Why Enterprise Analytics - and why now?
It's big and it's heading your way
There are markets where undifferentiated, dog-eat-dog competition has existed for decades. The best real-world example of this type of environment is the financial services industry. Think about portfolio managers: they’re competing against hundreds of others, making decisions based on the same publicly available information, and trying to eke out a few more basis points than the other guy.
To make their trading decisions and guide their short- and long-term strategy, they have built huge and highly complex models that attempt to mimic reality and predict future prices based on current trends and conditions. These models are their crown jewels, and are closely guarded.
They don’t stop at analyzing particular companies but also consider economic trends, demographics, societal evolution, and just about anything else that might have an effect on what they’re trying to predict. Regression helps them characterize the relevance and predictive strength of a particular factor, while techniques like Monte Carlo are used to compute the probability of the result or condition.
This type of analysis has gotten a black eye due to the recent financial meltdown arising from the mortgage-backed securities debacle. Our research into the root causes lead us to believe that the failure wasn’t in the models, but in the values input into them and the idiots who relied upon them (see our previous article). The bottom line is that the math works, but the models and the humans interpreting them can often be flawed.
But just as the crappy bookshelves I build aren’t an indictment of power screwdrivers as tools, bad results from mistaken modeling doesn’t mean the modeling techniques don’t work – they do. And it’s the next phase in business. We’re going to see more and more companies attempting to model everything from their supply chain to long-term customer behavior in a much deeper and more predictive sense than before.
While some may say that we’ve been doing this all along, I’m not so sure. Most of the business intelligence that I’ve seen in the real world tends to use company-generated data to make short-term predictions. Many companies still don’t have their internal BI systems to a point where they can be used to run through scads of data (1 scad = a hell of a lot of data) to find non-intuitive relationships.
In many cases, management uses their decision support systems to support their hunches and speculations rather than let themselves and their strategies be guided by the data. Part of this is because they don’t feel that the data is reliable – and they’re often right about this. But they are also fearful that letting data actually drive their strategy means they are less valuable. Like most important trends, the adoption of predictive enterprise analytics and what we’re calling the “Analytics-Led Enterprise” will be hindered more by internal culture than by technical factors.
As the trend emerges it’s going to have definite implications on the IT market as a whole and HPC in particular. These business problems look awfully HPC-like when you consider the workloads and scale. The same folks who can analyze and predict how gas molecules will react in a pipe can use their tools to model how shoppers will move through a crowded store… and come up with predictions of how changes will impact traffic at the cash register. It’s all data and math, right?
Sponsored: Hyper-scale data management