Industry Group vs. Complexity

The specific challenges and solutions required for supply chain transformation vary from company to company, but for modeling and optimization specifically, there is a step change in complexity between three high-level industry types with process industry companies being the most technically complex.  I define the different industry groups as follows:

• Resellers / Distributors: Either distributors or resellers that purchase goods for resell with no conversion done to the product being sold.  Retailers are the biggest subcategory of this group but the key characteristic is no in-house manufacturing of products is involved.  Retailers with their own branded line(s) of products that are produced by third parties would still fall into this category, especially where those branded products are a small percentage of total SKUs carried.  (Examples: hardware/home improvement retailers, department stores, grocery stores, etc.)  While the focus of this article is primarily on manufacturers and the accuracy of production models within S&OP/IBP and detailed scheduling solutions, I include resellers and distributors because they help to illustrate how the applicability of vendor solutions varies by industry group.

 

• Discrete Manufacturers: These are manufacturers with production processes where the finished goods are assembled from individual pieces or components.  Also, components involved in discrete manufacturing can be counted and itemized at any part of the process whether as raw materials, work-in-progress, or finished goods.  Assembly also necessitates ample factory floor space where more human labor (or robotics) is typically required throughout the process.  (Examples: appliances, automobiles, computers, etc.)

 

• Process (or Continuous Flow) Manufacturers: These are manufacturers with production processes where finished goods are derived from recipes or formulas and raw materials are blended, cooked or chemically transformed as solutions or batches.  These solutions or batches also commonly move through the production process via networks of pipes and tanks which are often controlled and monitored from centralized control rooms.  These production environments require a relatively limited amount of human labor amongst the pipes and tanks and then only to adjust, inspect, change-over, or maintain the physical plant as needed.  However, even process industry products usually end up in the form of discrete products toward the end of the manufacturing process as the items take their final form or are packaged for transport and sale to customers. (Examples: processed foods and beverages, chemicals, pulp and paper, plastics, etc.)

It is important to note that most process industry manufacturers technically have a ”hybrid” manufacturing operation (i.e. part process, part discrete) as many of these products continue to undergo some processing or packaging even after the work in progress (WIP) recipe, formula, or solution is transformed into discrete units.  For instance, in producing a Hershey’s Kiss, the blended chocolate recipe is cast into individual Kisses which continue through discrete-like processes as they are wrapped, bagged, and packaged.  Alternatively, melted plastic resin or composite mixtures are molded or extruded & cut into discrete units which may continue to undergo cutting, trimming, or recombination until they are in their final form and are packaged for transport to customers.  For our purposes, let’s call the point in a process industry production process where the WIP goes from continuous form (batches or solutions that flow through tanks and pipes) to discrete units as the “switchover” point.  We’ll come back to this term a little later.

The complexity of models required to drive multi-factor optimization and objective-driven scenario modeling differs among these three industry groupings.  If we were to place the inherent technical complexity of required optimization models on a scale from low to high, resellers and distributors would be on the lowest end of the scale, discrete manufacturers would be in the middle, and process manufacturers would be on the highest end of the scale.  Not that resellers and distributor networks are simple to optimize, but the models and the inherent sets of constraints, logic, and interdependencies required within them are less technically complex.  One might even substitute the word “technically complex” with the word “scientific”.  Likewise, the models that process industry manufacturers require are more technically complex (or “scientific”).  (See Diagram 3.1)

 

Diagram 3.1 – Optimization Model Design Complexity by Industry Group

 

So what exactly does technical complexity really mean?  Let’s take a company like Home Depot, which would fall into the reseller/retailer grouping.  With so many products, vendors, stores, and distribution centers, the solution or model that helps optimize their supply network will be complex on some level.  But in a rudimentary description, this type of solving and modeling complexity is primarily driven by sheer numbers—numbers of stores, number of SKUs, SKU/quantity allocation, etc.  The type of complexity refered to in this article involves the logic within the model.  The logic contained within a model that optimizes a relatively small process-industry plant can be more technically complex than one used to optimize the product movement and inventory for a large, national retail chain.  So let’s drill down another layer.  What exactly is meant by “logic”?  Let’s say the model has to accurately account for many factors where chemistry, flow options, switchover points (i.e. conversion from flow to discrete), etc. pose unique considerations.  Some of these include:

