You're struggling to analyze decisions in Operations Research. What tools can help you get unstuck?

It is essential to understand a range of models and techniques in the field and the application problem space. Brown (2004) discusses five aspects to consider before embarking on an operations research problem. These include (1) clearly defining the problem, (2) the importance of the problem, (3) how the problem would be solved without your approach (relativity measure), (4) what you are doing to solve the problem (methodological approach) and (5) how you know when you have succeeded (success criteria). If you cannot answer these aspects, stop and revise your problem definition. If the flow could be more coherent, reflect on why or how you should solve the problem.

Decision Analysis tools can be highly beneficial. Employing tools such as Decision Trees can prove to be extremely useful, particularly when dealing with decision-making under risk. One pivotal aspect to consider is the decision maker's attitude, whether they tend to be optimistic, pessimistic, or moderate, particularly it influences decision-making under uncertainty. Multi-criteria decision-making (MCDM) methods are also extremely useful while making decision in the presence of several available alternatives/options that are to be evaluated based on various criteria.

Complex decision-making problems often involve multiple interrelated variables and constraints. Problem structuring allows you to decompose the problem into smaller, more understandable parts. By breaking down the problem into manageable components and clarifying the relationships between them you can focus on addressing each component individually, which can make the overall problem more tractable.

Scope: what is included within the system and what exceeds it. It's crucial to define the limits for later analysis, particularly when validating a problem. For the components that remain within the scope, it's important to determine the level of detail. This could range from very aggregated variables to highly detailed ones.

One tip I've found incredibly useful in structuring problems is applying the concept of "frame-shifting." It involves looking at the problem from various perspectives and challenging our initial assumptions. This approach can reveal hidden opportunities and risks, and can drive creativity and innovation in finding solutions. By incorporating frame-shifting into our problem-structuring phase, we encourage a broader and more flexible thinking process. This can lead to more robust and adaptable solutions, especially in complex or uncertain decision-making environments.

Formulate first, then solve. This should be one and only approach when it comes to deterministic and stochastic OR problems. Objective, decisions and constraints are main building blocks. Uncertainty in parameters is super important to consider in the case of stochastic models.

Se podría partir de los modelos clásicos identificando el problema a solucionar: Problema de transporte, de dieta, problema de corte, problemas en redes. Esto es una gran ayuda para iniciar con la estructura de formulación del problema.

In light of interest for publications, some try to limit themselves to solve a problem using a pre specified method. We always need to keep in mind that for someone to be able to solve a problem effectively, they need to have knowledge on various methods and techniques, and the ability to identify the best method for solving a problem on hand

Here are some mathematical programming solvers: -CPLEX -Gurobi -Xpress -CBC (Open Source) -GLOP (Google's Linear Programming solver)

Another open source alternative that I have used as a solver both in a research project but also for personal use is "GLPK". Works very well for both LP and MILP problems. It actually solves smaller optimization problems faster than commercial solvers such as Gurobi. However, for larger problems Gurobi is faster.

Remember always, it is not about the tools, at least not primarily. Sometimes, by relaxing the problem (e.g. most well-known is using reals instead of integers) you can already get a good idea on what the solution will look like. Often this will lead to solutions that are “good enough” rather than optimal. If not, think of heuristics that help you get closer, until it is satisfactory. And then there is the whole field of meta heuristics…

One of the best software tools is Python using libraries such as “NumPy” and “Matplotlib” to conduct graphic Monte Carlo simulations across two or more variable decision simulation scenarios. Another is the R package using the “IPsolve” package to calculate Linear Programing problems with a well defined objective (Desired output) provided with the corresponding constraints.

Si el modelo es multiobjetivo se podrían utilizar modelos de optimización generativos, como algoritmos genéticos, PSO, Ant Colony Optimization, Simulated annealing. Para tener valores cercanos al optimo.

The best tool to use is your brain. Too many people are becoming dependent on things like ChatGPT and can no longer have their own thoughts. Tools should supplement your own critical thinking.

Also, consider using cloud-based BI tools such as Amazon QuickSight to seamlessly visualize your outputs. It integrates with well-known data sources.

Mixed Integer Linear Programming and Goal Programming address most of the optimisation problems. Besides, Queuing models are important but it is crucial to understand the arrival and service time distributions. Understanding Erlang distribution theory can help in applying the right model. Simulations do away with this complexity but runs matter. Complexity increases when there are capacity constraints and probabilities associated. Heuristics are applied for NP hard problems. Understanding Genetic Algorithms is essential for implementing such problems. Non-parametric approaches such as Data Envelopment Analysis is simple and useful to compare inter-organisation performance and efficiency. Inventory and project management tools work well

Once the key metrics and important features are identified for a problem, it is crucial to analyse how those behave in the solution provided by the model. A full-fledged solution to a business problem is when one can give a detailed comparison between the benefits that optimization models have added as opposed to when they weren't present in the organisation. These could be just in terms of high level numbers or charts of how different features are behaving in the pre & post scenarios. Another important aspect is to also involve the users or stakeholders who can share the business insights and validate if the model's solution is actually making sense as per their domain knowledge & experiences.

Consider using tools like Tableau to creatively analyze the output of OR models. This tool can show data in ways that help the average person really understand the implications of the analysis.