Abstract:
The available-to-promise (ATP) model is a mechanism that provides recommendations
about when to accept customer orders that takes into account both product availability
information, current customer orders and future orders in order to maximize overall
profits. It becomes an important tool in a decision making process for manufacturing
businesses. In this thesis, we present the stochastic available-to-promise problem which
addresses the problem of needing to accept-or-reject in real-time orders for customizable
computer configurations where the manufacturer cannot predict when the most profitable
customers might arrive, but does have some probabilistic information about the
likelihood of order arrivals and their requirements. Because the problem is stochastic,
modeling all possible future scenarios results in an exponentially large problem. Even
when one limits the total number of scenarios considered, solving the problem by off-the-shelf
commercial solvers such as CPLEX results in computation times that are too large
to be usable. We study the underlying structure of the model and propose decomposition
methods to solve it. We test both a Dantzig-Wolfe decomposition (“Column-generation
approach) and a Bender’s decomposition (a row-oriented decomposition). We compare
solution times of both methods to solution times of CPLEX.
