Sight Machine and Microsoft use AI-driven optimization to increase manufacturing productivity by 10% with Microsoft Foundry

Like many high-mix manufacturers managing shifting demand, supply constraints, and plant-floor disruptions, a major beverage manufacturer was replanning production schedules 10–15 times per week, losing hours to manual meetings while conditions changed faster than plans could keep up. Sight Machine and Microsoft changed that with an AI-driven scheduling agent built on OptiMind and Microsoft Foundry that converts real-time plant data into optimized schedules automatically. The generalized approach enabled dynamic manufacturing optimization, increasing overall plant productivity by 10% or more while reducing scheduling time by 75%.

Manufacturing plants generate enormous amounts of operational data every day, but most of it remains difficult to act on in practice. Equipment across the plant floor was rarely designed with analytics in mind, leaving manufacturers to make production decisions based on averages, spreadsheets, and operator experience.

Sight Machine was built to solve that problem. The company's industrial AI platform connects every plant data source—IT, OT, cloud, and edge—and continuously structures and analyzes that information so manufacturers can improve uptime, throughput, and quality in real time. To extend that foundation into broader supply chain challenges, where decisions about what to produce, when, and where carry enormous operational consequences, Sight Machine partnered with Microsoft and integrated OptiMind through Microsoft Foundry.

Together, the companies developed a new approach to dynamic manufacturing optimization that continuously adapts production schedules using real-time operational data, helping a major beverage manufacturer reduce non-value-added production time by 75% while improving overall production efficiency and responsiveness.

Bringing visibility to manufacturing blind spots

Production scheduling—assigning specific jobs and quantities to specific machines based on customer demand and raw material supply—might sound like a planning exercise. In practice, it's a daily act of damage control.

When machines slow down, orders change, or materials arrive late, the data that could inform a response is scattered across disconnected systems, and by the time it reaches the people who need it, the schedule is already wrong. The role is so reactive it's sometimes called "rescheduling," because that's what the job actually requires.

"One of the biggest challenges is that reality never works out the way you originally planned and scheduled," says Kurt DeMaagd, Co-founder and Chief AI Officer at Sight Machine. "Machines slow down, maintenance issues happen, customer orders change, input materials are delayed, and manufacturers have to constantly adapt."

For one major beverage bottler, those replanning sessions occurred 10 to 15 times per week. When a line slowed or a new order arrived, schedulers had to quickly pull together plant managers, engineering leaders, and operations teams to revise production plans without sacrificing output targets or adding downtime. Sequencing added further complexity—get the product order wrong and cleaning time increases, production slows, and quality risks follow.

"Manufacturing works best when there is consistency," says DeMaagd. "But the reality is that something is always changing, and every disruption creates a ripple effect across the operation."

The challenge was compounded by workforce change. As experienced process engineers approached retirement, the bottler faced a growing need to preserve decades of scheduling expertise that lived in people's heads, not in any system others could access or build on.

Working with Microsoft, Sight Machine set out to change that.

Building the solution: OptiMind meets the factory floor

Mathematical optimization models have long offered a path to smarter production scheduling, but translating real-world manufacturing constraints into optimization models traditionally required specialized expertise. That changed when Sight Machine was introduced to OptiMind through the Microsoft Foundry team. Developed by Microsoft Research, OptiMind is a small language model that converts business problems described in plain language into the mathematical formulations that optimization software needs. Rather than requiring a specialist to manually encode scheduling constraints, OptiMind generates Mixed-Integer Programming (MIP) code dynamically from natural language inputs.

"We'd been looking at scheduling problems for a long time, but they're incredibly difficult because every manufacturing environment has different constraints," says DeMaagd. "With OptiMind, we could describe the problem in words, connect it to real operational data, and generate optimization models much more dynamically."

After testing the technology with customer production data, Sight Machine recognized the opportunity to apply it at scale and began collaborating with Microsoft teams to integrate OptiMind into its manufacturing workflows.

The resulting solution works in three layers. Sight Machine's data foundation supplies the real-world inputs: per-SKU line speeds, changeover times, ramp-up durations, and live production status. Data from supply chain and demand forecasting tools is incorporated, and the optimization can also include other factors such as energy, labor, or cost—the span of optimization opportunities is broad. OptiMind takes those inputs, along with customer orders, existing schedules, and natural language instructions, and generates MIP code that encodes the full scheduling problem. 

When conditions on the floor change, live plant data triggers a re-optimization. If a production line slows down or equipment goes offline unexpectedly, teams receive a revised schedule based on current priorities and estimated downtime, without convening a cross-functional meeting.

