Have you considered there may be a better way to structure supply chain? I have sure have and it doesn’t look much like the linear Supply Chains of today. Why not? Because the Supply Chains of today suffer from delays, poor forecast accuracy and the silo mentality. The good news? Advances in technology like Machine Learning and AI can fix all this.

Traditionally, we view the Supply Chain as a chain of sequential links each with behavioral attributes which act both separately and together to cause demand variation from historical performance. The “linked” supply chain is characterized by linear time delays in communicating of variance to the forecast, and amplification in volume as demand is placed back from many demand points to fewer supply points. What is this known as? That’s right, the Bullwhip Effect.

As demand variations are communicated sequentially through the Supply Chain, the time delays and signal variations that cause error propagate throughout the network as per the following illustration:

Do linear Supply Chains really work? They are responsible for the negative bullwhip effect.

Statistical forecasts are based on historical data, and are not representative of causal factors associated with new product introductions

The Core Problems Of Linear Supply Chain

Planners are making daily multi-million dollar working capital decisions based on little more than spreadsheets and tribal knowledge.

Statistical forecasts are based on historical data, and are not representative of the existing and future conditions (causal factors) associated with new product introductions that influence demand. The departments responsible for creating, marketing, and selling (Demand Creation functions), do everything they can to change history. As a result, the forecast for sourcing, making, and delivering (Demand Fulfilment functions) is always wrong.

The time delays associated with S&OP processes exacerbates forecast error

S&OP Makes Bullwhip Effect Worse

Sales & Operations Planning (S&OP) is a noble attempt to capture causal factors. However, the time delays associated with S&OP processes exacerbates the error. The statistical forecast is only useful when determining a basis for segmentation, analyzing patterns, and creating baseline forecasts. As a result, statistical forecasts about Stock Keeping Units (SKUs) at the location level, for example, will always be inaccurate on a day to day basis.

The resulting culture accepts the inaccuracy, moves on, and no one identifies, collects or explains the reasons why the forecast was wrong and to consider those recurring causes in the future. To improve forecast accuracy, organizations need to collect and study the causal factors that are likely to increase or decrease demand, across the business, to determine and manage variability from baseline demand, for each item, at each location.

In the past, we theoretically knew the impact that promotions and controllable factors would have on the forecast, especially in consumer goods where syndicated data and analytics are readily available. However, from an operations perspective, obtaining the marketing and promotional plans was cumbersome and arduous.

To improve forecast accuracy, organizations need to collect and study the causal factors that are likely to increase or decrease demand

Improved Forecast Accuracy Is Here – But We Need A New Supply Chain Model

With digitalization, the Internet of Things, Cognitive Analytics and Machine Learning, consideration of external and internal causal factors are a new opportunity to dramatically improve forecast accuracy based on new insights. Cloud deployed connectivity, computing power, data share, communication, and digital technology are breaking the barriers of siloed functions. You can’t break down the silos but, you can connect them in near real time from the points of final demand to the points of original supply. With “digital connected commerce”, omnichannel demand management enables daily forecasting and recasting. Time delay is replaced by near zero information latency and amplification synchronized and nullified taming the Bullwhip Effect.

To get started though requires a new business model, based on new mental models, advanced analytics maturity, and a new culture that demands planning excellence that avoids or overcomes the first pitfall of Supply Chain planning. New organizational models and structures to leverage new and more abundant sources of demand and supply data are emerging. In the coming weeks, we’ll explore more of the common pitfalls to Supply Chain planning.

[Ed: For further debate on this topic, IBF thought leader Eric Wilson argues that S&OP isn’t working, exploring the efficacy of the traditional approach to managing Supply Chain.]

When it comes to Machine Learning in Supply Chain Planning, how many of us can say we are ready to reap the benefits? If we are not quite there yet, what do we have to do get started? The use of Artificial Intelligence (AI) and Machine Learning in Supply Chain Planning has started and early adopters are already optimizing supply chain processes and gaining greater insight for smarter decision making. Here’s how you can get started with Machine Learning in your company.

Should You Believe The Machine Learning Hype?

Results from early Machine Learning successes are driving the hype to a fever pitch. Typing “Machine Learning AND Supply Chain” into Google delivers more than 3.8 million results in less than half a second.  There’s no question that Machine Learning is a topic that supply chain planning people are thinking, talking, and writing about. The real question is are we, as a profession, ready to embrace Machine Learning?” If so, what does that mean and how do we get there?

What Exactly Is Machine Learning?

While most organizations are still in the early stages of exploring machine learning, a thorough understanding of what it is and how it can be applied is vital. Terms like data science, advanced analytics, Artificial Intelligence and cognitive computing have been used interchangeably with Machine Learning in current periodicals. So what exactly is ‘Machine Learning’?

Machine learning is a type of artificial intelligence where computers have the ability to learn without being explicitly programmed. Machine Learning programs teach themselves to grow and change when exposed to new data.

