Journal of Business Forecasting Articles Issues

There are several realities demand planning teams must face when trying to implement AI for the first time, including how AI is more about the ‘plumbing’ than algorithms, how it may highlight structural weaknesses in your existing processes, and how successful planning is more about the right response than it is about the right number. This article reveals how to take your first steps in AI, including how to build a pilot, and how the real value AI in demand planning may not be the technology itself but the way it quietly reshapes how organizations think, talk, and decide about the future.

In 2017 I presented a vision for AI adoption in demand planning in my book The Fundamentals of Demand Forecasting, whereby the Demand Planner of the future would be responsible for managing multiple AI systems. In this article, I revisit the applications of AI that I predicted then, and assess their viability of them now in light of recent developments and use cases. Of the five applications I outlined, I highlight the effectiveness of three agentic approaches that seem likely to add value to an organization, namely validating manual forecast overrides, explaining demand variability, and automatically gathering information affecting the demand plan. I also reveal which potential applications of AI are more difficult to implement and less likely to yield tangible benefits.

AI is a stack of predictive, prescriptive, and generative capabilities embedded in workflows to drive measurable outcomes like higher uptime, lower OPEX, faster cycle time, and better customer experience. It learns from telemetry, transactions, and text to forecast, diagnose, decide, and act. The highest value comes from closed-loop application of insights, where models trigger automated control actions and continuously learn from outcomes to self-improve. Success is judged by business KPIs, not model metrics.

I recently read a Wall Street Journal article that detailed how food manufacturers are using Artificial Intelligence to optimize recipes according to desired characteristics such as flavor, aroma and appearance, as well as ingredient costs, environmental impact, and nutritional profile. Such optimization techniques are not new, in fact they employ Linear Programming methods developed as early as 1947. In this article, I discuss how the Simplex model can be used for product optimization, using case studies from my UMass Business Analytics undergraduate course. I also discuss whether LLM’s like ChatGPT are capable of developing new and better models than those developed by humans or whether they will always rely on existing approaches.

AI is everywhere from conference stages to newsfeeds and it’s often described as a mythical, world-changing force. This article brings AI back down to earth, starting where it matters most: demand planning. Among all supply chain disciplines, demand planning offers the perfect entry point for AI because it blends data, human judgment, and measurable outcomes. This article reveals three actionable steps: start small, find an approachable pain point, and define ROI early. We discuss how AI could have solved a real life problem planning French fry demand, and how even modest AI experiments in forecasting and planning can unlock speed, accuracy, and confidence that once seemed out of reach.

Demand Planners face mounting pressure: deliver greater value with fewer resources in an environment that changes daily. This article explores how simple, free AI Agents can reduce manual work, improve agility, and help Planners save time and focus on more value-added tasks. You’ll also see a practical example of a Copilot agent that saves hours of analysis, proof that quick wins are possible without technical expertise. I reveal the prompts I use for Microsoft Copilot to create an AI agent that identifies trends and anomalies in a product portfolio and outputs qualitative and quantitative insights.

Artificial Intelligence is reshaping expectations in planning, yet many growing organizations still find its benefits difficult to access. Planners often operate in environments defined by fragmented data, spreadsheet-based processes, and inconsistent decision rhythms. In this article I reveal how these realities limit the impact of AI and increase the burden on planning professionals. I discuss how meaningful progress begins not with new systems but with readiness: stronger data structures, clearer interpretation of signals, and a consistent cadence of decision-making.

As large language models (LLMs) become common in business “vibe coding” is helping SMEs automate demand forecasting, inventory planning, and S&OP without heavy IT support. Users describe tasks in plain language and LLMs generate the code, enabling faster iteration and cost savings, especially in Excel- based settings. However, this approach might also introduce risks like shadow IT, compliance issues, and fragmented planning. This article shares practical insights from using LLM-driven vibe coding in supply chain planning to maximize benefits and minimize pitfalls.

Consensus forecasts indicate that GDP growth will increase by 1.48% from Q1 2026 to Q4 2026, signaling a modest economic expansion. Over the same period, the M2 money supply is projected to grow by 4.00%, rising from $22,768.82 billion in Q1 2026 to $23,679.87 billion in Q4 2026, compared with $22,298.10 billion in October 2025. This continued expansion of the money supply is expected to support economic growth while simultaneously exerting upward pressure on inflation.

Large Language Models (LLMs) offer a conversational approach to planning. Such LLMs, trained on internal and external data, don’t just provide forecasts but can answer questions such as why demand dropped in a particular region, and provide scenario plans if a given event were to occur. Further, they can provide a rationale for their forecasts, making this a highly interactive tool that facilitates deep understanding of demand drivers. In this article, I discuss how LLMs learn from human overrides and analyze data including emails, campaign notes, and disruption alerts that traditional models overlook. I also discuss the practical applications of LLMs in retail, apparel, consumer goods, and logistics that are driving real-world operational efficiency.

The proliferation of systemic volatility has stretched the capabilities of traditional Advanced Planning Systems which, despite their optimization power, were not designed for today’s highly dynamic and data-rich environment. This article posits that agentic artificial intelligence represents the next evolutionary step, shifting the paradigm from human-led optimization to AI-driven autonomous action. I present a practitioner’s guide for successful implementation, including a strategic matrix to help leaders select high-value use cases and validate genuine agentic capabilities. I also discuss how the role of the Supply Chain Planner must evolve from system operator to a strategic conductor of an AI-powered enterprise.

I recently attended an IBF New England Chapter meeting where a manager discussed the S&OP process at her company which was having difficulty convincing colleagues in finance to assess whether demand-supply plans were aligned with annual financial performance targets. In this article, I discuss my response to her, namely that if supply chain managers want to get buy in from finance and executive leadership, they must speak their language—that is, in dollars and cents. I introduce the Dupont Model of Return-on-Assets (ROA) as a blueprint to use when developing a financially-oriented business case, which can effectively present demand and supply plans in financial terms, and facilitate involvement of finance colleagues.

In this article, I provide a practical guide to building up and managing safety stocks, drawing from my hands-on experience with a high-tech client. I emphasize the importance of aligning statistical models with operational realities, and share insights into navigating the complexities of different safety stock types— statistical, factory-requested, and strategic—through a clear, well-maintained policy. I address practical elements that made the implementation both sustainable and successful, such as cross-functional coordination, effective communication, and managing workload.

This article details how a global medical devices manufacturer faced significant challenges integrating its expanded supply chain network following a major acquisition. I reveal the challenges involved in such a merger, including tripling the product portfolio and expanding the distribution footprint from a few regional centers to over 30 global warehouses. This case study explores how the organization redefined its demand management and forecasting processes to enable scalable, efficient planning across the new enterprise structure, thereby unlocking post-acquisition value.

In this article I present a case for using real-time search spikes and other unstructured data as early indicators of operational risk. It highlights how terms like “fuel long queue” or “black market dollar rate” can precede formal disruptions in supply and demand. I introduce a tool under development designed to translate publicly available data into proactive disruption alerts, especially for users in low-data or informal environments. By mapping search trends to operational risks like fuel shortages, border closures, or protests, the tool offers a practical, accessible alternative to traditional reactive models.