"What's powerful is that manufacturers now have the ability to replan around changing conditions much more quickly," says DeMaagd. "In the past, that process could take hours and involve a large group of people. Now it can happen in minutes."

Anyone on the team can also explore what-if scenarios in plain English, evaluating the impact of maintenance events, demand shifts, or production constraints before disruptions occur. Sight Machine also uses predictive analytics models built with Azure Machine Learning to help manufacturers anticipate potential slowdowns before they affect production targets.

The broader Microsoft ecosystem helps Sight Machine extend these capabilities across manufacturing workflows. The company uses Microsoft Fabric to centralize operational data. Sight Machine is also exploring expanded use of agents within Microsoft Foundry and Microsoft 365 Copilot to give operators on-demand access to scheduling and operational insights through the tools they already use every day.

Putting AI-driven scheduling to work on the factory floor

For the beverage bottler, the results were immediate and measurable. The biggest improvements came from giving manufacturing teams the ability to continuously optimize production decisions using real-time operational context. Product transitions, ramp-up periods, cleaning processes, and line sequencing all became opportunities for dynamic optimization rather than static planning. By using real-time operational context instead of static averages, the system generated schedules that minimized downtime while maintaining production quality.

Manufacturers have traditionally relied on operational heuristics and static scheduling assumptions to manage production variability. While those rules of thumb often reflect decades of process expertise, they struggle to adapt quickly when demand shifts, equipment performance changes, or production disruptions occur in real time. By grounding scheduling decisions in actual changeover times observed in production and applying mathematically optimal sequencing, the solution uncovered opportunities that rules-based approaches consistently missed.

The numbers show the operational impact of dynamic AI-driven optimization. By continuously adapting schedules around real-world production conditions, the manufacturer significantly reduced operational inefficiencies across the plant. As one example of the many optimization opportunities enabled by the system, changeover-related downtime dropped by nearly 80%. Other improvements included reducing ramp-up delays by nearly 60% and cutting clean-in-place (CIP) sanitation downtime by almost 90%.

"When you're dealing with manufacturing, every percentage point matters," says DeMaagd. "These facilities involve enormous, fixed costs, so even small operational improvements can translate into millions of dollars in value over time."

Dynamic replanning gave teams the ability to respond faster when production conditions changed. Instead of repeatedly pulling engineering leaders and plant managers into manual scheduling meetings, teams could adjust schedules with greater speed and confidence.

"Manufacturers need the ability to adapt quickly when conditions change," says DeMaagd. "The more visibility and operational context they have, the more effectively they can respond without sacrificing efficiency."

Helping manufacturers build more resilient and adaptive operations

For Sight Machine, the work with the beverage bottler points toward a broader opportunity. The company sees AI as a tool not just for optimizing schedules, but for helping manufacturers preserve operational expertise.

"We're not trying to replace people with AI," says DeMaagd. "We're trying to help manufacturers keep operating successfully as experienced workers retire and fewer people enter manufacturing roles."

By embedding that expertise into AI-assisted workflows, manufacturers can scale what previously lived only in the heads of their most experienced operators, making it available across shifts, sites, and teams.

Sight Machine is also exploring how AI agents within the Microsoft ecosystem could eventually support more continuous optimization, where production schedules adjust automatically in response to machine slowdowns, maintenance events, or changing demand, with human oversight guiding rather than driving every decision. The company is actively investigating integrations with Microsoft Fabric IQ, which turns unified plant data into operational intelligence that AI agents can act on, and Foundry Agent Service, Microsoft's fully managed platform for building and deploying agents at scale—bringing Sight Machine closer to a future where scheduling optimization runs continuously in the background, across shifts, sites, and teams.

"There's a huge push right now toward modernizing manufacturing operations," says DeMaagd. "Manufacturers know the world is changing quickly, and they need tools that help them adapt faster, preserve expertise, and operate more efficiently."

By combining Sight Machine's semantic layer with Microsoft Foundry, OptiMind, Azure Machine Learning, and the broader Microsoft ecosystem, manufacturers are moving on from reactive, manually intensive operations toward production environments that can anticipate, adapt, and scale, with or without a full room of engineers in the loop.

Developers and researchers looking to explore what comes next can find OptiMind and other cutting-edge AI experiments from Microsoft Research on Foundry Labs, available at labs.ai.azure.com. Foundry Labs offers a glimpse into potential future directions for AI, from optimization and reasoning models to agentic frameworks and beyond.