CFOs can improve the accuracy of their sales estimates — and, ultimately, earnings forecasts — by supplementing statistical modeling with the human-judgment side of forecasting, Karen Sedatole, an accounting professor at Emory University, told CFO Dive last week.
Sedatole and her co-authors have research coming out early next year showing improved sales estimates at an agriculture chemical company that used what she calls disaggregated forecasting. She’s also working on human-judgment research that looks at ways to better align incentives to more accurate forecasts.
In her research on disaggregated forecasting, she had the chemical company split its sales forecast into two estimates: one for sales to regular customers, and another for sales that are influenced by events that are hard to model ahead of time but that sales managers would typically be aware of.
“In the agriculture industry, much of the demand for your product comes from your regular customers — the Home Depot, the large growers — and it's pretty predictable,” Sedatole told CFO Dive. “You can forecast that a little easier. Then there are other sources of demand that are volatile. There might be a huge increase in demand for a product because of a drought in a certain part of the world, or an outbreak of a pest.”
For her test, Sedatole had sales managers keep logs tracking anomalous events they believed could affect demand. Once they believed an event would occur as predicted, they would upload their updated assumption to the company’s enterprise resource planning (ERP) system for use in the revised forecast.
“Once they decide that this event is pretty likely — we’re 90% sure this is going to happen next month — it gets moved into the official part of the forecast and the production planners incorporate it into the production planning process,” she said.
Sedatole measured results using a before-and-after analysis and determined managers’ input mattered. “We found it actually did reduce overall forecast error,” she said.
She's looking at other research that suggests CFOs can better align incentives to improve accuracy by understanding the bias human judgment introduces to the forecasting process.
Sedatole said there’s a built-in incentive for the sales organization to underestimate demand while the budget’s being prepared and to overestimate demand when the forecast is rolled out and updated throughout the year.
“Budgets are a little bit of a negotiation process,” she said. “When I’m negotiating my budget targets, with people higher up in the organization, I negotiate a sales target of 90, because my bonus is tied to meeting my targets and I think I can sell 100. That way, I make sure I meet my budget by the end of the year."
When it's time to forecast, the incentive on the sales side is to convince the production planner to make more, not fewer, units.
"If I tell them 90, the production manager is only going to produce 90 and then I will lose out on 10 units of sales, so I tell the production manager 100, or maybe even 105, because I want to make sure there’s enough," Sedatole said. "So, I set my budget at 90, but now I’m forecasting 105 to try to influence a decision on the operations side of the organization.”
Tapping human insight
Much of today's focus is on using data analytics to improve budgeting and forecasting, but more work on the human-judgment side is needed to tap into the potential to improve both, Sedatole said.
“The human aspect is where the biggest improvements in forecasting can come from,” she said. “Of course, data analytics are getting more sophisticated, but some of the improvement is going to come from the managerial judgment side, which may be a byproduct of changes in the way we incentivize managers when they generate those forecasts.”