Heuristics vs. Linear Programming (LP): Why Heuristics are better for solving complex supply chain planning problems in a Tactical time Horizon?

“Which  is better; LP or Heuristics?  Wouldn’t a LP solution yield the optimal plan whereas heuristics results come somewhat close?”.

We keep getting these questions.  My answer is that in most short-term planning situations, where plan results have direct impact to execution, Heuristics based solutions have far greater advantages over LP based solutions. Here are some of the advantages that I have seen:

1. Modeling Realistic constraints:  Heuristics based algorithms offer much more flexibility in adding detailed constraints to the model.  Although modeling such constraints is possible in LP, getting the results in a realistic time would be a challenge.  Most solution vendors that provide LP based solutions make assumptions and simplify the problem.  Consequently, the plan is no longer representative of the customer’s problems and business.  One of the common simplifications is modeling discrete manufacturing as non-discrete manufacturing, because of the limitations of linear programming.  Moreover, details and modeling of constraints like Campaign planning, WIP management, shipping calendars, are far from reach for a LP based solution.
2. Transitional Stability:  Most LP based solutions start from a clean slate, but in most instances, such planning results causes a number of problems:
   1. There is a lot of variability in the output.  Huge variances from plan to plan in many instances, cause organizations to lose faith in the system
   2. Interaction with other systems:  Planning results should take input from other systems into consideration.  This includes schedules created by scheduling systems/ MES, as well as manual modifications
3. Root Cause Analysis:  LP is black box solution.  It is difficult, if not impossible to tell the specific reasoning for a certain result.  I have seen a number of instances where planners lose trust in the system because it is not easy to explain why the results are a certain way.  Secondly, since the results cannot be explained in an easy manner, it is difficult to assign accountability to the plan results.
4. Performance:  LP algorithms take significantly longer to run than heuristics based solutions.  Hours compared to minutes.  With the speed of a heuristics based solutions, planners have much more flexibility to run a number of “what-if” scenario analysis using varying parameters such as demand, supply, capacity, etc.
5. Business Workflows:  Heuristic solutions enable a level of interactive planning that is not possible for a LP based solution.  Workflows where users review the plan, freeze a certain part of the plan, perform quick reruns, etc are not a viable option in a LP based solution.

In conclusion, in the short horizon the key is not which solution yields a more optimal solution. Any plan is no longer perfect the moment it is generated, due to the continuously changing circumstances. It is more important, to go with a solution that helps you to proactively plan and change course with the changing conditions.  To this, I contend that the solution needs to contain speed, integrated analytics and modeling flexibility.

So the question is, is there an instance where LP based solutions are better suited in supply chain planning?  The answer to this question is yes.  LP based solutions are generally suited for longer term planning where the objective is to determine major investments, product strategy, long term supplier contracts, Asset Optimization etc.  This is the topic of discussion for another blog. Stay tuned.