutely need to be able to rely on the quality of the underlying data.
Ensuring that data quality is at a high level is one of the most important functions of any data management unit. It is an interminable process that sometimes gets underrated. The backbone of broader quality initiatives is a sound master data management. Master data, as opposed to transactional data, describe items like product and customer names, addresses etc. These data will be stored and reused in the long run but very often change over time.
For example, people change their names or move their residence to new cities. Here they may make a new contract with their previous telecommunications provider at a different point-of-sale than before and may be registered as a new customer, not as the one already known. Thus the quality of the master data deteriorates, unless there are organizational processes put in place that make sure that the customer data are permanently reconciled. Without such procedures, any analysis of the provider’s customer wins and losses would be flawed from the start. So let’s not forget: Ceci n’est pas un client.
Obviously, the EDW approach of storing all data in one place, organized by a logical data model, facilitates quality-ensuring procedures of master data and other data. Still it’s not a trivial matter. The logical data model must be flexible enough to capture unexpected categories like, for example, official (as opposed to biological) birthdays in some countries. In retail, product data must capture temporary changes in package sizes, which are typically used during special offers, to make any meaningful analysis of the sales results possible. And just to return to the financial sector, detection of money laundering would be a great deal harder if banks couldn’t figure out whether a single person own several accounts within (and beyond) their organization. An effective master data management means that the data betrays the fraud, not the bank. And that’s my idea of la trahison de données.
