Predictive Analytics is still searching for a home in the world of performance management
IBM's recent announcement of its intent to acquire of SPSS has gotten me thinking about predictive analytics and its potential application in the world of performance management.  Pundits have been touting its potential for years without much traction in the real-world.  Why is that?

Predictive Analytics remains a Black Box

Business users don't trust an algorithm to generate their forecast.  "How can a calculation engine know my business better than me?  I'm out there in the field every day.  I know what's happening out there more than some propeller-head crazy regression analysis!"  The psychology of it is that they are afraid of what they don't know (or understand) and certainly don't want to be held accountable to something they don't know or understand.

We Expect More and Better Forecasts
Everyone agrees that businesses are forecasting more frequently than ever before because of our current challenging economic times.  All are trying to predict the bottom and gauge the intensity of a recovery.  Will it be a hockey stick, a V-shaped or a  U-shaped recovery?  More frequent forecasting will help companies see the recovery coming and react to it more quickly than they did to the downturn.  However, the bigger question is if we are expecting the business to forecast more frequently how far out should we be expecting our lines of business to forecast?  Can we expect accurate 18 month rolling forecasts every month, every week?  I think it is unrealistic to expect a detailed forecast that someone is willing to hang their hat on that looks more than six to nine months out.

Can Predictive Analytics Provide the Answer?
As we move from annual and quarterly forecasting to monthly, weekly or even daily forecasting.  Let the line managers build the short-term forecasts and hold them accountable to them.  But don't bog them down trying to forecast what their sales will be 18 months from now.  This is where I think predictive analytics can fill the gaps.  Predictive analytics can model the outlying periods and create forecasts for those periods using predictive algorithms rather than gut feel and pipeline (both of which lose accuracy significantly when looking more than a few months out).

Combined with Field Knowledge
The results of these predictive models could then be provided to the field.  If they disagree with the forecasts fine, but they need to document why.  Why are the numbers we provided not realistic?  Is there a big deal looming 12 months out?  Is there a costly initiative that the model didn't take into account?  If so let's consider the impact and adapt the model, if not then the model stands.

The bottom line is that if we are expecting managers to forecast much more frequently we need to be conscious of the effort involved and the accuracy expected.  Having them forecast deviation from the predictive model rather than developing a full long-term reforecast saves time and effort and more than likely more accurate rolling forecasts.  I think incorporation of predictive analytics into the planning process would take full advantage of the gut instinct and business knowledge in the field and the predictive capabilities of technology without trying to make the business user think like a PhD.

What do you think?  Is this a way to use predictive analytics effectively without having to push it out to the masses?