• Curves, Formulas, & Recipe Variations: Portions of production operations may need to vary from batch to batch to produce the exact same recipe or formula. The changes may be driven by slight differences in raw materials, environment, or plant operating conditions and the functions that control them may be depicted graphically by curves rather than tables or step functions.
• Routing: A portion of the downstream production line, say a bottle filling line in a beverage plant, could either (a) be fixed to a single source tank containing the beverage formula, (b) have maximum flexibility to be instantly rerouted to tap any one of many source formula tanks within the plant, or (c) have a degree of flexibility that falls somewhere in between (a) and (b)—for instance, bottling line “A” can only connect to tanks 1, 3, and 4 out of tanks 1 through 10.  This type of routing constraint can also appear at multiple sequential points in a production process.
• Changeover/Flush Factors: When switching from one product to another, depending on the product and plant configuration, a certain volume of material at the “end” of a batch and at the “beginning” of the subsequent batch may be lost to waste due to flushing or cleaning.
• Parallel Dependencies: Two or more downstream production lines may be fixed to a single upstream material/formula supply tank enabling the downstream lines to function only semi-independently.  For instance, the downstream lines can produce different SKUs at the same time as long as the different SKUs are reliant on the same material supplied by the one tank.
• Coproducts and Byproducts: These products are sometimes a key consideration in process production planning as they may be sold independently and thus become part of the demand signal or revenue potential that supply chain planning solutions are solving for.
• Recursive Interdependencies: Outputs (coproducts or byproducts) from a production batch (usually a mixture or solution) may become a required input (raw material, WIP, or fuel) for a subsequent batch on the same production line.
• Time-Flow Constraints: Typical process production operations cannot be paused for an extended period of time without damaging the WIP in the tanks and pipes.  In contrast, a typical discrete operation can hypothetically pause for hours or days and then resume with minimal impact on WIP.  Once started, a process production batch typically must continue to completion or the WIP becomes waste or scrap.  Pipe flow rates and cooking or chemical reaction dependent hold-times may also factor into the times and rates for a given batch of production.

These and other factors drive more advanced design requirements in terms of how the software represents each production point, the relationship with preceding and successive points, and how it performs its solving calculations.  However, most supply chain planning solutions (even those offered by some of the biggest and well-known vendors) do not contain the advanced logic required to even closely model these or other types of process industry constraints.  As the required technical complexity of a manufacturing environment increases, the number of vendors whose solutions can appropriately model the environment decreases.  By the time the scale reaches the high level of technical complexity characteristic of many process industries, the vendors with applicable, well-fitting solutions become the exception rather than the norm.  Diagram 3.2 illustrates this dynamic.

 

Diagram 3.2 – Optimization Model Design Complexity vs % of Capable Vendor Solutions

 

The dotted line representing the percent of vendors is illustrative, but the curves generally reflect the relative quantity of vendor options by level of required technical complexity.  The curve drops off gradually as it moves to the right through the low complexity/resellers and distributors range.  It continues to drop but more precipitously as it continues further through the moderate complexity/discrete manufacturers range, and then drops sharply and begins to level out again as it enters the high complexity/process manufacturer range.  While there are dozens of vendors that offer detailed scheduling and S&OP/IBP layer solutions overall, if a manufacturer that falls in the right-most quartile, there are relatively few vendors with solutions capable of accurately modeling the environment.

As discussed in Part 1 / Principle 1, when a manufacturer implements a technology that is incapable of accurately modeling key constraints in the production environment, an “artificially adapted model” is created to make the solution work given the manufacturer’s typical operating norms.  Diagram 3.3 provides a visual of the relationship between vendor capability and the “artificially adapted model”

 

Diagram 3.3 – Complexity That Exceeds Solution Capability & Artificially Adapted Models

 

In this diagram, I’ve maintained the industry grouping bars but moved them to the bottom of the chart for reference.  In the top portion of the chart, a vendor is represented by a horizontal bar extending from end to end of the technical complexity scale.  This bar represents the fact that a vendor could conceivably sell and implement their solution to any company.  However, there is a point at which a vendor’s solution reaches its limit in terms of technical complexity and fails to accurately model key production constraints and options.  This zone, indicated by red shading, correlates to the yellow “Artificially Adapted Model” area from Diagrams 1.3 and 1.4.

