Success Impaired: Common Pitfalls of S&OP and Supply Chain Planning Technologies...And How To Avoid Them

Part 2: The Subtle Blast of a Solver’s Approach

robot ai figuring out problem 4in
Kenton Phillips

In the prior section, we discussed modeling in greater detail, but there is a separate but closely related piece of the planning and optimization puzzle that stakeholders must consider to ensure a technology is appropriate for a given production and planning environment.  Let’s call it “solver logic” which refers to the logic by which the system sorts through all available options to arrive at the recommended plan.

Solver Logic

While the model defines the time and situation-dependent parameters at every step in a production process, solver logic determines how the system works through those parameters across all production steps to arrive at a recommended scenario.  More plainly, within the parameters defined by the model, the solver “solves for” the best outcome given a user-defined objective.

There are two sets of attributes that define how a solver works—I’ll refer to them as the solving method and solving (or planning) direction.  The two attributes are closely intertwined and sometimes inseparable, but it is important to identify and distinguish the two since they both have substantial impacts on planning results and financial outcomes.  Also, either of the two attributes can serve as a yellow flag to a buyer if the other attribute is somehow overlooked initially.


Diagram 2.1 – Solving Methods


A solver can use either heuristics methods or mathematical optimization methods to develop a recommended scenario, plant schedule, or supply chain plan.  The differences in operational and financial results generated by the two can be significant.  The two methods are defined as:

Heuristics:  A heuristic is a practical problem-solving method that relies on “rules of thumb” to produce a fast and feasible solution that may be good enough to solve a problem, but it does not necessarily provide an optimal solution.

Optimization:  Optimization (sometimes referred to as mathematical optimization) considers all decision variables, constraints, and objectives and generates a globally and mathematically optimal solution.

A heuristics function might be as simple as logic suggesting it is generally better to do A before B, and B before C and C before D (resulting in A->B->C->D) and so on.  But if circumstances create an exception to that rule to such that it is now most optimal (i.e. least costly, most profitable, etc.) to flip B and C and instead plan for A->C->B->D, the simplified heuristic approach might chug along with it’s rule of thumb, never “knowing” that flipping B and C provides a better outcome.

Alternatively, a heuristic might use a changeover matrix or table to create a more informed sequence based on changeover cost or time which is better.  However, this prioritization often occurs independent of other factors and other production points.  This can be problematic because a changeover matrix may suggest that a certain changeover sequence is best at that production point in isolation, but conditions at a different production point could create a situation where it’s better to accept a moderately higher changeover cost or time at this production point in order to achieve a better overall result. This is some of what isolated heuristics will miss.  Mathematical optimization considers all points and all objective functions simultaneously to arrive at a truly optimized result for the whole system.  

“In the past, many companies opted to use heuristics because they were seen as both easier and faster to implement and able to quickly provide “good-enough” solutions. Historically, mathematical optimization technologies were often viewed as too complex (requiring specific technical expertise) and not able to scale well for real-world problems.

This perception has changed immensely over the last few decades with the improvements in computing power and parallelization. In addition, the introduction of cutting-edge algorithms have made mathematical optimization technologies much more widely accessible and applicable. Now, we see a rapidly increasing number of companies across an ever-expanding range of industries (including supply chain, electrical power, finance, aviation, retail, telecommunications, and sports scheduling) applying mathematical optimization and achieving significant business benefits.

Still, many companies continue to develop and use heuristic solutions – which may be a part of legacy systems that have been in place for some time – to address their business challenges. More and more of these companies, however, are finding that heuristics are not able to meet their needs and are looking for an alternative algorithmic approach.”

– Gurobi Optimization

In demonstration models, optimization-based supply chain planning and scheduling methods regularly outperform heuristics-based ones.  Depending on the complexity of the problem, the differences in the outcomes of the objective function (e.g. minimum time, minimum cost, highest profitability, etc.) between the two methods can be significant.  Yet, if a system is using heuristics—particularly the more rudimentary forms of heuristics—the scheduling and S&OP/IBP scenario modeling results are likely leaving plenty of money on the table.


