Before we discuss abandoning forecasting entirely, it's worth acknowledging that the industry has made progress. Although traditional "point forecasting"—predicting that demand will be exactly 47 units next month—is still the predominant method used by ERP purchasing modules and Supply Chain Planning vendors, there are a handful of software vendors who are now offering a smarter approach called probabilistic forecasting.
The Probabilistic Approach: Better, But Still Not Enough
Companies like Smart Software (acquired by Epicor in 2024) pioneered probabilistic forecasting for wholesale distribution. Rather than generating a single-number forecast, probabilistic forecasting outputs a full range of possible item demand levels, using probability theory to attach odds to the possibilities, then runs thousands of scenarios to expose the full range of possibilities.
This is a significant improvement over traditional point forecasting. For a spare part with a 33-week replenishment lead time, while the average lead time demand might be 3 units, the most likely demand is actually zero, and a reorder point of 14 units is needed to ensure only a 1% chance of stockout. A point forecast would have suggested 3 units—completely wrong for inventory planning purposes.
The advantage of probabilistic forecasting is that it doesn't rely on direct point forecasts as the basis of inventory replenishment and stocking levels—instead, it determines probabilities for every possible demand and aligns demand probabilities to percentiles (like 52%, 75%, or 89%) as inputs to inventory control models based on desired service levels. A critical spare part might be planned at an 89% service level, while a non-critical part might use a 52% service level. The system goes directly from the probability distribution to the recommended inventory replenishment amount.
Probabilistic forecasting is better for wholesalers than point forecasting because it is a better approach to uncertain demand. Instead of pretending you can predict a single number, it generates a range of potential outcomes with associated probabilities, creating a more realistic picture for decision-making. This potentially generates better results for low-volume intermittent demand, which is prevalent for wholesalers.
Why AI-Generated Inventory Targets Beats Probabilistic Forecasting
While probabilistic forecasting is superior to point forecasting, it's still fundamentally trying to predict demand—even if it's predicting a probability distribution rather than a single number. The breakthrough with AI-generated inventory targets is that they stop trying to forecast entirely.
The fundamental difference is philosophical. Probabilistic forecasting asks: "What's the probability distribution of demand?" For items selling 8 times a year, this is still the wrong question—even with sophisticated probability distributions, you're trying to predict the unpredictable.
AI-generated inventory targets ask fundamentally different questions:
The Customer and Cost Dimensions That Forecasting Misses
This is where AI-generated inventory targets reveal their power. Consider two SKUs that both sell 8 times per year:
SKU A: Sells to 1 customer, 8 times per year, at $50 cost, 20% margin
SKU B: Sells to 8 different customers, once each per year, at $5 cost, 40% margin
Probabilistic forecasting would treat these identically—they have the same demand pattern. But the optimal inventory strategy for these two items is radically different:
AI-generated inventory targets incorporate customer concentration, item economics, and demand patterns to determine optimal stocking levels—or whether to stock at all. This is SKU rationalization combined with inventory optimization.
For that automotive part selling 8 times a year, AI doesn't try to predict which month those 8 sales will occur (point forecasting) or generate a probability distribution of possible sales (probabilistic forecasting). Instead, it calculates the optimal inventory position that balances:
All of this adjusts continuously based on actual sales patterns rather than forecasted demand or probability distributions.