The Operating Model of the AI Native Global Supply Chain : Part 1

An operating model is the system of arrangements by which an enterprise converts its assets into an outcome most likely for its customers. The useful definition of the operating model is structural. An operating model is a stack of five layers bound by downward causality.

The five layers of the operating model

1. Asset base: What the enterprise owns and operates the physical plant, the workforce, and now the digital estate. Assets define the outer boundary of what is possible.

2. Business model: How those assets are converted into value and margin make-to-stock or configure-to-order, owned manufacturing or contract manufacturing, regional or global, direct or channel.

3. Business process: The repeatable sequences that execute the business model: plan, source, make, deliver, return and enable.

4. Organization structure and RNR: The grouping of people into functions and levels, and the allocation of decision rights, accountabilities and responsibilities among them.

5. Performance system: The KPIs, targets, service-level agreements (SLAs) and rewards through which the organization steers itself and demonstrates value to the customer.

Why the direction of causality matters The direction of causality matters more than the list. Assets constrain the business model. The business model dictates which processes must exist. The processes dictate which roles must exist and where decisions must sit. And the role structure determines what can be measured and attributed at all—you cannot hold a layer accountable for a number that no role in it controls. A change at any layer forces the re-derivation of every layer below it. This is why the operating model is the right unit of analysis for the shift that AI is forcing on operations. A technology that only changed tooling would sit inside Layer 3 and could be absorbed. A technology that changes the asset base enters at Layer 1 and obliges the enterprise to re-derive everything below it. Agentic AI is the second time in fifty years this has happened. The first time, we handled it badly.

When the asset base was purely physical, management was direct. A plant manager walked the floor. A warehouse supervisor counted pallets. The gap between the state of the world and the manager's knowledge of that state was a few meters and a few minutes. The addition of digital assets did something more consequential than automating clerical work. It inserted a representation between the physical world and the decision-maker. The enterprise stopped managing the factory and started managing the record of the factory. Physical flow and information flow decoupled—and every material decision from that point forward was made against the record, not against the world. The Information Loop Two governing mechanisms operate inside this layer. The Information Loop describes the circuit every operational decision must complete before the physical world responds: Data → Analysis → Decision → Communication The 3C Framework—Content, Context, Choices describes what the loop must deliver for a decision to be made well: • Content: The information must be accurate. • Context: The information must be sufficient to interpret it. • Choices: The available choices must be genuinely actionable. These mechanisms explain why the decoupling layer matters to operational responsiveness. The supply chain bottleneck has traditionally been considered to be at the asset layer i.e. a truck, a plant or other physical capacity. The paper argues that the gap in the decoupling layer, and the resulting decision latency between digital and physical assets, is also critical. “So far in operations, supply chain bottleneck has always been considered as the asset layer i.e truck, plant or the truck however in reality it is the decoupling layer gap caused the decision latency between digital and physical assets. “ Once the decoupling layer exists, the binding constraint on responsiveness becomes the cycle time of the loop through it not the throughput of the plant or the transit time of the truck. Where AI enters the operating model Because the decoupling layer is where decisions are actually made, it is where AI enters. Agentic AI does not primarily automate physical work in the supply chain, it occupies the decoupling layer and begins executing the Information Loop autonomously. This has a direct implication for the operating model. The layers most disrupted are precisely the two that were built to staff and measure that loop: organization structure and roles and responsibilities and the performance system. To understand why, it is necessary to look at how the supply chain operating model evolved.

The operating-model stack did not change all at once. Each era added capability to the asset base and forced an adaptation down the stack. What is instructive is not the additions themselves, but which layers were redesigned and which were merely extended. Era I: Physical (1975–1990) The asset base consisted of factories, warehouses, fleets and labour. The business model emphasized scale, make-to-stock and vertical integration. Business processes were manual, paper-based, batch-oriented and local. The organization structure was a tall functional hierarchy, with span limited by human information processing. The performance system focused on cost, output and on-time delivery. Era II: Systems of Record (1990–2010) ERP, MRP, EDI, barcodes and servers expanded the asset base. The business model shifted towards global outsourcing, multi-tier operations and low-cost country sourcing. Processes became transaction-standardized and enforced through systems of record. Planner, master-data, centre-of-excellence and regional or global roles were added to the organization. Performance systems gained measures such as on-time-in-full (OTIF), inventory turns and cost-to-serve. Era III: Systems of Insight (2010–2020) Cloud, IoT sensors, data lakes, dashboards and robotic process automation expanded the digital estate. Business models became segmented and omnichannel, with differentiated cost-to-serve. Processes became exception-based, analytics-informed and partially automated. Analytics and data science teams were added alongside existing structures. Performance systems incorporated forecast accuracy, plan attainment and data quality. Era IV: Disruption and Machine Learning (2020–2024) Machine-learning models, control towers and digital twins expanded the asset base further. Business models became more resilience-weighted, with dual sourcing and near-shoring. Processes moved towards scenario-driven, probabilistic planning. Risk and resilience roles were added, alongside measures such as agility, time-to-recover and variability. Era V: Agentic (2024 onward) Agents enter the asset base as actors, not tools. The business model becomes continuously re-optimized, with service promises set dynamically. Processes become agent-executed, with humans intervening by exception. However, the paper identifies two unresolved questions. Organization structure and RNR: Agents have no established place in the hierarchy. Performance system: Decision-level and agent-level measures remain unresolved.

