Worst Planning Practices in SAP IBP — WP #2: A “Best Fit” Model Pool with No Design Thinking

Continuing our Worst Planning Practices in SAP IBP series — The Best Fit configuration checked all the right boxes during implementation, but its real-world performance introduced considerable forecasting challenges.

The Scenario

A consultant configured Best Fit statistical forecasting in SAP IBP with an impressive-looking model pool:

• Auto-ARIMA/SARIMA
• Automated Exponential Smoothing
• Simple Moving Average
• Croston Method
• Seasonal Linear Regression
• Gradient Boosting of Decision Trees

Six models. Looks thorough. The problem? Not even one was selected based on analysis of actual historical data patterns.

What Actually Happened

The models were randomly assigned to the Best Fit pool, and the system was set to reselect the “best” model every month — applied uniformly across all 2,000+ SKUs, regardless of volume and demand variability. To make matters worse, there was no lower limit set on the forecast.

The result was exactly what you’d expect: Automated Exponential Smoothing projected a decline of nearly -55,000 units by early 2025. Other models clustered near zero. Planners had no consistent number to trust because the winning model changed every month.

Three Root-Cause Failures

1. Poor model pool design: Models were included without evaluating whether they matched the behavioral patterns in the SKU portfolio. A Croston Method, designed for intermittent demand, has no business competing on high-velocity, seasonal SKUs — and vice versa.

2. No SKU-level segmentation: A single Best Fit logic was applied uniformly to thousands of SKUs. An ABC-XYZ or similar segmentation approach would have differentiated model pools based on volume and variability — steering each SKU toward candidates that actually fit.

3. No forecast floor: The system was never told that a negative forecast is operationally meaningless. Without a floor set at zero — and without period-over-period change thresholds — the engine was free to produce numbers that no planner could act on.

This directly violates one of the foundational requirements of demand planning: every month, the business needs a robust, defensible forecast it can build execution decisions on.

What Good Looks Like

Best Fit is a genuinely powerful SAP IBP capability — but only when used deliberately:

• Candidate models are selected based on your portfolio’s actual data patterns
• SKUs are segmented and model pools are differentiated accordingly
• Guardrails are configured: floor at zero, outlier capping, max period-over-period change limits
• Model performance is actively monitored, not assumed

A Best Fit model is only as good as the models you put into it.

Deliberate model governance is a core part of how we approach SAP IBP usability consulting — because configuration decisions made at setup define forecast quality for years.

What’s Your Experience?

Have you encountered a poorly configured Best Fit model — in SAP IBP or another planning tool? If your team is running or preparing for an SAP IBP implementation, this is exactly the kind of configuration detail that separates a high-performing forecast from a monthly fire drill. Share this with your planning team and let’s keep the conversation going.

If your SAP IBP configuration needs a second look — reach out to us. We’ve seen this pattern across hundreds of implementations.

Read: SAP IBP Worst Practices WP1 — Applying an Aggressive Forecast Model to a New SKU

💬 Have you run into poorly configured Best Fit models in IBP or other planning tools? Drop your experience below.