Sometimes vendors do get it (mostly) right. Hewlett-Packard put together a brief white paper in February of this year laying out their view of Business Intelligence (BI) for 2009 (and beyond). I think that they got it largely right. Their #6 trend is black-box analytics. Below is a summary of the trend, my thoughts on whether HP got it right and what the trend may mean for HIT.
HP predicts: that complex statistical methods such as data mining and machine learning will be packaged in such a way that any front office user will be able to request multiple data elements and string them together to make a report – amongst the Gen Y set, this is what is known as a mash-up. HP didn’t stoop so low as to call it a mash-up, but this is what they are talking about when they say “…the user will go to a portal and request data elements. Provisioning will be automated rather than manual…”.
The Verdict: Sadly, inevitable. Mash-ups are all the rage these days, right up there with Service-Oriented Architectures (SOA), Cloud-Computing, and Social Networking. All of which are somewhere between the upswing from Technology Trigger to the Peak of Inflated Expectations and the downswing from this peak heading towards the Trough of Disillusionment on the Gartner Hype Cycle. Okay, maybe SOA is on it’s way up the Slope of Enlightenment – but that is for another post. Back to The Verdict. Mash-ups are a terrible idea whose time has come.
Mash-ups begin with the basic concept of a black-box with defined inputs and outputs. The black box may be an ETL routine, it may be an analytic function, it may be a graphical visualization, or any combination of these or similar snippets of code. These black boxes are then strung together, lego-like, by matching the outputs of one box with the inputs of another (a la “insert Tab A into Slot A”). Before you know it, a non-technical user has assembled a very complicated bit of code to deliver advanced analytics (vendors usually tout things like black box neural network analyses, as an example) with high-end graphical representations. End-users, young and old, love this. I hate it – and here’s why.
My objection begins with: just because you can feed data from one mash-up into another doesn’t mean that you should feed data from one mash-up into another.
We are all familiar with the apples and oranges metaphor, but I believe most non-technical data analyses fall apart at the apples-to-apples level, say McIntosh to Red Delicious. Both the McIntosh and the Red Delicious are red apples of medium size, often packed in North American lunches. However, if you’ve ever eaten them side-by-side you know they are totally different apples, what is more, McIntoshes are superb for cooking in pies and sauces, Red Delicious are terrible for either use. In short, without a detailed knowledge of the precise apple cultivar, you could very easily make the mistake of throwing Red Delicious apples into your grandmother’s apple pie recipe and be very disappointed in the results. Or worse, if you had never eaten your grandmother’s apple pie, you may well take one bite and decide against ever eating any type of cooked apple again – a bad decision, based on bad analysis from a flawed mash-up.
Lest you consider my objection too fruity, you can find much more technical treatments of the same basic phenomenon in many systems engineering (e.g. Tactical Air-Launched Weapons) and enterprise architecture (e.g. Ariane 5 Flight 501) tracts.