Configuring Forecast Profiles and Selecting Forecast Model Parameters

Explanation

Once the business rationale for forecast models is understood, the next step is translating that understanding into actual configuration within SAP IBP. The central configuration object here is the forecast profile, which is maintained as part of the demand planning configuration and referenced by the forecast run (either an interactive run in an Excel planning view or a batch job scheduled through the application job framework). A forecast profile bundles together the forecast model choice, its parameters, the history horizon, the forecast horizon, and settings related to outlier correction and error measurement. When configuring a forecast profile, the consultant typically starts by deciding between a fixed model (explicitly choosing, for example, a triple exponential smoothing model that handles level, trend, and seasonality) versus automatic model selection, where IBP evaluates a candidate set of models against history and selects the best performer based on a chosen error metric such as MAPE or MSE. Automatic model selection is attractive for large product portfolios where manually assigning a model per SKU is impractical, but it requires sufficient clean historical data to make a meaningful comparison; for new or sparse-history products, automatic selection can behave unpredictably or default to a naive model. Parameter tuning is where intermediate-level judgment comes in. For exponential smoothing style models, smoothing constants (alpha for level, beta for trend, gamma for seasonality) control how quickly the model reacts to recent changes versus how much it relies on longer historical patterns. Higher alpha values make the forecast more reactive to recent demand spikes, which can be useful for products experiencing genuine shifts but harmful for products with noisy, non-trending demand. IBP allows these parameters to be set manually or estimated automatically through optimization during the forecast run; many implementations start with automatic optimization and only override manually for specific problem SKUs identified during pilot testing. History horizon and forecast horizon are also profile-level decisions. The history horizon defines how many past periods (weeks/months, depending on the planning time profile) are fed into the model. Too short a history can miss seasonal cycles; too long a history can force the model to weight outdated patterns that no longer reflect the market (e.g., pre-disruption demand behavior). The forecast horizon defines how far into the future the model projects, which must align with the planning horizon needed by downstream supply and S&OP processes - forecasting further out than the model can reliably support introduces false precision. Outlier correction settings matter significantly in real data sets: promotional spikes, stockout-driven demand drops, or one-time bulk orders can distort a model's perception of the 'normal' pattern if not corrected before or during the forecast run. IBP provides mechanisms to flag and adjust for outliers as part of the forecasting process, and consultants need to decide, in collaboration with the business, which historical events should be treated as noise versus genuine signal (for example, a recurring annual promotion should likely be treated as seasonal signal, not an outlier). Finally, configuring forecast profiles is not purely a backend/IT activity - it directly shapes what planners see in their Excel-based planning views and what numbers feed the consensus demand review. Intermediate consultants need to test forecast profile changes against a holdout period (running the forecast as if 'today' were some months in the past and comparing model output to actual known results) before deploying changes broadly, since forecast profile changes affect live planning data once activated.

Code example

# Conceptual example of forecast profile parameters (illustrative, not a real IBP configuration file)# This represents the kind of parameters a consultant reviews when setting up a forecast profileforecast_profile:  name: FCST_PROFILE_FASTMOVERS  model_selection: AUTOMATIC          # or FIXED_MODEL  candidate_models:    - EXPONENTIAL_SMOOTHING_SEASONAL    - EXPONENTIAL_SMOOTHING_TREND    - MOVING_AVERAGE  error_measure: MAPE                 # used to pick best model when AUTOMATIC  history_horizon_periods: 104        # e.g., 104 weeks of history  forecast_horizon_periods: 26        # 26 weeks forward  outlier_correction: ENABLED  outlier_threshold_std_dev: 2.5  parameter_optimization: AUTOMATIC   # alpha/beta/gamma auto-estimated  # Manual override example for a known problem SKU segment  overrides:    - segment: SLOW_MOVERS_INTERMITTENT      model_selection: FIXED_MODEL      fixed_model: CROSTON_INTERMITTENT      history_horizon_periods: 156

Real project scenario

During a rollout for a mid-size industrial parts distributor, the initial forecast profile used automatic model selection with a 52-week history horizon for the entire product portfolio, including thousands of low-volume spare parts. Pilot testing against a holdout period showed unacceptably high error for the spare parts segment because the automatic selection kept choosing trend-based models unsuited to sporadic, lumpy demand. The team created a second forecast profile specifically for the intermittent-demand segment, extended its history horizon to capture rare demand events, and applied an intermittent-demand-appropriate model, which measurably reduced forecast error for that segment before go-live.

Common mistakes

• Applying automatic model selection uniformly across products with very different demand patterns without segment-specific profiles • Setting an overly long history horizon that includes outdated or disrupted demand periods without adjusting for known anomalies • Enabling aggressive outlier correction without business validation, accidentally smoothing out genuine seasonal spikes like promotions • Deploying forecast profile changes directly to production without holdout-period testing against known historical outcomes • Ignoring the alignment between forecast horizon length and the actual planning horizon needed by supply and S&OP processes

Best practices

• Create separate forecast profiles for distinct demand pattern segments rather than one universal profile • Validate any forecast profile change using holdout-period backtesting before production deployment • Review outlier correction settings with business stakeholders to distinguish genuine anomalies from recurring seasonal events • Keep forecast horizon length aligned with the actual decision horizon of downstream supply and S&OP processes • Document parameter choices and rationale so future consultants or support teams understand why a given profile was configured a certain way

Interview angle

Candidates are often asked to explain the trade-off between automatic and fixed model selection, and how they would validate a forecast profile change before rolling it out. A strong answer references holdout testing, error measures like MAPE/WMAPE, and the importance of segmenting products rather than using one profile for an entire portfolio.