To further illustrate this dynamic, Diagram 3.4 provides an example where solutions by three vendors (A, B, and C) are matched to three different companies (1, 2, and 3), with each company requiring a different level of model technical complexity.  Company 1, a reseller with a lower model technical complexity could choose either of the three vendors and have good results.  In reality, the solution offered by Vendor C might be viewed as overkill, Vendor A or B, targeting this industry and type of operation, might be deemed a better fit, or Vendor C may even have a separate application that is more suitable for other challenges experienced by Company 1.  However, all three vendors could accurately handle the modeling and optimization if needed.  Company 2, a discrete manufacturer, would be okay choosing either Vendor B or C, but if they choose Vendor A, they would run into modeling limitations and an artificially adapted model would be used in an attempt to compensate for limits in Vendor A’s solution.  Company 3, a process manufacturer with the most technically complex modeling environment of the three, would only be okay choosing Vendor C.  Both Vendors A and B would both pose modeling limitations an artificially adapted model would again be used in an attempt to compensate for limits in the solutions offered by Vendors A and B.

 

Diagram 3.4 – Vendor Capability vs. Manufacturer Complexity Example

 

Incompatibility of a vendor solution to a given production environment will sometimes be clear, particularly if there is a tremendous gap, and a vendor will decline the opportunity or the manufacturer will rule out the vendor.  However, there are at least as many situations where the incompatibility is subtle and the manufacturer does not perceive a significant problem or a problem is not detected at all.  In these cases, the manufacturer that makes such an ill-fitting choice doesn’t always notice a blatantly failed implementation although some do.  What they get is a presumably functional implementation, but under the hood, the model has been artificially adapted (or hacked) to work within a limited range of typical operations and business objectives.

This becomes more obvious if, at a later date, a user tries different objective functions, seeks a more granular level of decision-making in the production process, or business circumstances drive a change in scenarios and priorities.  In such cases, the user will eventually hit a brick wall where the technology is incapable of supporting the new demands.  This is because the model was artificially adapted or force-fit and is thus inflexible or otherwise incapable of accurately modeling and optimizing scenarios under a new set of parameters or conditions.

It would be understandable to think: “Just make sure buyers appropriately define requirements and ensure the vendor can meet those requirements and they should be set, right?”  Hypothetically yes, realistically no.  The issue is that requirements wording is slippery even in simpler cases.  When that is combined with the highly nuanced modeling, mathematical, multi-factor optimization, and software design aspects of this problem, it becomes difficult to write requirements that clearly discern these types of capabilities in a concise manner and that many buyer-side screeners or even vendor-side sales teams will reliably grasp.  What usually happens is once the requirement is boiled down to a 1-2 sentence bullet using terms that stakeholders who are not well-versed in multi-factor constraints, optimization, and software design fully appreciate, it’s typically imprecise enough that reviewers and vendor sales teams alike perceive (or conjure up) enough justification to consider the requirement met, even if it is not.

The common and imprecise use of the term “optimization” as applied to solutions that actually use heuristics is a good example of this dynamic.  It can appear to be an inane distinction that’s ripe for dismissal, but the implications are significant, particularly when combined with the linear/forward-only planning that usually accompanies solutions that use rudimentary heuristics.

 

Real Technology Selection Implications

To take this a step further and highlight a key issue that may be unintuitive to some supply chain leaders and technology decision makers, I’ll illustrate this point with a handful of common vendors and compare them to some recent Gartner Magic Quadrant ratings.  In Diagram 3.5, I’ve adapted the vendor technical complexity limitation chart to include actual vendors and my view of their approximate limits on the modeling technical complexity scale.

 

Diagram 3.5 – Example of Select Vendors and Approximate Modeling Limits

 

Placement of vendors on this scale is based on many years of my first-hand experience in screening, implementing, or refining vendor solutions as well as discussions with other implementation professionals and users/stakeholders at companies where they were implemented.  These ratings also consider the combined capabilities for both S&OP/IBP/scenario modeling and detailed scheduling and are therefore weighted heavily on the ability to both plan and execute.  The listed vendors can have other strengths or limitations beyond this focused perspective.  This chart is not intended to provide a thorough assessment of all vendors although I may seek to accomplish that in the not-too-distant future.  For now, it highlights only select vendors to illustrate the central point of this article.

This vendor list is especially condensed in the low-to-moderate modeling technical complexity ranges since there are dozens of vendors that could be listed there.  In the high modeling technical complexity range, the vendor list is not as significantly condensed since there are relatively few vendor solutions (probably in the mid-single digits) that can adequately support these environments.

I’ll also point out that in the “Discrete Manufacturer/Moderate Complexity” range there is a broader range of limits where some vendors reach these limits further to the left (low-moderate) side of this range while other vendors reach their limits closer to the right (high-moderate) side of this range.  To further highlight this point, SAP’s PP/DS and IBP solutions together contain enough limitations that they hit their limits in the low-moderate portion of this range.

