What does the Solution Space look like
– Problem solvers and analysts use the term “Solutions Space” to describe the graphical
relationship between the inputs (levers) they experiment with in their analysis and the
outputs (results) achieved .
– Originally, people thought of the solution space as being line ar .
– Then, our thinking evolved to understand that the solution space is no-linear .
– Lately, our thinking evolved to understand that the solutions space is not just
non-linear, but actually has multiple “local optimals”
– Add to that , the understanding that the solution space is not two dimensional. There
are numerous factors (operational levers) that impact output, resulting in a
multi-dimensional solution space. Picture the solution space as a “Mountain Range”
(pretty intimidating to think of it that way).
Reasonable Expectations of the Solution Space
– With any kind of analysis, you can never know the “Optimal” optimal solution, we can
only discover (through analysis or experience) a number of “local” optimal solutions.
– The number of “local” optimal solutions we discover is limited by our perception of
reality, and our understanding of the operational levers (dimensions) impacting the
Solution Space Challenges
– First, to identify which of the “controllable” operational levers (solution space
dimensions) have the greatest impact on operational performance (output).
– Second, Narrow this list of operational levers to a manageable number. This is optimally
two to three (at most five) levers.
– Third, experiment with this manageable number of levers to identify the “Best Practical”
Solution Space Strategy
– If cost effective, first simulate the operations to 1) understand process behavioral patterns, so
as to 2) identify Best Practical Solution(s).
– If cost effective, follow up the simulation with an optimization-based analysis, so as to
identify as many Local Optimals as possible.
Note: when I use the term “simulation”, I am talking about what would be considered a
“statistically accurate” simulation. This implies that:
1) you utilize input Distributions,
2) run the simulation model for multiple replications,
3) you eliminate non-steady state statistics from the output results,
and 4) you develop Confidence Intervals for all critical output statistics.