While other analytics address other areas of concern - establishing cause and effect in the marketplace, identifying obscure influences on industry trends, and so on - predictive analytics have the unique burden of informing tomorrow's decisions.
Shrinking the landscape
A typical predictive analytics problem/solution, commonly cited as an example, is the behavior of homeowners, per their mortgages. It's useful for a mortgage company to analyze payment histories and determine the various markers in homeowner payment that indicate a homeowner in trouble, a homeowner about to default, a homeowner likely to make an early payoff. Being able to plan for those events months ahead of time is a huge financial advantage for the mortgage company, on several fronts.
When those predictive numbers are delivered, they reconfigure the near future as a concise new landscape, with small vistas of constrained, well-articulated probabilities. Knowing how many mortgages are likely to go away via early payoff is good for bottom-line planning; knowing how many mortgages will need to be refinanced so that they don't vanish is good for resource allocation.
The overall point: each of these areas of the future landscape is uncertain, and uncertainty is expensive, and the management of that uncertainty makes good economic sense.
Through the window, into the mirrorIt's often possible to take this general concept a step further. What about the uncertainty to be found in the results themselves, beyond the uncertainties they address?
Even if you've spent a great deal of time working with business intelligence, let alone analytics, it's hard to keep in mind that the results of any predictive process are tentative, and exist as probabilities. Only hard-core scientists really have this in mind at all times, and even they can slip. That means that the results we act on, when using predictive analytics as decision support, are probabilities at best.
How does that affect the way we use them? Consider the mortgage problems above: slow payment, impending default, early payoff. Each of these problems has a substantial administrative cost, potential revenue losses, and opportunity consequence. If I'm the one delivering the analytics, I want to be able to not only report the most accurate projections of what will happen next quarter, I want to be able to recommend where effort is most effectively applied.
In other words, I can do the same analysis on my quarter-to-quarter results in each of these predictions and be able to add a meta-recommendation: our default recommendations are more accurate than our payoff guesses; more time and money should go into foreclosure relief next quarter than acquiring new business!
The principle is clear: predictive analytics gives us a handle on the near future, and helps us strategically deploy our time and resources most effectively to do the best possible job in the months ahead. But those are probabilistic results, and carry uncertainty of their own. By measuring their reliability against each other, we can make our near-future time and resource allocations that much more fine-tuned, and increase our efficiency all the more.
Scott Robinson is a 20-year IT veteran with extensive experience in business intelligence and systems integration. An enterprise architect with a background in social psychology, he frequently consults and lectures on analytics, business intelligence and social informatics, primarily in the health care and HR industries.