Read the five-era comparison vertically and the evolution looks healthy. Read it horizontally, row by row, and the asymmetry is unmistakable. The top three layers were genuinely redesigned in each era. Era II did not append global sourcing to a domestic business model—it replaced one with the other. Exception-based planning in Era III did not sit alongside manual planning; it displaced it. The bottom two layers were only ever extended. Era II did not redesign the functional hierarchy—it added planner and master-data roles inside it. Era III did not redesign the scorecard—it added forecast accuracy to it. Forty years of asset-base change produced a taller hierarchy and a longer scorecard, but the same shape of hierarchy and the same logic of scorecard. Why this pattern was survivable This asymmetry was survivable for four decades for a specific reason: in every prior era, the new digital asset was a tool operated by a human role. ERP was operated by a planner. A dashboard was read by a manager. An RPA bot executed a task defined by, and owned by, an identifiable person. The role remained the unit of accountability, so the accountability architecture never had to change. Adding tools to roles worked, so extending the hierarchy and extending the scorecard was a rational response. Where the pattern breaks An agent is not a tool operated by a role. An agent performs the role: it senses, analyses, decides within a scope, and communicates. It completes the Information Loop without a human in it. The moment that is true, extending the hierarchy stops working—because the thing being added is not a capability that a role uses, it is a candidate occupant of the role itself. And a scorecard that measures only outcomes cannot tell you whether the agent decided well, or merely decided quickly.

Assessing each layer against what agentic systems actually change gives a picture of where enterprise effort has and has not gone. Asset base Agents join the estate as operating assets with running costs, capacity, failure modes and a lifecycle. The typical enterprise response described in the paper is to treat them as software licences in the IT budget, rather than as operating capacity. Business model Service promises can be set per customer and re-optimized continuously; cost-to-serve becomes dynamic. The paper describes active exploration and pilots in dynamic allocation, pricing and segmentation. Business process Deterministic judgement steps collapse; the process becomes exception-triggered rather than cycle-triggered. The paper describes substantial investment in copilots, exception engines, automated replenishment and expediting. Organization structure and RNR Agents occupy execution-layer decision scopes. The managerial layer shifts from supervising work to designing and assuring it; span of control decouples from headcount. The paper describes agents being deployed into existing roles as productivity aids, while job architecture, decision rights and grade structures remain unchanged. Performance system Decision quality, latency, autonomy and attribution become measurable for the first time. Outcome KPIs alone become insufficient. The paper describes the continuation of the same scorecards and targets, with agent value reported separately as project savings. The conversion problem The consequence of this split is a predictable failure described in the paper: pilots demonstrate genuine capability at Layer 3, but stall on the path to enterprise value because Layers 4 and 5 were never re-derived. The agent works. Nobody owns it, nobody's decision rights changed, and no metric on anybody's scorecard moves when it succeeds. This connects directly to a broader thesis: “Productivity is converted, not adopted.” Capacity released by an agent becomes enterprise value only if some structure converts it into throughput. That structure lives in Layers 4 and 5—who is accountable for the freed capacity, and which metric registers its redeployment. Leaving those layers untouched returns released capacity as slack rather than converting it into output, no matter how high adoption climbs.

AI for Work is a practical book for COOs, operations leaders, and supply chain professionals who want to understand AI, its capabilities, and its limitations as they navigate the transformation of global operations in a landscape of constant disruption. Reviewed by a panel of senior experts from top global companies like Snowflake, Accenture, IBM, PwC, FourKites, Inc., and Lenovo, this book is written from the perspective of a transformation leader who has spent many years driving digital transformation and has seen the same patterns repeat: how decisions are made, where systems fail, and why so many operations leaders still struggle to build AI that can scale beyond a long list of pilots and proofs of concept. This book explains AI and work through the structure of operations itself by shifting the mindset from seeing operations as a collection of processes to seeing operations as an information loop — sense, decide, act, learn. Every decision in that loop is governed by three forces: Content, Context, and Choices. That single shift is what separates leaders who see AI as just another tool to optimize operations from those who see AI as a core component of structural design. Across fifteen chapters and five parts, the book moves from the foundational physics of operations through AI technology, into the human–AI division of labor, organizational redesign, and the future of the workspace. It covers ANI, AGI, and ASI in clear operational language; the three roleplay levels of AI in real workflows; the shift from management to design; the four-dimension ROI model that can survive a CFO review; and the trust path that helps AI agents move beyond the pilot stage into production. Some of the original frameworks introduced in this book to help COOs, operations leaders, and supply chain professionals build a practical roadmap for their AI journey include: Information Loop 3C Framework of Content, Context, and Choices From Mass to Bit to Algorithm Management to Design Three AI roleplay levels of Assistant, Coworker, and Leader Domain AGI The book offers three reading paths: The COO Fast Track — ninety minutes to the working model The Transformation Leader — a one-to-two-week deep read with the toolkit The Operations Practitioner — the full manuscript, chapter by chapter The examples are drawn from two decades of work in global operations, and the frameworks are intentionally original, practical, and built for leaders trying to turn AI into something operationally useful. Link to purchase : https://www.amazon.in/dp/B0H472BZWH/