Configuring Statistical Forecast Profiles and Selecting Forecast Models

Explanation

A forecast profile in SAP IBP ties together the historical key figure, the forecast model, model parameters, and the horizon over which statistical forecasting runs. Getting this configuration right is central to demand planning accuracy, because a mismatched model (for example applying simple exponential smoothing to a strongly seasonal, intermittent-demand product) produces systematically biased forecasts that ripple downstream into S&OP, supply planning, and inventory targets. SAP IBP offers several forecast model families accessible through the forecast profile app or the planning view forecasting settings: single and double exponential smoothing for level and trend patterns, seasonal models (triple exponential smoothing / Holt-Winters style) for products with repeating seasonal cycles, Croston's method and its variants for intermittent or lumpy demand where many periods have zero demand, and composite/ensemble approaches that combine multiple models and select the best fit statistically per product-location combination. IBP also supports automatic model selection, where the system evaluates a candidate set of models against historical data using an error metric (such as MAPE or a weighted error score) and picks the best performer per combination, which is useful when planning thousands of SKUs with heterogeneous demand patterns but requires careful monitoring because the chosen model can change between forecast runs as history evolves. Configuration typically involves: defining the historical key figure (often actuals or corrected history), the forecast key figure that will store output, the forecast model type and its parameters (smoothing constants, seasonal periodicity, damping factors), the history horizon used for model fitting, and the forecast horizon and granularity (weekly, monthly) for the output. Outlier correction and history cleansing should generally be configured upstream of model fitting, because unaddressed spikes (promotions, one-time events) distort the parameter estimation, especially for seasonal and Croston models which are sensitive to irregular history. A critical intermediate-level skill is interpreting forecast error and bias metrics after a test run: reviewing MAPE, MAD, or bias by planning level (product, location, product group) to detect where automatic model selection is underperforming, and understanding when to override the automatic choice with a manually assigned model for specific segments (for example forcing Croston for spare parts or seasonal models for weather-driven categories). Causal or driver-based forecasting, where demand is explained by external factors like price or promotions, is a more advanced extension typically layered on top of these base statistical models rather than replacing them. Integration-wise, the forecast model configuration lives entirely within the IBP planning area and time profile; it does not depend on S/4HANA transactions, but the quality of the underlying history is often sourced from S/4HANA billing or sales order data through the standard integration (CPI-DS or SAP IBP integration add-on), so upstream data extraction settings (which document type, which date field) directly affect which history the forecast model consumes. On S/4HANA private cloud/on-premise, history extraction jobs are typically scheduled batch jobs; in fully cloud landscapes, integration is usually via SAP Integration Suite or the IBP add-on with similar scheduling constraints. Differences in extraction timing or filters between environments can cause forecast profiles that work well in a sandbox to behave differently in production if the actual history volume or granularity differs, so validating the profile against production-representative history before go-live is essential.

Code example

Example forecast profile configuration (conceptual, planning-area settings, not executable code): Forecast Profile: ZFC_FINISHED_GOODS History Key Figure: CORRECTEDHISTORY Forecast Key Figure: STATISTICALFORECAST Model Selection: AUTOMATIC (candidate set: SES, DES, Seasonal, Croston) History Horizon: 24 periods (monthly) Forecast Horizon: 12 periods (monthly) Seasonality Periodicity: 12 Outlier Correction: ENABLED, threshold = 3 std dev Error Metric for Model Selection: Weighted MAPE Manual override example for intermittent spare parts segment:Forecast Profile: ZFC_SPAREPARTS Forecast Model: CROSTON Alpha (smoothing constant): 0.2 Forecast Horizon: 12 periods (monthly) Outlier Correction: DISABLED (zero periods are meaningful, not outliers)

Real project scenario

A consumer goods client ran automatic model selection across 15,000 product-location combinations. During UAT, planners noticed several seasonal beverage SKUs were being fit with simple exponential smoothing instead of seasonal models because only 18 months of clean history existed, below the minimum the seasonal model needed for reliable parameter estimation. The team resolved this by extending the history horizon using legacy ECC sales data migrated into IBP, and by manually assigning seasonal models with a fixed periodicity for the affected product group while leaving automatic selection active for the remaining assortment, then re-validating bias and MAPE by segment before the profile was approved for the production planning cycle.

Common mistakes

• Enabling automatic model selection without segmenting history minimums, causing seasonal or Croston models to be skipped for lack of sufficient data • Leaving outlier correction on for intermittent-demand items where legitimate zero-demand periods get treated as anomalies • Fitting models on uncorrected raw history that still contains promotional spikes or one-time returns • Not reviewing error metrics by segment, only looking at an aggregate MAPE that hides poor performance in specific product groups • Assuming a forecast profile validated in a sandbox with sample data will behave identically once run against full production history volumes and real extraction timing

Best practices

• Segment the product portfolio by demand pattern (steady, seasonal, intermittent, new product) before deciding on automatic versus manual model assignment • Apply outlier correction and history cleansing before model fitting, and disable it where zero periods carry planning meaning • Validate forecast error metrics (MAPE, bias) at a granular level, not just in aggregate, before promoting a profile to production • Re-test forecast profiles against production-representative history volumes and extraction timing prior to go-live • Document manual model overrides and the business rationale so future planners understand why automatic selection was bypassed for specific segments

Interview angle

Interviewers commonly probe whether a candidate understands when to trust automatic model selection versus overriding it manually, how outlier correction interacts with intermittent demand models like Croston, and how to diagnose a forecast bias problem by tracing it back to history quality or model mismatch rather than assuming the algorithm itself is at fault. Be ready to explain a concrete example of segmenting products by demand pattern and assigning different forecast models accordingly.