There's no replacement for the "innate business intelligence associated with people in the field who know of the one-off situations that are coming up," he says. "Being able to integrate information affecting customers into a data-driven forecast is always going to drive a better result."
Still, the human factor can skew a forecast. "Politics and bias" are always an issue, says Karelse. "Every person brings to the table, no matter how pure their intentions are, biases that will color their judgment." The political side emerges when different departments adjust the forecast to suit their own needs. To combat that tendency, companies need to deploy "situation-based bias correction," involving collaboration among all parts of the business.
Bias correction is not a new idea. The problem with the traditional approach, Karelse says, is that it "assumes people are always wrong for the same reason, and to the same extent." Yet one individual might be wrong 50 percent of the time on new products, and only 3 percent on mature items. "If you assign a co-efficient for each [person and category]," he says, "you get much closer to actual demand."
Of course, it would be impossible to parse the decisions of every individual for every product. The answer, says Karelse, is to focus on items of high value that are "innately forecastable." They can be broken down into half a dozen situations, based on their position in the product lifecycle. For products with more stable or all-season demand, by contrast, trying to adjust the numbers according to multiple factors probably isn't worth the time. "You're choosing your battles," he says.
.webp?height=100&t=1780416625&width=150 "018_how_3pls_can_get_started_with_ai_v1-(540p).png")
