For much of the past decade, manufacturers have focused on improving visibility across increasingly complex supply chains. But while businesses now have access to more data than ever before, this does not necessarily translate into faster or better decisions.
To achieve that, businesses must turn this information into action, says Alfonso Rodriguez, head of marketing for Microsoft for Manufacturing. Rather than operating as a decision-support tool that periodically provides recommendations, AI is becoming an active participant in supply chain operations – sensing what is happening, planning a response, executing it and learning from the outcome.
“Today, most supply chains still operate in silos – fragmented data across systems, manual decision-making, and planning cycles that run periodically rather than continuously,” says Rodriguez. “What’s changing is that AI is moving inside the operating rhythm itself: sense, plan, execute, monitor, learn.”
He describes this as a closed loop rather than a sequential process, where “every signal updates decisions, every decision triggers action, and every outcome improves the next cycle.”
Rodriguez believes that, within three to five years, humans will increasingly operate at the strategic decision level while AI handles autonomous execution at the operational level. People will continue to set strategy, define outcomes, approve exceptions and make trade-offs between service, cost, inventory and resilience, while agents orchestrate and continuously optimise day-to-day operations.
That transition, however, will not happen everywhere at once. Rodriguez says the fastest adoption is likely to come in high-frequency, lower-stakes decisions such as carrier selection and re-routing, warehouse labour allocation, invoice matching and high-volume, low-attention procurement categories. “These are decisions made hundreds of times a day where the cost of being slow exceeds the cost of being wrong,” he says.
Manufacturers are understandably more cautious when AI decisions affect uptime, safety or process control. For these areas, Rodriguez advocates a gradual approach, beginning with human-in-the-loop recommendations inside governed guardrails and only introducing greater autonomy once confidence, controls and operating boundaries have been proven.
“Autonomy is earned incrementally, not switched on,” he says.
Manufacturers are increasingly combining intelligent automation, connected data and human oversight to create supply chains that continuously sense, plan, execute and improve
This shift from visibility to action is particularly important because the investment in supply chain visibility has not always delivered the expected results. Rodriguez notes that Microsoft expects around 85 per cent of organisations to require real-time end-to-end visibility by 2030 but argues that visibility has not solved a fundamental problem: companies can see disruption without necessarily being able to quantify its consequences, simulate alternatives or respond quickly enough.
“The next wave is decision intelligence and autonomous execution sitting on top of that visibility layer,” says Rodriguez.
That means enabling AI to reason across competing priorities rather than optimise a single metric. Rodriguez points to policy-aware trade-off analysis that considers cost, service, cash and risk simultaneously; multi-scenario simulation that can show the cascading effects of a disruption and rank potential responses; and self-healing workflows that can detect exceptions and resolve them within defined guardrails.
For Rodriguez, one of the clearest signs an organisation remains stuck in the visibility era is the way it manages exceptions. “The tell that a company is still stuck in the visibility era is simple: exceptions are still handled by email and phone, with manual overrides nobody can explain afterwards,” he says. “That’s where the value is trapped.”
Unlocking that value requires manufacturers to connect functions that have traditionally operated separately. Supply chain decisions rarely affect only the supply chain: a change to a production plan can alter margins, an inventory policy affects working capital, while a sourcing decision can carry a carbon cost depending on the region and sourcing path.
“This is the problem we spend most of our time on, because supply chain decisions are financial decisions that happen to have an operational expression,” says Rodriguez.
He identifies two foundations for making cross-functional optimisation possible. The first is a unified data platform with a semantic layer and supplier systems. This allows AI agents to reason within the appropriate business context rather than simply analysing raw, uncontextualised data.
The second is multi-agent orchestration, allowing planning, sourcing, logistics, warehouse and risk agents to collaborate instead of each optimising its own function.
The impact becomes particularly visible at board level through integrated business planning (IBP). Rodriguez argues that many manufacturers still cannot run product, demand, supply, financial and executive reviews using one connected set of numbers. AI has the potential to compress that cycle, connecting operational plans with financial outcomes with greater accuracy and agility.
“None of this requires rip-and-replace,” says Rodriguez. “We orchestrate across the systems customers already own. The foundation supports incremental adoption without requiring wholesale replacement of existing systems.”
This is where Microsoft’s partner ecosystem becomes important. Rodriguez points to Blue Yonder and o9 Solutions as examples of how expertise can be combined with Microsoft’s technology foundation.
Blue Yonder has built its next-generation applications on Microsoft Azure, helping customers move from siloed configurations towards an intelligent operating network. One example comes from a global ingredients manufacturer, which reduced inventory levels by 16 per cent while increasing service levels by 15 per cent.
For a global food manufacturer, o9 Solutions used an enterprise knowledge graph representing products, plants, suppliers, carriers and their relationships, running on Azure to support forecasting, scenario planning and digital IBP. This resulted in a 10 per cent improvement in short-term forecast accuracy at item-location level and a 25 per cent reduction in excess inventory.
“The pattern across both is the same,” says Rodriguez. “We don’t win by owning the application layer. We win when the data foundation, the AI and the governance model are strong enough that partners with 20 years of domain depth can build decision intelligence on top of them – and the customer gets a connected supply chain without a rip-and-replace programme.”
Technology, however, is only part of the transformation. As AI moves into everyday operational decision-making, manufacturers also need to rethink how their organisations work.
“The technology question is largely settled; the operating model question isn’t,” says Rodriguez.
He identifies three changes that separate organisations scaling AI from those remaining in pilot projects. “First, manufacturers need explicit decision rights for agents, defining which decisions can be taken autonomously, which require approval and how the system should escalate when confidence falls below a defined threshold.
“Second, data trust needs to become a prerequisite rather than another workstream. If planners do not trust the numbers – particularly when the system of record and planning tool disagree – they will override the agent, and the value of automation quickly disappears.
“Finally, roles themselves will change. Planners will increasingly stop producing plans and instead manage exceptions and policy. That is a different job, and one that needs to be named, trained for and rewarded.”
Connected supply chains rely on real time data, AI-driven decision intelligence and coordinated execution across production, warehousing and logistics
The manufacturers succeeding with AI tend to scale one decision loop at a time rather than attempt to transform everything simultaneously. They measure outcomes in financial terms from the outset and ensure supply chain and finance leaders are working from the same numbers.
Looking further ahead, Rodriguez sees several emerging capabilities becoming standard practice. “Physical AI is one of the most significant, with robots and humanoids increasingly working alongside warehouse operators,” he says. “Beyond that lies the prospect of autonomous facilities where humans determine strategy while AI manages execution and fleet maintenance.”
Manufacturers, he says, will also increasingly need visibility into multi-tier and material-level risk: understanding dependencies between Tier 1 and Tier 3 suppliers, identifying products containing materials that could face future regulation and spotting leading indicators of supplier distress.
Rodriguez also expects always-on network simulation to become commonplace, with digital twins continuously testing the impact of tariffs, demand fluctuations and port closures. And, there is the opportunity to build foundations that allow intelligence to move from observation to decision to action.
“Build the unified data foundation and semantic layer, because every capability above depends on it,” says Rodriguez. “Manufacturers should also establish autonomy and governance policies before deploying agents and select one decision loop – such as demand and sales and operations planning or logistics execution – to take all the way through to autonomous action.
“The companies that wait for the full picture will be buying capability from competitors who didn’t.”
Read more like this, including news, interviews, case studies and insight from industry experts, in the Autumn 2026 issue of Technology Record. To get content like this on a regular basis, subscribe to the print edition or free digital edition of the quarterly Technology Record magazine.