Supply chain analytics has matured to the point where advanced methods – machine learning, data science, and the specific implementations of modern analytical science and technology – are both relevant and necessary: relevant because the value gained from improved analytics offers an impressive ROI relative to the cost of analytics development; necessary because more and more enterprises are creating assets from data that improve business operations and profit margins. Companies that are not investing in advanced analytics for their supply chain are condemned to lower profit margins, because analytics combined with the right decision framework always creates more value, so long as the application is curated well and applied to the right point of leverage.
Supply chain analytics has many such levers – demand planning, sales and operational planning, and inventory and supply planning, for example, are three common use cases for advanced analytics. The complexity of the supply chain management use case for advanced analytics requires a trustworthy solution that capitalizes on the same (or similar) advances in computer science that have driven the value creation of firms such as Google (search, navigation, digital ad optimization), Apple (user experience), and Amazon, NetFlix, and Uber/AirBnB (recommender systems, digital transactions, customer/performer reciprocal rating methods). At Halo, we are investing in developing similar methods for specific use cases in Supply Chain Management. Along the way, we are educating our customers on how to make the best use of these methods.
This is the first report in a series written to inform business leaders on how to identify the right use cases for supply chain advanced analytics, and to start that journey with a cohesive plan, a rational budget, and improved likelihood of success in their data and analytical agenda. But what starts the discussion is a definition of machine learning that is relevant to supply chain management:
Machine Learning in Supply Chain Management capitalizes on operational big data to create dynamic forecasts that improve the allocation of resources and capital that the organization manages today and in the future. The necessary components are curated data and appropriate forecasting models, reporting that informs decision makers, a method of updating assumptions and decisions, and a human-computer interface that facilitates the analytical, decision management, and learning workflows.
Foundations
What business leader in food and beverage, manufacturing, or retail has looked at the innovative data and analysis systems described above and not wanted a solution for their own organization? And at the same time, how many business leaders find this to be a leap too far, due either to past failure to capitalize on the value of data, the inability to hire and retain the right internal team, internal turf issues that inhibit innovation, or failed attempts to leverage third-party providers to create change? Simply stated, if machine learning were easy, it would already be happening in your organization.
One goal of this series is to highlight the foundational issues that, once solved for, help to divide the larger problem into parts that can be solved for incrementally, and in an order where quick wins provide insights and the confidence to continue the work and the necessary investment. These parts include defining the problem, building the analytical process, studying the process to identify and solve for likely failure points while automating processes, and creating ease of use through a user interface/user experience that leads to optimal human-computer interaction.
Many overviews would start with discussions about advanced algorithms, programming languages, and data processing, storage, and retrieval. We at Halo prefer to start with the user experience – we identify the optimal way for an expert to interact with data to improve decisions. We then work backwards to find the appropriate solutions that enhance the user experience so that the expert has more data at hand for making better decisions. The expert user interacts with data, reports, and forecasts, and the variance from forecast identifies business processes that decision makers will want to understand and take management action on. This is the essence of supply chain analytics; it is two steps removed from programming languages and the particular algorithm used to build the forecast. But it cannot work if the decision maker’s user experience is sub-optimal; this is why Halo has invested in an exceptional user interface.
Halo clearly emphasizes the human-computer interaction model. We avoid using the phrase artificial intelligence because supply chain analytics requires human oversight due to the high cost of bad decisions that could arise from an unmonitored purchase order or inventory forecast. We appreciate the variety of machine learning techniques, programming language, and IT solutions, but we see those as aids to the forecasting solution that in many cases are interchangeable. We focus on well-established technology, proven algorithms, and experimental methods for improving prediction over time as more and better data are found.
But what can’t be swapped is the role of the expert who understands their business model, their products, and their distribution channels, and the importance of meeting that person’s needs when it comes to data access, data visualization, report creation, report management, and decision monitoring. This is why Halo has invested in foundational tools that organize data, visualize and manipulate data for knowledge discovery and reporting, and why we have focused on creating an intuitive user interface that facilitates the translation of insights and forecasts into new decisions and distributed reporting.
We encourage business leaders to regard the user experience as a top differentiator when selecting an advanced analytics solution. If the user experience is solved for, improving the analytics and managing the data can follow. If the user experience does not facilitate knowledge discovery, decision, and report management, and continuous learning, then the value of the data and analytics will not be realized.
We hope that this focus is appreciated by our readers. But even more importantly, we want the reader to appreciate the role of expert judgement and corporate governance in the application of advanced analytics, both broadly applied and in the supply chain management use case. We emphasize the user experience because an exceptional user experience will gain greater buy-in across the organization, will reduce dissatisfiers of adopting a new software solution and new reporting processes, and thus will have a higher likelihood of success.