Machines that could learn were first introduced in the 1960s with much of the early focus on probabilistic algorithms and forecasting. However, it wasn’t until the late 90s with the emergence of more powerful PCs that computer programs that analyzed large amounts of data and drew conclusions became more widespread. Some of the early successes came in the form of Neural Network programs. In more recent years, the explosion of ‘Big Data’ and the exponential increase in computer power led to significant growth of available solutions offering Machine Learning capabilities. Companies have discovered, however, that it takes more than abundant data and advanced computer capabilities to be successful with Machine Learning initiatives. Success also depends on having the appropriate talent, infrastructure and business focus.

A basic, foundational component of any effort to mature supply chain capabilities is to identify, secure and organize the necessary executive support. Real change happens from the top down, and it is imperative to secure an executive sponsor that has both the organizational power and vision for Machine Learning capabilities. It is important to ensure that any envisioned machine learning capabilities align with higher-level business goals and objectives.

[bar id=”6270″]

You Need A Machine Learning Champion To Define Requirements And Drive Things Forward

Who will lead the initiative and what skills do they need? Finding or assigning the right person to champion the initiative is critical for success. This leader needs a diverse set of skills including communication and influence to build support for adoption; team-building skills to convince existing employees of the benefits; and the ability to recruit new talent as needed. To establish organizational credibility, this Machine Learning champion needs to understand the various available technologies and their applications to the supply chain.

Working together, the executive sponsor and initiative leader can now envision how supply chain roles need to change to embrace an analytics driven supply chain. What new roles are needed (Database Engineers, Data Scientists, etc.) and are the required skills available within the company? How should the team be organized to efficiently run the business while also driving innovation?

What infrastructure or foundational components are required to enable the envisioned machine learning capabilities? Often the first place to start is to determine whether you have the data. What will it take to build accurate and consistent data to support the envisioned machine learning applications? How will the data be maintained and who is responsible for doing so? Do you need a Master Data Management (MDM) solution for supply chain data?

The next step is to determine whether your supply chain platform supports the envisioned Machine Learning capabilities. If so, how are these capabilities enabled? If not, what solutions are available on the market and is there budget available to acquire and implement a new solution? A critical step at this point might be to build an ROI based business case to obtain the funds needed to invest in a solution that enables machine learning capabilities.

Don’t Be Cutting Edge, But Copy What Cutting-Edge Leaders Are Doing

If your company is ready to explore machine learning capabilities, where should you start? Although it can be an advantage to be on the leading edge, I believe it is more productive to first review what has worked for other companies. Building Artificial Intelligence capabilities like machine learning is an evolutionary process. Getting started is probably more important than where you start. As you build experience, continue to explore new areas where additional applications of Machine Learning can add value. A few examples of how your company could get started with Machine Learning in the supply chain are provided below.

Application 1: Best-Fit Algorithms

Considering the fact that forecast accuracy continues to be a problem for many companies, a high value application of Machine Learning could be in the area of a “Best-Fit” algorithm for forecasting. Best-Fit algorithms automatically switch to the most appropriate forecasting method and parameters based on the latest demand information, to ensure you create the best forecast for every product at every stage of its life cycle. The algorithm evaluates forecast error for each cycle and recommends or automatically changes to the forecasting method that will produce the best result.

Application 2: Supply Chain Optimization

Another high value application area of Machine Learning is found in applying algorithms that continually analyze the state of your supply chain and recommend or automatically execute changes to meet customer requirements while maximizing company objectives. Optimization driven by algorithmic planning has also been in practical use for many years. Supply chain optimization relies on a set of provided information (supply chain facilities and capacities, transportation lanes and capacities, customer service requirements, profit requirements, etc.) and real-time operational updates (orders, shipments, unplanned events, etc.) to suggest optimal responses to planned and unplanned events.

Application 3: Multi-Echelon Inventory Optimization

One powerful application of Machine Learning is Multi-Echelon Inventory Optimization (MEIO), which automatically adjusts inventory parameters to meet stated customer service requirements while minimizing inventory investment. Using the latest demand and inventory projections and variabilities, MEIO seeks the optimal balance of component, Work In Process and finished goods inventory at the right locations. Embracing MEIO can reduce total inventories by upwards of 30% while maintaining or improving customer fill rates.

The Key To Success? Learn To Crawl Before You Walk

Attaining the full benefits of Machine Learning capabilities will be an evolutionary process. We must learn to crawl, then walk, then run. The introduction of Machine Learning into most supply chain organizations will take time, but that shouldn’t stop supply chain professionals from planning for the future or taking advantage of Machine Learning capabilities available today. Implementing foundational data and system platforms and taking advantage of existing algorithmic planning optimization technologies builds the expertise and experience needed to pursue more advanced Machine Learning in the future.

Are you considering applying machine learning solutions to your supply chain?  If so, what steps are you taking to get there?

Learn to create your own Machine Learning Models at IBF’s Demand Planning & Forecasting Bootcamp with w/ Hands-On Data Science & Predictive Business Analytics Workshop. Join us for this world-class training event in Chicago, Illinois on 14-16 March 2018.