Diagram 2.2 – Solving (or Planning) Direction


Solving Direction refers to how the solver moves through the modeled production process to plan schedules and create scenarios.  You could also think of this as “planning” direction.  The two directions used include: “forward-only” (which is also referred to as linear, sequential, or asynchronous) and “bi-directional” (which is also referred to as forward & reverse or synchronous).

With forward-only planning, the solver addresses one production point at a time in sequential order (see Diagram 2.3).  Specifically, it examines production point #1 and creates a schedule or plan for that point.  The work in progress (WIP) schedule or sequence coming out of production point #1 becomes a fixed input upon planning production point #2.  Using the fixed, pre-determined WIP sequence from production point #1, the system then creates a schedule or plan for production point #2 which becomes a fixed input for planning production point #3. Using the fixed, pre-determined WIP sequence from production point #2, the system then creates a schedule or plan for production point #3, and so on.


Diagram 2.3 – Forward-Only/Asynchronous Planning Example


The key here is that as it moves forward through the production process creating schedules for each point, it only solves one point at a time and it cannot look “backwards” to adjust plans at an earlier production point even if it hits a constraint at a later production point that would normally prompt an upstream change.   It is common that a production environment will have a downstream constraint that would need to inform and drive a change to an earlier (upstream) production point to arrive at an optimal plan.  Sometimes, having that downstream constraint adjust the upstream plan is a necessity for the production plan to be feasible.   This approach not only produces sub-optimized results, but it can produce unworkable plans that must be adjusted manually.  Yet, many major vendors, including some of the biggest ones, have supply chain planning modules that use forward-only planning and deliver the limited results that come along with it.

Bi-directional or synchronous planning, on the other hand, factors in all production points simultaneously or solves in a manner where it moves forward and backward from point to point, reversing as many times as needed until the plan for all production points is feasible and optimized for the whole process. (See Diagram 2.4).  In these cases, the system could find a constraint at production point #3 that suggests that it needs to readjust the plan/schedule at production points #1 and #2.  When the final solution is created, the result is mathematically optimized and always feasible even where downstream constraints necessitated adjusting upstream plans.


Diagram 2.4 – Bi-Directional/Synchronous Planning Example

These two different attributes, solving method and solving direction, are commonly connected such that systems that use mathematical optimization will, by definition, have to use bi-directional/synchronous planning to create an optimal result whereas systems that use heuristics more commonly also use forward-only/asynchronous planning.  This concept is illustrated in Diagram 2.5.  There is an exception though.  It is possible for a supply chain planning system to use advanced heuristics and bi-directional/synchronous planning, but that’s relatively rare among today’s supply chain planning systems.


Diagram 2.5:  Relationship Between Method & Direction


As you may instinctively detect, there is a huge difference in the level of sophistication of systems that use optimization and synchronous/bi-directional planning vs. those that use heuristics and forward-only/linear planning.  By extension, vendor technologies that rely upon the less sophisticated techniques in this area tend to encounter greater limitations in their modeling capability as that same, less sophisticated programming logic is typically used throughout their detailed scheduling and S&OP/IBP solutions.  Likewise, given the more sophisticated software design required to enable optimization and bi-directional planning, vendor solutions that use these techniques tend to have fewer problems accurately modeling more complex environments as a similar level of design sophistication is typically applied throughout their detailed scheduling and S&OP/IBP solutions.

Implications of Different Solving Approaches

Now consider the two attributes together and the business outcomes they would deliver.  As you may appreciate, a solver that uses both heuristics and forward-only/asynchronous planning to create a production or supply chain plan incorporates sub-optimization on two levels—first in the forward-only direction of planning, then in the heuristics used to create the plan at each given production point.  When these methods are used together the potential sub-optimization effects are compounded, providing scenarios that can be substantially suboptimal and potentially unfeasible.

