Business Optimization

Efficient operations in complex business networks

Turning innovation into reality for today’s complex business networks requires — more than ever before — a holistic understanding, the right mixture between cutting-edge and established techniques, and an end-to-end solution approach.

All members of our team have a deep understanding in one or more of the following key disciplines — in addition to their industry expertise. We aprtner closely with universities, the worldwide IBM Research organizations and other IBM business units.

  • Combinatorial optimization

    Combinatorial optimization

    Simply giving an integer programming formulation does not always help solve a problem, because even the most advanced solvers like CPLEX will not terminate in an acceptable amount of time. In such cases, it is important to come up with a customized combinatorial algorithm that is tailored to the specific problem structure. Such algorithms are usually based on classical design principles such as dynamic programming, greedy, or branch-and-bound. The goal is to design such an algorithm that runs in low-order polynomial time and, in the best case, returns an optimal solution. However, approximate solutions are also acceptable.

  • Control

    Forecasting

    "Prediction is very difficult, especially about the future" is a well-known quote from Niels Bohr, among others. Nevertheless, today's business strongly relies on suitable forecasts — just like basically any project for decision support or optimization in an uncertain business context. Organizational forecasting today "rests on solid theoretical foundations while also having a realistic, practical base that increases its relevance and usefulness to practicing managers" (Makridakis et al.: Forecasting). Upfront, neither the most trivial extrapolation nor the most complicated neural network is superior or inferior for forecasting. Basically, every problem requires a unique combination of different techniques, whereas Occam Razor is a good guiding principle for exploring feasible solution paths.

  • Robust optimization

    Optimization under uncertainty

    Usually, the complex optimization problems we want to solve are subject to various types of uncertainty (e.g. uncertainty of the data, of the outcome) and we want to ensure that we take this into account in our decision process, i.e. that the solution we provide to a specific problem remains feasible and reasonably good under various perturbations. Robust optimization is a framework that allows one to include uncertainty as part of the formulation. Starting with some reasonable hypotheses regarding the uncertainty sets, one can solve those problems efficiently. We have experienced robust optimization formulations of corporate portfolio optimization problems and risk management problems in the pharma sector.

  • Combinatorial optimization

    Simulation

    There exists a plethora of different simulation tools — from general-purpose simulators all the way to specialized domain simulators. They are well suited to study a system of standard processes and components, especially as they provide instant statistical evaluation and often a good visualization. Additionally, many simulation solutions also offer an underlying (scripting) language to extend or adjust the provided portfolio of processes, components and evaluation methods. Nevertheless, it can be wise to build a special-purpose simulator from scratch in situations where only a small proportion of the provided functionality is suitable. We have a long history of stochastic discrete-event simulation, simulation-based optimization, including infinitesimal perturbation analysis (automatic differentiation), to model and evaluate business processes, especially in the domains of supply chain management and transportation systems.

Contact the experts