Lora Cecere of Supply Chain Insights, LLC wrote an insightful piece titled “Three Reasons Why SAP Supply Chain Planning Is a Risk to Your Business” (link here) where she discusses problems with SAP’s supply chain planning and strongly recommends going with “best-of-breed”.  While Lora wrote this article in 2015 and a few nuances of the landscape have changed since then, the essence of that commentary still holds true.

Diagram 3.6 provides a listing of Gartner’s most recent supply chain planning-related Magic Quadrants and the vendors landing in the “Leader” quadrant for each.  Comparing Diagrams 3.5 and 3.6 reveals key differences in vendor modeling capability versus their appearance as a Leader in Gartner’s Magic Quadrants.  Keep in mind that this modeling capability is a proxy for how well the technology will plan and execute in each environment.

As I mentioned earlier, Gartner Magic Quadrants for S&OP/IBP and Supply Chain Planning were at one time separated into industry groupings—similar to the ones I’ve listed in Diagrams 3.1 through 3.5.  But, for the last 10+ years, that distinction has been dropped and the narrative is generally silent on these differences.  As a real-world example of the implications, a smaller regional retailer, an appliance manufacturer, and a chemical or pulp manufacturer will have drastically different levels of technical complexity in their operations modeling, and thus different requirements when choosing a solution for either the detailed scheduling layer or the S&OP/IBP/scenario modeling layer.  Yet, I’ve seen many organizations using the Leader section of the Magic Quadrant as a vendor selection starting point when it is possible, or even likely, that many vendors selected from that quadrant could land them in the “red zone” of Diagrams 3.3 and 3.4.

 

Diagram 3.6 – Gartner Magic Quadrant Leaders for S&OP and Supply Chain Planning Technologies

 

Is Gartner incorrect in calling these vendors “Leaders”?  It depends.  I’d say that there is a critical dimension—the technical complexity required in modeling the operation—that Gartner remains largely silent on.  The result is that vendors may be “Leaders” according to Gartner’s stated definition, but those “Leaders” may be well-suited only for a subset of less complicated operating environments and could be lacking in more complex manufacturing environments.  That begs the question that, if a vendor’s application is intended to address a given industry group but is limited in terms of its ability to accurately model many environments in that industry group, should they even be considered a “Leader”?  I’d argue no, especially in this day and age as most industries have their sights on digital transformation and Industry 4.0.

To be clear, my objective here is not simply to criticize either SAP or Gartner.  To the contrary, I believe they are both good companies with many good people doing good work.  However, they are not infallible and there is a perception—too often inadvertently seconded by Gartner—that some of these solutions are far more capable than they really are while under-representing capabilities of other vendors whose solutions are quite advanced and well thought through.  This results in buying decisions that lead to problematic implementations and disenchanted stakeholders.  Not to mention the sneaking perception among business leaders that “this is as good as it gets” with supply chain planning capabilities and results.

A 2019 S&OP study performed by Supply Chain Insights LLC found that the two biggest gaps that organizations with S&OP processes experience are (1) “The ability to run what-if analyses to determine alternatives”, and (2) “Manage opportunities and risk analysis”. (See Diagram 3.7)   The two capabilities are among the first casualties of the modeling and optimization problems highlighted in this article.  Meanwhile, the fourth biggest gap is “Use technologies to determine the next most profitable plan”.  That this gap is slightly smaller than the others, though still large, comes as no surprise since even ill-fitted technologies are implemented to provide a plan–they just may not do it well or have extensive limitations.  And whether they provide the “next most profitable plan” vs. the next best sub-optimally profitable plan the technology could resolve depends on the technology’s model and solving logic, but in many cases it will provide the latter.

 

Diagram 3.7 – S&OP/IBP Capability Gaps

 

With Industry 4.0 and digital transformation being the prominent trend for the foreseeable future and eventually becoming a benchmark in terms of supply chain excellence, technology vendors that profess to offer solutions addressing the detailed scheduling and S&OP/IBP/scenario modeling layers should be able to more accurately capture and model these operations or be held to account for not doing so.

The features of Industry 4.0 are based on a holistic approach to manufacturing where the physical and digital are connected, providing business stakeholders with greater control and insight into all aspects of the operation.  And the promise of Industry 4.0 is that business stakeholders can leverage this connectedness to boost responsiveness and flexibility, increase productivity, improve processes, and drive growth.  But at the very foundation of these features and promises lies the need for solid fit and accuracy in modeling the processes, flows, options, and constraints involved in the manufacturing operation.  This leads to Principle 4.