For more insights on financial planning and analysis (FP&A) check out AFP’s latest FP&A Guides. DOWNLOAD

While more companies are trying to perfect the science of forecasting with techniques like driver-based modeling and technology, there’s one thing that continues to stand in the way of making pure data-driven decisions: human nature (see the AFP FP&A Guide, Forecasting: Best Practices for Common Challenges).

Managers continue to massage the forecast to sync it with the kind of decisions they’re trying to make. “Companies make a lot of smaller judgements that waste time and incrementally hurt the forecast,” said Paul Goodwin, Professor Emeritus of Management Science at the University of Bath in England, who has written extensively on the use of judgement in business forecasting.
 
Protecting the integrity of the process

There are at least seven things FP&A executives can do to insulate the forecast from the effects of human judgement.

1.    Appoint an independent partner. Sales forecasts reflect sales expectations. Marketing forecasts reflect marketing expectations. To get at an objective and accurate forecast, companies must appoint an independent group to aggregate and run the forecasting numbers. At most organizations, that’s the role of the FP&A group.

2.    Separate forecast from decision. According to Goodwin, one way finance can ensure the forecast remains unbiased is to separate it from the act of decision-making. “The forecast should be a genuine expectation of will happen based on the information you have at the time,” he said. The consequences are the decision. If forecasters consider the consequences, personal (I may get fired if my forecast is wrong), or corporate (I will have to deliver on that forecast) consideration will taint the numbers that are produced by the most sophisticated models.

“Break the problem into small parts,” Goodwin advised. “Don’t think of the two things at once. That can get overwhelming.” In many companies, forecasters fall into the trap of feeding senior managers the numbers they want to see. While new tools may help improve the accuracy of the forecast, they’re only good as long as they’re being trusted. A famous academic experiment showed the majority of managers still trust their gut more than a statistical model.

3.    Concentrate on history. It’s amazing how many explanations managers can come up with to explain recent random changes in numbers. “There’s always this expectation that next week will be different,” Goodwin said. By looking at historical data, it’s possible to dispel notions of a “brand new world” just because there’s a twitch in the grass.
 
4.    Design a careful feedback loop. Just telling managers their forecast was less accurate than the model is not enough, according to Goodwin. Finance needs to supply managers with more actionable information. For example, it’s more useful to let managers know that they’ve been consistently over-forecasting by 10 percent or under-forecasting by 5 percent. That’s information they can act on. “Provide bias feedback vs. accuracy,” Goodwin said. In addition, if possible, provide feedback on what particular elements are over-weighted in the forecast.

5.    Clearly define the forecast. Another way to combat human judgement is to clearly define what the forecast is—and isn’t. “The forecast is strictly the best expectation of what will happen,” said Goodwin. “There should be a clear definition of what the forecast should be.”

6.    Combine multiple views. Another approach is to collect anonymous forecasts from multiple people. It’s important that the data is collected anonymously because if done in a face-to-face meeting, people are often shy or concerned about contradicting the higher ups.

7.    Design the right support system. Finally, Goodwin suggested software companies begin to design forecast-support systems that take into consideration the human element, not just data. Most tools rely on ever-more-precise algorithms. That approach may produce more accurate numbers but it doesn’t guarantee that these numbers will make it into the decision-making process. In a way, that’s a self-defeating programming objective, Goodwin noted. Vendors are touting their tools based on their accuracy. To incorporate human adjustments would be counterproductive.

As more FP&A groups invest in new technologies, and are facing demands from management to deliver actionable information faster and more frequently, they need to protect the integrity of the forecasting process from human interference. While tools that allow more scientific forecasts, using algorithms, scenario analysis and Monte Carlo simulations are becoming more prevalent, one cannot ignore behavioral science findings about how we all make decisions and view the future.