Much of the current conversation about artificial intelligence in manufacturing focuses on what AI can do. My work evaluating AI in manufacturing and AI in operations management has highlighted a different question: Is the answer operationally realistic enough to use?
AI can produce a polished answer in seconds.
It can explain Lean manufacturing, build a project plan, recommend inventory levels, outline an ERP implementation, analyze a supply chain problem, or suggest ways to improve production throughput.
The problem is that a polished answer is not necessarily a workable answer.
I’ve spent much of my career in manufacturing and operations, from industrial engineering and quality through engineering management, operations leadership, ERP/MES development, supply chain, medical devices, and executive management. More recently, I’ve also worked as a domain expert evaluating AI-generated material.
That combination has reinforced something I already suspected.
AI is getting very good at explaining how businesses are supposed to work. It still has trouble understanding how they actually work.
That distinction matters.
A recommendation that looks perfectly reasonable on a screen can fall apart quickly when it meets a real production floor, an overloaded work center, an unreliable supplier, an ERP system with bad master data, a regulatory constraint, or a customer who needs an order tomorrow morning.
Here are some of the problems I see repeatedly.
1. AI Often Treats Manufacturing Like a Clean Mathematical Model
Manufacturing rarely behaves like a textbook example.
A model may assume that a machine produces 50 units per hour because that is its theoretical rate. In the real world, that machine may require setup, cleaning, inspection, material replenishment, tool changes, maintenance, operator breaks, quality holds, and rework.
Those aren't minor details.
They determine actual capacity.
The same problem appears when AI analyzes labor. It can divide required hours by available employees and produce an answer that looks precise. But does every employee have the required training? Can they work every operation? Are there bottleneck work centers? Are certain processes dependent on one experienced operator?
Capacity is not simply available hours divided by standard hours.
Real operations have constraints.
2. AI Can Confuse ERP, MES, MOM, and WMS Responsibilities
Enterprise systems are another area where technically correct definitions can lead to poor recommendations.
ERP, MES, Manufacturing Operations Management, and Warehouse Management Systems overlap, but they do not serve identical purposes.
An ERP system may contain work orders, inventory balances, routings, purchasing, demand, and financial data. MES is much closer to actual production execution. MOM may encompass broader manufacturing activities such as quality, performance, maintenance, and production operations. WMS focuses on warehouse execution and inventory movement.
AI sometimes recommends adding functionality to one system without considering where that information should originate, who owns it, how frequently it changes, or which system should be the system of record.
That creates duplicate data, conflicting transactions, unnecessary integrations, and confused users.
The technology question should not be:
Can the system do it?
The better question is:
Where should this process live, and why?
3. AI Underestimates the Importance of Bad Data
AI loves data.
Manufacturing systems are full of it.
That does not mean the data is good.
An organization may have inaccurate routings, obsolete BOMs, incorrect lead times, duplicate items, bad units of measure, outdated supplier information, inaccurate inventory balances, or standards that haven't been reviewed in years.
Adding more analytics to bad operational data doesn't fix the problem.
It can make the wrong answer look more convincing.
Before recommending advanced forecasting, optimization, scheduling, or AI-driven decision support, I want to know whether the underlying data can be trusted.
Sometimes the smartest AI project begins with something much less exciting.
Clean the master data.
4. AI Tends to Optimize One Problem at a Time
Operations management is full of tradeoffs.
Reducing inventory can improve working capital but increase stockout risk.
Increasing batch size can reduce setup costs but increase lead time and work-in-process.
Adding inspection can improve defect detection but slow throughput.
Increasing machine utilization can actually hurt overall flow if it creates excess WIP in front of the next constraint.
Choosing the lowest-cost supplier can increase transportation cost, lead time, quality problems, or supply risk.
AI often identifies a reasonable local optimization without fully accounting for what happens somewhere else in the system.
Operations leaders can't optimize isolated metrics.
They have to optimize the business.
