THE QUANTITATIVE SUPPLY CHAIN IN A NUTSHELL (LECTURE 1.2 SUMMARY)

Though quantitative supply chain (QSC) and mainstream initiatives both seek to generate the best business decisions (and financial returns), the former deviates from the latter in several consequential ways. These distinguishing features, as outlined in Lokad’s Supply Chain Manifesto, summarize the core principles that guide Lokad’s approach to supply chain optimization. Beyond software intervention, QSC advocates an overall mindset recalibration – one that refocuses attention on the more important though less immediately visible forces that actually exert the greatest influence on supply chain.

Watch the Lecture

All possible futures

By definition, a supply chain initiative is an attempt to identify and serve future demand. The problem is that the future (in all contexts) is inherently and irreducibly uncertain; there is an extraordinarily wide range of possible future outcomes, each of which possesses its own probability of occurring. The crux is not every outcome is equally probable. The same is true for supply chains, however, traditional solutions to the problem of demand uncertainty, such as time series forecasts, simply  dismiss uncertainty. Instead, traditional solutions focus on producing a single future demand value, later fortified with a predefined safety stock formula.

This approach fundamentally ignores the multitude of possible future demand values, leaving a business completely exposed if demand does not meet expectations. QSC, however, embraces uncertainty and identifies all possible future demand values (with nonzero probabilities). This insight is the product of probabilistic demand forecasting, itself the foundation of QSC, and provides a far more detailed picture of future demand than the classic time series.

All feasible decisions

At its core, a business is the sum of an extraordinary array of decisions and constraints. In terms of decisions, a business must contend with choices at both the macro- and microlevel; nearshoring a factory is a significant macrolevel decision, while choosing to increase or decrease one’s purchase quantity by one unit represents a routine microlevel one. Each decision comes with its own opportunity cost - one cannot spend the same dollar twice - and consequences - how it impacts the business, directly and indirectly.

Generally speaking, a supply chain practitioner encounters a greater volume of microlevel decisions than macrolevel ones. These microlevel decisions are often the most mundane ones, yet they represent a troubling layer of complexity, further compounded once a business considers its constraints (not to mention its suppliers’ and clients’). These can be minimal order quantities (MOQs), economic order quantities (EOQs), lot sizes, available shelf space, expiration dates, etc. In the presence of these myriad parameters and the specter of future uncertainty, the concept of a perfect supply chain decision is fanciful at best.

Rather, QSC seeks to identify all feasible decisions. In this context, a decision is “feasible” if it is immediately actionable, which means it is fully compliant with a business’ constraints. Ranking these feasible decisions (in search of the most optimal one) requires not only a sophisticated grasp of a business’ constraints, but a very granular understanding of its economic drivers.

Economic drivers

Overarchingly, QSC prioritizes reducing dollars of error over increasing forecast accuracy. Though perhaps counterintuitive, a more accurate forecast does not, in and of itself, automatically translate to greater profit or business performance. For example, one could guarantee a 99.99% service level simply by ordering vastly more stock than one could conceivably sell. In terms of customer satisfaction, the business is a success. However, this policy would result in colossal write-offs, adversely affecting the company’s bottom line.

Thus, to one degree or another, there is an ineluctable tradeoff between higher service level and economic return. QSC not only focuses on reducing dollars of error, it takes an even more fine-grained economic view, factoring in both first and second order drivers. First order drivers can be considered the immediately obvious and ordinary ones commonly found in accounting ledgers and mainstreams ERPs: cost of materials, gross margins, carrying cost, etc. Second order drivers are more nuanced, less immediate, and entirely absent from traditional enterprise software. These drivers represent the second-order effects of one’s decisions, and are a more abstract class of concern.

Consider the downstream effects of a stock out event. In a B2B context, a company may have contractual penalties for these situations, which represent a clear financial incentive to avoid missing service level targets. In a B2C context these incentives are far less clear. There is no explicit service level agreement between a business (for example, a supermarket) and its customers, thus there is no traditional mechanism to measure the impact of a stockout event. This may lead some practitioners to underestimate – or completely dismiss – the negative consequences of not having enough milk on the shelves.

QSC, however, argues that stockout events for some SKUs carry unexpectedly high financial impacts, and these are disproportionately high relative to their direct margin contributions. In other words, some items, such as fridges, are typically bought in isolation. Others, like milk and bread, are typically bought in baskets, i.e., in combination with other goods. Thus, the unavailability of certain SKUs may influence a customer’s overall purchasing decisions.

For example, a person might be perfectly happy to wait for their preferred fridge model to be in stock, but a lack of milk in a store may cause the same person to leave and complete their grocery shopping elsewhere. These latter SKUs, though perhaps not significant margin drivers in a direct sense, have significant inventory worth given their indirect value: they facilitate the sale of other goods. Therefore, in this example, the stockout penalty for milk is not limited to the milk itself; it includes the loss of all the other items in the basket.

In QSC, this less obvious value is expressed as stockout cover (a reward driver), and is factored into prioritized inventory policies¹.

Control requires automation

Once a company has identified all possible future demand values, considered feasible decisions, and ranked them with respect to all their economic drivers, the next step in QSC is to fully automate the supply chain decision-making process (or, at minimum, autogenerate recommended decisions). This automation stands in direct opposition to common practice, namely departments of clerks with spreadsheets.

In reality, a supply chain is a densely distributed system of actors (e.g., wholesalers, suppliers, customers), constraints (e.g., lead times, budget, service levels), and external forces (e.g., seasonality, natural disasters, competitors’ prices). Expecting a human mind (or even a team of minds) to contend with all of these variables for even a single SKU is simply unreasonable, never mind for a catalog of thousands of SKUs for multiple stores.

Furthermore, any attempt at innovation within such a framework is destined for bureaucracy and costly retraining, both of which will produce delays and inefficiencies. On the other end of this spectrum, QSC seeks to implement an end-to-end numerical recipe that generates all the trivial, mundane supply chain decisions for operational management. These are the kinds of decisions that consume too many dollars of attention, and misdirect bandwidth from far more pressing concerns.

QSC, as such, treats supply chain as an asset rather than an expense; it is a process that should be optimized (and automated) in order to yield its greatest value².

The Supply Chain Scientist

A piece of supply chain software, no matter how impressive, cannot govern itself, let alone take responsibility for the results it generates. The effectiveness of a numerical recipe is, in fact, limited by the expertise of the data scientist implementing and monitoring it. At Lokad this role is performed by the supply chain scientist (SCS).

An SCS is charged with, amongst other things, processing the data for the QSC initiative, and taking ownership of the successful implementation of the numerical recipe. Establishing a valid data semantic (what the data actually means) requires considerable skill, as the success of the QSC is predicated upon not only processing data, but making sense of it in the first place. For all the advancements in AI, this is still a human-led process.

For example, analyzing simple historical sales data may seem relatively straightforward, but this dataset may be misleading given any number of nested and overlooked factors. The data may unintentionally contain promotions, thus it does not reflect the true demand for full-priced goods. Alternatively, the history may contain returns, giving yet another false impression of demand. The term quantity per day is also subject to any number of interpretations; it might reflect the day a sale was made, or the time a pre-order was accepted, or when the client’s payment was received. This says nothing about the additional complexity a company’s ERP might introduce to the process.

All of this is to say, making sense of the data is tricky, and requires a highly-trained supply chain scientist to take charge of the process, as well as oversee the day-to-day running of the numerical recipe³.

Notes