Rolling Horizon Planning

The core mechanism involves solving a multi-period optimization model over a defined horizon (e.g., 12 weeks), but only the first period's decisions are physically executed. The model then rolls forward one period, incorporates newly observed data, and re-optimizes. This creates a continuous feedback loop where each iteration benefits from the most recent demand signals, inventory positions, and supply disruptions, preventing the plan from diverging from reality.

A critical governance mechanism where the immediate upcoming period is designated as frozen or firmed, meaning no algorithmic changes are permitted. This provides operational stability for:

• Warehouse picking and packing schedules
• Carrier booking and dock-door assignments
• Manufacturing line changeovers Subsequent periods remain slushy (minor adjustments allowed) or liquid (fully re-optimizable), balancing execution certainty with planning flexibility.

A known pathology where the optimization model makes irrational decisions near the end of the planning window because it sees zero value in holding inventory beyond the horizon. Mitigation strategies include:

• Salvage value constraints: Assigning a terminal value to ending inventory
• Extension buffers: Adding dummy periods beyond the true horizon
• Terminal constraints: Forcing ending inventory to match a target safety stock level Without these, the model will artificially deplete stock, creating a bullwhip effect in subsequent runs.

Advanced implementations replace a single deterministic forecast with a branching scenario tree that represents multiple possible futures. At each decision point, the model considers probabilistic paths for demand spikes, supplier delays, or price fluctuations. The optimization then finds a non-anticipative solution—a decision that is feasible regardless of which scenario materializes—while minimizing expected cost across all weighted outcomes. This transforms the plan from fragile point estimates to robust hedges.

Re-solving a large-scale mixed-integer program every period demands careful engineering. Techniques to maintain speed include:

• Progressive hedging: Decomposing the stochastic problem by scenario
• Warm starts: Seeding the solver with the previous period's solution
• Horizon aggregation: Bucketing distant periods into coarser time blocks
• Variable fixing: Locking binary decisions (e.g., plant openings) for near-term periods The goal is sub-hour solve times to fit within nightly batch windows or event-triggered replanning cycles.

Beyond the standard periodic cadence, intelligent systems initiate an out-of-cycle replan when deviation metrics breach thresholds:

• Demand signal: Actual orders deviate >15% from the forecast envelope
• Supply disruption: A supplier force majeure or port closure event
• Inventory excursion: Stock levels fall below safety stock minimums
• Financial signal: Spot market transportation rates spike beyond model assumptions Event-driven replanning ensures the frozen period doesn't lock in decisions based on obsolete assumptions during volatile conditions.

The holistic framework that rolling horizon planning operates within. MEIO simultaneously optimizes stock levels across all network nodes—from suppliers to retailers—to minimize total system-wide costs. Rolling horizon planning provides the temporal execution mechanism, re-solving the MEIO model with updated data at each iteration.

• Key Link: MEIO defines the network model; rolling horizon planning executes it iteratively.
• Benefit: Prevents local optima by considering echelon interdependencies over time.

An optimization framework that explicitly models uncertainty through a discrete set of probabilistic scenarios. Rolling horizon planning often uses stochastic programming as its core solver to make decisions that are robust across multiple possible futures. At each re-solve, new scenarios are generated from updated demand and lead time forecasts.

• Two-Stage Recourse: Decisions made now, with corrective actions modeled for later.
• Scenario Tree: A branching structure representing how uncertainty unfolds over the planning horizon.

The application of machine learning to short-term, high-frequency data signals like daily point-of-sale transactions. Demand sensing provides the critical near-term forecast update that feeds the rolling horizon model at each re-optimization cycle, dramatically improving the accuracy of the immediate period's execution decisions.

• Data Sources: POS data, weather, social sentiment, web traffic.
• Impact: Reduces the bullwhip effect by anchoring plans to real consumption signals.

The algorithmic calculation of precise buffer inventory levels required at each echelon to absorb variability. In a rolling horizon context, safety stock targets are not static; they are dynamically re-calculated at each planning cycle based on the latest demand and supply uncertainty forecasts, ensuring service levels are met without excess capital lockup.

• Dynamic Buffers: Safety stock adjusts as lead time volatility changes.
• Service Level: Directly linked to the cycle service level or fill rate target.

A time-phased planning methodology that applies dependent demand logic to distribution networks. Rolling horizon planning is the modern, optimization-based evolution of DRP. While traditional DRP uses sequential, single-echelon logic, rolling horizon MEIO solves all echelons simultaneously, considering capacity constraints and cost trade-offs.

• Evolution: DRP calculates net requirements; rolling horizon optimizes them.
• Constraint Awareness: Unlike DRP, rolling horizon models finite warehouse and transportation capacity.

A phenomenon where small demand variability at the retail level causes amplified order oscillations upstream. Rolling horizon planning directly mitigates this by sharing a unified, frequently updated plan across all echelons. By re-optimizing based on actual end-customer demand rather than distorted downstream orders, the feedback loop dampens variability.

• Cause: Batch ordering, lead time delays, and information distortion.
• Mitigation: Frequent re-planning with shared demand visibility.