AI in Manufacturing: Where It Works Today, What It Needs, and What Comes Next
AI adoption is no longer the hard part for manufacturers. Scaling it is. RSM found that 88% of 129 manufacturing respondents in its 2026 middle-market AI survey had AI at least partially integrated, yet McKinsey found that 46% of surveyed manufacturing COOs still cite data or IT/OT limitations as a barrier to implementation.
That gap changes the useful question. Instead of asking where AI can be used, operations leaders need to know which use cases are mature enough to pursue, what data each one depends on, which KPI proves value, and where ERP is actually part of the answer.
Key Takeaways
- AI in manufacturing already spans maintenance, quality, production planning, supply chain, engineering, and ERP analytics. The technology is broad; the data and operating context for each use case are different.
- Adoption is high, but maturity is uneven. RSM reports 88% partial or full integration among its manufacturing respondents, while MLC says manufacturers still describe themselves as early in the AI maturity curve.
- The bottleneck is usually data, workflow, or governance. McKinsey found 46% of surveyed manufacturing COOs cite data or IT/OT limitations, and nearly 60% of manufacturers lack clear targets for measuring AI value.
- ERP matters most when the AI decision needs commercial and transactional context – orders, inventory, costs, suppliers, schedules, margins, or work orders. Machine telemetry, images, and process signals usually come from
- MES, historians, QMS, CMMS, cameras, or OT systems.
- The practical path is use case -> data -> decision -> workflow -> KPI. A pilot is useful only if the team knows what decision improves and how the output reaches the person or system that can act on it.
- Fully autonomous factories are not the near-term baseline. In MLC’s 2026 Industrial AI survey, 0% expected all factories to be fully autonomous by 2030; most expected higher automation with people still playing a meaningful role.
What Is AI in Manufacturing?
AI in manufacturing is the use of machine learning, computer vision, language models, generative AI, and agentic systems to predict outcomes, interpret operational data, recommend actions, or automate defined tasks across engineering, production, quality, maintenance, planning, and supply chain workflows.
Those technologies solve different problems. Predictive models estimate what is likely to happen. Computer vision classifies images or detects defects. Generative AI creates or summarizes content. Agentic AI can execute multi-step tasks within defined permissions. Treating all four as one category makes vendor comparisons and ROI discussions much harder than they need to be.
RSM’s 2026 manufacturing respondents reported current use of generative AI (79%), language AI (63%), prediction AI (57%), and agentic AI (56%).
Where Manufacturers Are Using AI Today
The strongest use cases share one trait: the model has a defined operational decision to improve. The table below separates the use case from the data and KPI behind it, which is the part most generic AI-in-manufacturing articles skip.
| Use case | AI task | Primary data | Operational decision | Useful KPI |
|---|---|---|---|---|
| Predictive maintenance | Estimate failure risk / remaining useful life | Sensor, historian, maintenance, asset records | When to inspect, service, or replace | Unplanned downtime, MTBF, maintenance cost |
| Visual quality inspection | Detect / classify defects | Images, labels, process parameters, QMS data | Accept, reject, rework, or inspect further | FPY, defect escape, scrap/rework |
| Production planning | Predict constraints and recommend schedules | ERP orders/materials + MES/APS capacity and progress | What to run, when, and with which resources | Schedule adherence, throughput, OTD |
| Demand and inventory | Forecast demand and exceptions | Orders, forecast, inventory, lead times, supplier data | What to buy/build/hold | Forecast error, stockouts, inventory turns |
| Engineering / generative design | Generate or evaluate design alternatives | CAD/CAE, design rules, materials, constraints | Which designs to test or release | Engineering cycle time, weight/cost, manufacturability |
| ERP analytics / GenAI | Query, summarize, detect anomalies | ERP transactions, master data, KPIs | What needs attention and why | Decision latency, analyst time, exception resolution |
Predictive Maintenance
Predictive maintenance works when machine condition data can be tied to maintenance history and a real maintenance decision. A model that predicts a bearing failure has limited value if the team cannot see the spare part, technician availability, planned production load, or work-order priority needed to act on it.
ERP may provide asset cost, spare inventory, purchasing, and work-order context, while sensor and condition data usually come from machines, historians, or CMMS/MES systems. The integration point is what turns a prediction into scheduled work.
AI-Powered Quality Inspection
Computer vision can inspect parts at line speed and flag defects that are difficult to catch consistently with manual sampling. The model still needs labeled examples, a stable imaging setup, and a decision path for reject, rework, escalation, or additional inspection.
For quality teams, the useful metric is not simply model accuracy. False accepts, false rejects, first-pass yield, scrap, rework, and defect escape show whether the system improves the process.
Production Planning and Scheduling
AI planning use cases combine demand, material availability, routing, capacity, labor, machine constraints, and current shop-floor progress. That makes planning one of the clearest examples of why ERP data alone is rarely enough: the order and material picture may live in ERP, while real production state comes from MES, APS, or the shop floor.
A recommendation matters only if it improves schedule adherence, throughput, on-time delivery, or constraint resolution without creating instability somewhere else in the plan.