Foundations applied to Supply Chain Analytics
As of now, there has been little written on what will make for an exceptional analytical revolution in supply chain management vis-à-vis machine learning and data science. Will proper data science in supply chain management simply borrow innovation from other paradigms, or are new tools and methods required? A thorough discussion of this topic will be covered later in this series, but for now we will start with general definitions of machine learning, data science, and business intelligence that are relevant to the supply chain analytics leader. With these definitions, the analytics leader can then tackle the use case and business requirements for his or her organization. Once the business case is tailored and vetted for the specific needs of that organization, the leader can then engage in a Design and Build the Future exercise by declaring the need for innovation, assessing the specific strengths and opportunities of the organization, and enlisting internal and external resources to create change, measure performance, and manage the analytics re-invention over time and across divisions of the organization.
Let’s return to defining the central terms of data science, machine learning and business intelligence. Unfortunately, there are already so many definitions of these concepts that the concepts themselves have become muddy, as every new entrant to the advanced analytics market seems bent on creating an arbitrarily unique definition that in some cases is overly specific, and in other cases simply wrong. Rather than re-hashing this loose use of jargon, here is a useful starting point for the analytical leader to begin the discussion:
Data Science is the application of the scientific method to understanding the utility of data, how to create useful information from disparate (and often messy) data, and understanding where advanced analytical methods create value above the value realized from classical statistical and business analysis.
Machine Learning is a broad domain that includes many advanced analytical methods that improve prediction over time as more data become available. This can include advanced “post-classical” predictive modeling algorithms, and the structuring of “data input → forecast output” methods that facilitate rapid learning and fine-tuning of supply chain management decisions. In some cases, the results are reports and forecasts for management review, and in other cases the output is an immediate input to a purchase or inventory management data system. Data quality assurance is always a key component, and automating data QA reporting is usually a Phase 1 deliverable. Machine learning becomes a closed system when the data flows are highly automated, modeling is continuously updated based on the latest information, and the output of the data processing is an automated input to a business decision. Most supply chain management use cases for machine learning will start with forecasts and reporting, and then progress towards automation.
Business Intelligence (BI) is the process of querying data, reviewing reports, and drilling into the data to uncover insights that require management review, or that could identify an issue with the analysis or an opportunity to improve prediction. We emphasize the importance of BI because there is often a knowledge gap between the core competencies of a data scientist or machine learning specialist and the knowledge of a business leader who understands the company’s P&L and where to make data-driven decisions. The business leader needs a BI tool, on his or her desktop, in order to drill down and understand the variances from forecast both visually and financially.
Machine learning use cases
One reason why supply chain management is ready for machine learning is because the typical use cases for advanced analytics are well understood, and big data analytical methods (post-classical advanced predictive modeling) work very well in these situations. Typical supply chain analytics frameworks include demand planning that is grounded in forecasting methods, the segmentation of products and customers, and the management of additional data sources that improve prediction and provide more actionable insights to business decisions. These additional data sources can be both internal data, like HR reports, and exogenous factors such as the economic cycle or a major weather event that are outside of the business’s control. More advanced applications could involve the correlation of marketing and sales activities with forecasting errors as a way of identifying a signal of the effectiveness of marketing spend, price discounts, and competitor activities. These use cases will be expanded on as we explore Halo solutions to these use cases later in this series.
If a company builds a foundation of data, analysis, reporting, and knowledge sharing, even more advanced goals can be pursued. At that point, the work is truly scientific – deeper data analysis, A-B testing of methods and data sources and segmentation, and even advanced experimental design methods that facilitate the most rapid learning. Most organizations are not staffed for this work, and most entry-level data scientists do not have the experience necessary to take on this work. This is where an organization benefits from advisory services that can identify gaps and offer solutions – especially solutions that are well informed by prior successful deployment, and where the solution is an elegant analytical framework that can be handed off to the organization’s junior data scientists for maintenance.
Supply Chain Data Science
We believe not only that half of our customers have enough value in their supply chain data to benefit from the deeper scientific analysis of their data, but also that the available talent pool of data scientists with supply chain management experience and acumen is very limited. Our offer to our customers is to design simple, elegant analytical solutions, document and then hand over the process, while maintaining a relationship through software support and training. Because we understand our customers’ data, and have collaborated in building their advanced analytical methods, we can help them select the right external hires or internal promotions to make the system run efficiently.
Next time, we will go deeper into how to select software for the supply chain analytics use case, how to ensure that your selection enables advanced analytics, and the value-add of advisory services as you build out your supply chain analytics agenda.