FSN: Predictive analytics makes its debut

Predictive analytics makes its debut

3rd October 2005
Many column inches have been devoted on FSN pages and elsewhere to the need for smarter budgeting, planning and forecasting processes. The advent of specialised web based budgeting and planning applications coupled with appropriate workflow technology has transformed the budgeting and forecasting ritual from a slow, laborious process of dubious quality to a much quicker process. Despite real progress in this regard, many companies have failed to seize the real prize; that is, the ability to reliably predict future business performance. Now there are signs that one or two software houses are beginning to grasp the nettle and take us into the next generation of solutions based on computer generated forecasts. Gary Simon, FSN Managing Editor investigates.

As with all matters to do with corporate performance management there is an overabundance of confusing terminology and almost all software vendors will claim, with some justification, to have a predictive capability. But in order to understand the new science of "predictive analytics" in the business world it is necessary to differentiate present day solutions from the vision painted by predictive analytics.

The preparation of a forecast is by definition predicting future outcomes. Often, this version of the future is based on the actual results to date plus the remaining budget adjusted perhaps by management's latest view of likely business performance for the rest of the forecast period. So it is definitely a forecast. But how insightful is the forecast?

According to Christian Gheorghe, Chief Technology Officer of OutlookSoft, who was in London recently to launch a new predictive analytics package, the performance management market is stuck in a rut. He pointedly asked the question, "Is management managing performance or achieving performance?" According to Gheorge, too much time is spent monitoring historic performance rather than examining what is going to happen next. "Instead of asking what is going wrong, we should be focussing on what we are going to do about it," he suggested.

"Another limitation of CPM is the time it takes to resolve a problem. By the time a variance is reported, the reasons for it are identified and appropriate management action is taken to remedy the situation, it is probably too late and a more urgent problem has surfaced," says Gheorghe. But it does not have to be this way.

Geoghe believes it is the "discovery" phase of problem resolution which takes so much management time. Once an adverse variance is uncovered, there are so many possible paths of enquiry to drill down or drill through that it can take many days to discover the composition of the variance and the root causes - even in a unified database holding all of the performance management data.

OutlookSoft's innovation is to suggest that the system can identify all of the relevant variances on an exception basis and the power of the computer should be harnessed to discover the reasons for the variance and the root causes. What is revolutionary about Gheorghe's proposition is that the system finds the causes of the variance itself and does not follow some pre-ordained business logic or rules defined by the users. "You simply do not know what you do not know" says Gheorghe. Furthermore, the system can be used to add context to the variance by searching unstructured supporting documents in word and Excel.

Brian Hartlen, VP Global Marketing at performance management specialist GEAC agrees with the computer generated approach. "A computer generated analysis of variances and trends can be very useful. It may lead management to areas that they wouldn't have contemplated. But you need human interaction as well. I see guided analytics or statistical forecasting as a form of assistance to management. It's not going to replace other methods of forecasting on its own."

"You also need to consider the auditability of computer generated forecast. With so much emphasis these days on compliance you have to follow a rigorous process. Management needs to know exactly how every number in the forecast was derived. You can't say 'we don't know because the computer generated it'. Basically you need a comb ination of techniques which is why we have a structured approach in our Strategic Finance product and a statistical forecasting engine as part of MPC. Essentially, the optimum approach is a sort of triangulation between insights gained from historic performance, 'bottom up' budgeting in traditional fashion and management's instinct or guesswork".

Paul McCunn of Atos Consulting believes the automatic identification of root causes could highlight the real key performance indicators in a business. "Businesses often have difficulty identifying performance indicators so this could be a significant help and potentially applicable to a wide range of businesses. It could also highlight changing KPIs," he said.

This might all sound a bit farfetched and fanciful but OutlookSoft have a working application which they proceeded to demonstrate. Not only could automation of the discovery phase accelerate problem solving and time to action but it could also save money by eliminating the cost associated with present day methods of problem resolution.

Gheorghe reasons that once the true causes of a variance are discovered and therefore the exact algorithms that drive key performance indicators, it is possible to use the same logic to extrapolate future performance. This could be more accurate than existing management projections.

OutlookSoft is not claiming you can run a business on a 'fly by wire' basis. A reality check quickly establishes that management still needs to use its judgement to assess the information provided by the system and to decide the remedial action it is going to take.

To this extent it is a bit like a weather forecast. Sensors around the country detect low pressure areas and the root cause is identified as a depression from the North. The meteorological office computer predicts rain for at least three days and individuals can take the decision based on the information provided to protect themselves from the elements with an umbrella. At this stage no one is suggesting that the steady advance of rain clouds can be held back! But who knows - one day businesses could be 'waterproofed' using predictive analysis techniques.

About Us