Stochastic Service Model (SSM)

Unlike the Guaranteed Service Model (GSM) which assumes a fixed, deterministic maximum service time, SSM models the service time at each node as a stochastic random variable. This captures the real-world phenomenon where a supplier's actual delivery performance is not a constant but a distribution influenced by their own inventory availability. The model propagates these probabilistic delays downstream, calculating the probability density function of the total replenishment lead time rather than a single point estimate.

The core innovation of SSM is modeling the dependency between echelon service times. When an upstream node (e.g., a regional warehouse) stocks out, it cannot immediately serve a downstream node (e.g., a retail store). This creates a stochastic delay that is added to the downstream replenishment time. SSM explicitly calculates the probability of this blocking event and the resulting conditional delay distribution, providing a far more accurate picture of network-wide lead time variability than models that treat each echelon independently.

In an SSM framework, the service level at each node is not an exogenous input but an endogenous output of the inventory policy. The model solves for the optimal safety stock levels by recognizing that higher safety stock at an upstream node reduces its probability of stocking out, which in turn shortens and stabilizes the effective service time experienced by downstream nodes. This creates a feedback loop where inventory investment at one echelon directly improves the operational performance of the entire downstream network.

SSM draws heavily on queueing theory and inventory theory to model each stocking location as a service facility. Common modeling approaches include:

• M/G/∞ queues: To model the production or replenishment process with ample servers.
• Phase-type distributions: To approximate complex service time distributions with a combination of exponential stages.
• Renewal theory: To characterize the stochastic process of order arrivals and inventory depletion over time. This analytical rigor allows for closed-form or computationally efficient approximations of the complex state-dependent dynamics.

While both SSM and Guaranteed Service Model (GSM) are multi-echelon optimization frameworks, they differ fundamentally in their treatment of time:

• GSM: Assumes each stage can guarantee a maximum service time, placing safety stock to cover demand variability only during this deterministic bound.
• SSM: Assumes service time is a random variable dependent on upstream stock availability, placing safety stock to cover both demand variability and the stochastic nature of the lead time itself. SSM is more accurate for networks with high utilization and frequent stockouts, while GSM is often a conservative, simpler approximation.

The SSM optimization problem seeks to minimize the total inventory carrying cost across all echelons subject to a target end-customer service level constraint (e.g., a fill rate or cycle service level). The decision variables are the base-stock levels at each node. The model must solve a complex, non-linear optimization because the cost function and the service level constraint depend on the convolution of probabilistic demand and the state-dependent, stochastic lead times generated by the upstream inventory policies.