I began this article discussing potentially misleading guidance and information that is ubiquitous around supply chain planning systems.  To drive this point home, I’ll provide a comparison between SAP and Infor based on my opinion which is driven by my experience and ancillary discussions or interviews.  Note that I have no relationship with either vendor.  Likewise, I have no incentive to promote either vendor (beyond their solution functionality), and I have no agenda for or against either vendor.  I pick these two vendors in this example because (1) I am familiar with their supply chain planning systems and the logic contained within them, (2) they have a similar breadth of offerings including ERP systems, and (3) these two examples, in particular, help to illustrate the central point made in this article.


Diagram 2.6 – My View of Two Different Vendors


As shown in Diagram 2.6, SAP, a major vendor regularly positioned in the “Leader” quadrant in Gartner’s supply chain planning-related Magic Quadrants, offers detailed scheduling and S&OP/IBP solutions that use heuristics rather than optimization and a detailed scheduling module (PP/DS) that plans in a forward-only/asynchronous manner as opposed to a bi-directional or synchronous manner.  Technically, IBP could employ bi-directional planning and hand the plan off to PP/DS to deploy in an asynchronous manner, but there are still drawbacks with that approach.

Conversely, Infor, a major vendor not frequently mentioned in Gartner’s supply chain planning-related Magic Quadrants, offers detailed scheduling and S&OP/IBP solutions that use mathematical, multi-factor optimization and bi-directional/synchronous planning.  The Infor solutions have also used these methods for at least the last 15 years (including legacy companies that owned these modules before being acquired).

SAP has many great products, but they were a late entrant into the supply chain planning world with fairly entry-level supply chain solutions that came out only about 15-18 years ago.  Mathematical optimization and synchronous planning are not new capabilities in the supply chain planning world, they’ve been available and practical in many business environments since at least the mid-1990s.  Unfortunately, many vendors have met the challenge of supply chain planning and optimization with solutions do not adequately model and do not actually optimize sometimes leaving customers with underwhelming outcomes–even if improved.

Potential Future SAP Redemption

Since I beat up on SAP and Gartner here and a bit more in Part 3 / Principle 3, I would be remiss if I didn’t mention a potentially favorable recent update.

“Under the 10-Year Enterprise Agreement, SAP Confirmed Gurobi as the Premier, Long-term Supplier for Mathematical Optimization Technology” (link)

On May 13, 2020, Gurobi Optimization LLC, a company dedicated to providing mathematical optimization solvers, announced that they entered into a long-term partnership with SAP to enhance and expand the use of mathematical optimization across SAP’s enterprise application software suite.  This is promising for SAP longer term.  As I mentioned earlier, if a company uses mathematical optimization holistically, by definition, they have to use bi-directional/synchronous planning and they’ll hopefully incorporate improved modeling capability.  But this is unlikely to happen in a plug & play approach where optimization capability is dropped into PP/DS and IBP and quickly deployed.  It is more likely that SAP’s PP/DS and IBP applications will need to be completely redeveloped just to properly facilitate mathematical optimization.

That leads to another consideration – if SAP started overhauling these applications right away, it could take at least 1-3 years to design, develop, and test new detailed scheduling and S&OP/IBP applications and have them ready to hit the market.  Exactly which optimized enterprise applications SAP plans to roll out first, what functionality those applications will have, and when they’ll be ready will be unknowns or rough estimates for a while.  Even so, this announcement is a welcome, even if overdue, preview into potentially better supply chain planning designs from SAP in the future.

Due its length, this article is published online in multiple parts. Click on the sections below to view other parts of this article, or get the full PDF version immediately by requesting the download link below.


Success Impaired: Series Table of Contents & Navigation

> Next Section: Part 3 – Different Industry Groups Have Distinct Needs

< Previous Section: Part 1 – Realistic Scenarios Require Accurate Modeling