Manufacturers and distributors rarely lack data, but they lack time to interpret it before a late receipt becomes a stockout, the stockout becomes an expedited freight bill, and the bill lands on the margin. Two articles by Sage published a week apart in September set out how AI closes that gap. Read together, they form a single argument. Supply chain AI works when it sits inside connected ERP data, close to a real decision, with a person still accountable for the outcome.
Three Layers of AI, One Operating Problem
The first article, AI in supply chain: How it works and where it adds value, frames the technology as three complementary layers:
- Predictive AI forecasts demand, inventory requirements, supplier performance, and equipment failures.
- Generative AI summarizes, drafts, and answers questions.
- Agentic AI completes multi-step workflows such as recommending replenishment quantities, evaluating alternative suppliers, and preparing purchase requests within predefined rules.
The evidence Sage assembles is pointed. It cites PwC research showing 53% of organizations already use AI to anticipate and mitigate supply chain disruptions, with a further 31% piloting it. It draws on McKinsey findings that AI-enabled supply chain management cut logistics costs by 15%, reduced inventory by 35%, and improved service levels by 65% compared with slower competitors. It also references Gartner’s forecast that by 2031, 60% of supply chain disruptions will be resolved without human intervention.
The second article, Generative AI in supply chain: Where it creates value and where to start, narrows the lens to the generative layer, and its argument is deliberately modest. The article notes that forecasting and optimization already rely on machine learning and mathematical models. Generative AI adds an interaction layer around those outputs. A planner asks why a forecast changed. A buyer extracts contract terms or compares supplier responses. An operations lead asks which orders relate to a reported inventory issue before choosing a response.
Sage draws a useful line between the two newer layers. Generative AI summarizes this week’s late shipments and explains which orders appear affected. Agentic AI identifies those shipments, assesses alternatives, and proceeds with an approved next step. That distinction matters more than any vendor label when ERP leaders decide what to switch on first.
Where The Two Converge
Both articles land on the same three conditions for value. Start with a measurable workflow rather than a mandate to use AI, and set a baseline for review time, manual steps, or output quality before the tool arrives. Connect the data, because disconnected inventory, purchase order, production, supplier, and cost systems leave both people and AI working from incomplete context. Define guardrails, since a supplier update summary carries different consequences from a change to a customer commitment.
That third condition is where Sage’s product roadmap meets its research agenda. ERP Today reported in February that Sage had embedded Sage Copilot into Sage X3 and Sage Operations to surface supply chain bottlenecks, delivery delays, and fulfillment exceptions before they escalate into revenue-impacting disruptions. In June, ERP Today covered the next wave of Sage X3 enhancements, including Sage X3 SaaS, AI-powered e-invoicing, and deeper Lynq production visibility, noting that Sage positioned Copilot as built into the flow of work while keeping people in control of decisions, approvals, and actions.
The Trust Question
The oversight theme is not incidental. In July, ERP Today examined Sage-sponsored IDC research finding that 71% of finance leaders would reject an AI tool that was 99% accurate if it could not explain its answers, and that 54% would pay a premium for visibility into how outputs are generated. Supply chain decisions carry the same exposure. A replenishment recommendation that cannot show its reasoning creates review work for planners and audit questions for finance.
Sage also acknowledges a readiness caveat that ERP Today has seen in the field. In April, ERP Today reported on two Sage X3 customers in food and beverage manufacturing, one of which said its data architecture was “not quite prepared” for AI-driven procurement and demand forecasting until it rebuilt foundational ERP structures.
What This Means for ERP Insiders
Treat the three AI layers as a sequence, not a menu. Predictive models produce the forecast, generative tools explain it, and agentic workflows act on it. Skipping the middle layer leaves people unable to interrogate what the system recommends.
Measure the workflow, not the technology. Sage’s own guidance is to baseline review time, manual steps, and response speed before deployment and expand only after results hold.
Budget for explainability as a supply chain requirement. The trust threshold finance leaders have already set will apply to inventory, procurement, and production decisions as agentic capabilities mature. Vendors that build reasoning traces and approval controls into the product, rather than selling them as governance extras, will find the easier path into mid-market operations.
