What are we trying to optimize?
It is essential to define the yardstick against which the various decisions will be evaluated. The aim may be purely financial (e.g. profit margin, cash flows, costs, …) or operational (throughput, stock-outs, market share, risks, treasury requirements …). The time frame and contours of the exercise should also be made clear (is the challenge operational, tactical or even strategic?).
Finally, the chosen objectives should translate into measurable business performance indicators. These KPI’s should then be used as criteria to evaluate solutions proposed by the model.
What are our main assumptions?
We decide how to integrate the diverse input into the model. A trade-off should be made here between capturing enough real- world complexity to ensure the model is robust, while keeping the problem manageable. These choices must be guided by the objective of the model and the decision level.
For example, when making a strategic review of production capacities for a network of plants, it might not be necessary to consider the detailed physics of the process. Aggregated cost data may be sufficient.
Finally, when some assumptions are highly subject to uncertainty, different scenarios can also be designed (e.g. variation of price, demand, weather…)
What are the improvement levers?
We identify and select the decisions to be optimized, those that could lead to a significant improvement in business performance. These decision levers can be of any kind (e.g. process control parameters, production levels, logistics design, purchasing mix, investments, selling prices…).
As these decisions form the output of the model, we make sure that they are implementable by management within the time frame considered. We also ensure that their major impacts on business are properly understood.
Do we have the necessary data to feed the model?
The objective and assumptions that shape the problem also define its data requirements.
Therefore, reliable information sources should be available: the company’s own data (e.g. cost structure, empirical process data) or industry-wide published know-how (e.g. physics of processes, market data…).
One should also be able to quantify this information as categorized data.
Finally, data should also be easily manageable (e.g. via interfaces with existing Information Systems such as ERPs, CRMs, Process Control tools …)