Supply Chain and Inventory
Manufacturers use AI to forecast demand, flag supplier risk, identify inventory anomalies, and recommend replenishment or expediting decisions. These models become more useful when they combine transactional history with current demand, lead times, supplier performance, and production requirements rather than forecasting from sales history alone.
Generative AI and ERP Analytics
Generative AI is particularly useful when the friction is getting an answer from existing data rather than predicting a physical event. Users can ask questions in plain language, summarize exceptions, compare performance, or investigate why a KPI moved without waiting for a new report.
WM Synergy’s InsightsAI is positioned around this narrower problem. It connects to supported ERP databases and provides conversational analytics, visualizations, anomaly detection, and role-based views across manufacturing, costing, inventory, procurement, scheduling, sales, and finance.
That is different from computer vision, predictive maintenance, or generative design. Keeping the use cases separate makes the product fit clearer and keeps the article from making ERP sound like the source of every AI initiative.
AI Adoption in Manufacturing: What the 2025-2026 Data Actually Shows
AI usage is growing quickly, but the surveys measure different populations and different forms of adoption. Read them as signals of direction rather than one universal manufacturing adoption rate.
- RSM: 88% of 129 manufacturing respondents said AI was at least partially integrated, including 32% reporting full integration across core operations and processes.
- Manufacturing Leadership Council: 70.9% of respondents said they currently use GenAI tools such as ChatGPT or Microsoft Copilot in manufacturing operations; 89.2% planned a moderate or substantial increase in GenAI use over the next two years.
- MLC governance: 73.8% said GenAI tool selection and use now follow a corporate policy, while 66% said they currently use or plan to use agentic AI in manufacturing operations.
- PwC / Strategy& / VDMA: among 247 industrial manufacturers in Germany, Switzerland, and Austria, 91% planned GenAI investment in 2025, but only 29% had rolled out at least one use case and 7% had a systematic implementation approach.
What Manufacturers Should Expect AI to Improve
Depending on the use case, AI can improve output, capacity, productivity, downtime, quality, forecast accuracy, inventory, engineering cycle time, or decision speed.
Deloitte’s 2025 Smart Manufacturing and Operations Survey of 600 executives from large U.S. manufacturers reported average net improvements since smart-manufacturing implementation of 10%-20% in production output, 7%-20% in employee productivity, and 10%-15% in unlocked capacity.
Reporting, Predictive AI, Generative AI, and Agents Solve Different Problems
The useful distinction is not “traditional analytics versus AI.” Modern analytics platforms can be real-time and predictive, while AI can also summarize historical data. Compare the job each tool performs.
| Approach | Primary question | Typical output | Manufacturing example |
|---|---|---|---|
| Reporting / BI | What happened? | Dashboard, KPI, drill-down | OTD, scrap, labor variance |
| Predictive / ML | What is likely to happen? | Forecast, probability, anomaly score | Failure risk, demand forecast |
| Generative AI | What does the data mean / how can I interact with it? | Answer, summary, explanation, content | Ask ERP why margin dropped |
| Agentic AI | What approved multi-step task can be executed? | Actions across tools with rules/permissions | Create an exception workflow or proposed replenishment action |
Why AI Pilots Stall Before They Scale
The common failure is not a shortage of ideas. It is the distance between a promising demo and a production workflow that has reliable data, an owner, a decision right, security controls, and a KPI.
McKinsey reports that 46% of surveyed manufacturing COOs cite limitations in data or IT/OT systems, while nearly 60% of manufacturers lack clear targets for measuring AI value. The same research also points to culture and reskilling as major implementation barriers.
For a mid-market manufacturer, that creates four practical readiness questions: Is the source data trusted? Can the model reach the systems needed for context? Does a person or workflow own the decision? Can the business measure whether the decision improved?
Where ERP Fits in a Manufacturing AI Architecture
ERP is often the business context layer for manufacturing AI. It holds orders, inventory, suppliers, costs, routings, work orders, schedules, customers, and financial outcomes. But it usually does not hold every signal an AI use case needs.
- Predictive maintenance may need historian/IoT telemetry + CMMS/maintenance history + ERP parts and cost data.
- Computer vision may need images/camera data + QMS labels + process settings + ERP lot/serial context.
- Planning may need ERP demand/materials + MES/APS capacity and execution status.
- Supplier-risk or inventory AI may lean heavily on ERP transactions plus external supplier/logistics data.
- Conversational ERP analytics can work primarily from the ERP data model when the question is financial or operational rather than physical.
This architecture view is more useful than telling every manufacturer to “integrate AI with ERP.” It shows what ERP contributes and where another system is the source of truth.
What the Future of AI in Manufacturing Looks Like
The near-term direction is more automation and more AI-assisted decisions, not a universal move to lights-out factories. MLC’s 2026 Industrial AI survey found that 0% expected all factories and plants to be fully autonomous by 2030. Most respondents expected higher automation while people continued to play a significant role, or said fully autonomous operations remained a long way off.
More Agentic Workflows – With Guardrails
Agentic AI will move some use cases from recommendation to execution: opening a workflow, gathering context from several systems, proposing a schedule change, or creating a replenishment action. In manufacturing, autonomy has to be bounded by permissions, approval thresholds, safety rules, and clear rollback paths.
More AI at the Edge
Inspection, anomaly detection, and machine monitoring will increasingly run closer to the equipment where latency, reliability, or data-volume constraints make cloud round trips impractical. That shifts part of the AI architecture toward cameras, controllers, gateways, and edge compute.
A Hybrid Human-AI Factory
The more realistic workforce model is people making higher-context decisions with AI handling pattern recognition, repetitive analysis, and tightly defined actions. Deloitte found 85% of its surveyed manufacturers expect smart-manufacturing initiatives to transform how products are made, improve agility, and help attract talent.
More Automation, but Economics Still Decide Where
PwC’s 2026 industrial manufacturing outlook found the median share of manufacturers expecting highly automated key processes to rise from 18% today to 50% by 2030. That points to wider automation, not identical autonomy across every plant or process.
How to Get Started With AI in Manufacturing
Start with a decision the business already makes and a metric it already cares about. Then work backward to the data and system changes needed to improve that decision.
- Choose the business problem before the model. Define the operational loss or delay: downtime, scrap, late orders, excess inventory, slow analysis, or another measurable problem.
- Write down the decision the AI should improve. “Predict a failure” is incomplete; “create a maintenance window before failure without missing the production plan” is a decision.
- Map the data sources. Identify which data belongs in ERP, MES, QMS, CMMS, historian, PLM, cameras, external feeds, or spreadsheets – and which source is authoritative.
- Baseline the KPI before the pilot. Measure the current downtime, FPY, OTD, forecast error, inventory turns, or decision time so the pilot has something to beat.
- Design the human and system handoff. Decide who reviews the output, what the model may automate, what requires approval, and how exceptions are handled.
- Scale only after the workflow works. Reusable data pipelines, governance, security, monitoring, and ownership matter more at scale than the pilot model itself.
Where WM Synergy Fits
WM Synergy is most relevant when the AI use case depends on manufacturing ERP data or when the ERP data model is part of the readiness problem. That includes conversational analytics, cost and margin analysis, inventory and purchasing exceptions, scheduling insight, and the work required to make ERP data usable alongside other manufacturing systems.
For ERP-centered analytics, InsightsAI supports Infor VISUAL, Infor SyteLine/CSI, Acumatica, and Infor XA, with conversational querying, visual answers, anomaly detection, and role-based intelligence. For the broader architecture question, see WM Synergy’s AI Solutions for ERP.
For maintenance vision, robotics, or other OT-heavy use cases, ERP is only one part of the stack. The discovery work should identify the systems and data needed before choosing the AI product.
Conclusion
AI is already useful in manufacturing, but value does not come from adding the word “AI” to an existing workflow. It comes from improving a decision that matters: when to maintain a machine, whether to stop a defect, what to run next, how much inventory to hold, or why margin moved.
The manufacturers that scale AI will be the ones that can connect those decisions to reliable data, clear ownership, and measurable outcomes. ERP often supplies critical context, but the right architecture starts with the use case rather than forcing every use case through ERP.
If you want to identify where ERP data can support a practical AI use case, Schedule Your 30-Minute ERP Strategy Call.
Frequently Asked Questions
Manufacturers use AI for predictive maintenance, computer-vision quality inspection, production planning, demand and inventory forecasting, supply-chain analysis, engineering and generative design, robotics, and conversational analytics over business data. The systems and data required differ by use case.
The measurable benefits can include less unplanned downtime, higher first-pass yield, lower scrap and rework, better schedule adherence, more accurate forecasts, lower inventory, faster engineering cycles, and shorter decision time. The relevant benefit depends on the use case and should be baselined before implementation.
Data quality, IT/OT integration, legacy processes, workforce skills, governance, and weak ROI measurement are common barriers. McKinsey found 46% of surveyed manufacturing COOs cite data or IT/OT limitations, while nearly 60% of manufacturers lack clear targets for measuring AI value.
ERP supplies business context such as orders, inventory, routings, suppliers, costs, schedules, and financial outcomes. AI use cases that depend on machine telemetry, images, quality data, or real-time execution typically need ERP combined with MES, QMS, CMMS, historian, PLM, OT, or external data.
Agentic AI can plan and execute multi-step tasks across defined tools and workflows rather than only returning a prediction or text response. Manufacturing examples may include gathering exception data, proposing a schedule or replenishment action, and routing it for approval. Permissions, safety constraints, human review, and auditability are important before allowing autonomous actions.
Some operations will become more autonomous, but current manufacturing surveys do not support a near-term shift to universally lights-out factories. In MLC’s 2026 Industrial AI survey, 0% of respondents expected all factories to be fully autonomous by 2030; most expected more automation with people still playing a role.
Start with one measurable operational problem, define the decision AI should improve, map the required data, baseline the KPI, and design the human or system action that follows the output. Scale after the workflow produces measurable value, not simply after the model works in a pilot.
By Synergy Resources