5. AI Can Recommend Best Practices Without Asking Whether They Fit
There is no shortage of management best practices.
Lean.
Six Sigma.
Agile.
Kanban.
JIT.
Predictive maintenance.
Automation.
Digital twins.
Advanced planning.
AI forecasting.
They can all be valuable.
They can also be misapplied.
I become skeptical whenever a recommendation jumps immediately to a methodology or technology without first establishing the operating problem.
A company does not need Lean because Lean is good.
It needs Lean when Lean methods address a specific operational problem.
The same applies to AI.
Start with the business problem. Then determine whether AI belongs in the solution.
6. AI Sometimes Ignores the Cost of Implementation
A technically superior solution may still be a bad business decision.
Every operational change has a cost.
Software licenses.
Integration.
Implementation.
Training.
Process disruption.
Validation.
Maintenance.
Consulting.
Internal labor.
Data conversion.
Change management.
Downtime.
The relevant question is not whether something can improve the process.
The question is whether the value of the improvement justifies the total cost and risk required to achieve it.
I’ve seen plenty of projects where the technology worked exactly as designed and the business case still failed.
Technology success and business success are not the same thing.
7. AI Underestimates Change Management
A process can be technically correct and still fail because people don't use it.
Employees create workarounds.
Supervisors maintain spreadsheets.
Planners override the system.
Operators discover that the new workflow requires twice as many clicks.
Managers stop trusting reports.
Soon the old process is running beside the new one.
That is not primarily a software problem.
It is an implementation problem.
A good operational recommendation has to consider who will use the process, what changes for them, how exceptions will be handled, and whether the new method actually makes the work easier or better.
8. AI Often Gives Answers Before Asking Enough Questions
This may be the biggest issue of all.
A strong consultant, project manager, engineer, or operations leader usually starts by asking questions.
What is actually happening?
What outcome are we trying to improve?
How is the metric calculated?
Where does the data come from?
What constraints exist?
What has already been tried?
Who owns the process?
What happens when something goes wrong?
What does success look like?
AI often jumps directly to the recommendation.
That is useful when the problem is simple.
It is dangerous when the problem is operationally complex.
A Simple Test for AI Recommendations
When evaluating an AI-generated recommendation for manufacturing or operations, I use a fairly simple mental checklist.
Is it factually correct?
The terminology, calculations, assumptions, and technical concepts need to be right.
Is it operationally realistic?
Could people actually execute this process under normal working conditions?
Does it recognize constraints?
Capacity, labor, quality, regulatory requirements, suppliers, systems, cash, and time all matter.
Does it consider the entire process?
Improving one metric while damaging another is not optimization.
Is the data trustworthy enough to support the recommendation?
Garbage in still produces garbage out, even when AI is involved.
Is there a reasonable business case?
The benefit should justify the cost, risk, and organizational effort.
Can it actually be implemented?
A recommendation without ownership, process design, training, controls, and adoption is just an idea.
AI Still Needs People Who Know the Work
None of this means AI isn't useful.
Quite the opposite.
I use it constantly.
AI can accelerate research, analysis, documentation, brainstorming, requirements development, process design, data interpretation, project planning, and decision support.
The productivity potential is enormous.
But the strongest results come when AI is paired with people who understand the work deeply enough to recognize when an answer is wrong.
That is why domain expertise becomes more important as AI improves, not less.
When AI produces obviously bad answers, almost anyone can identify the problem.
When it produces an answer that is 90 percent correct, the missing 10 percent becomes much harder to see.
And in manufacturing and operations, that missing 10 percent may include the bottleneck, the regulatory requirement, the quality risk, the bad assumption, or the implementation constraint that determines whether the recommendation succeeds or fails.
The Opportunity Ahead
I don't think the future of manufacturing and operations is AI replacing experienced operators, engineers, managers, project leaders, or supply chain professionals.
I think it is something much more useful.
Experienced people using AI to analyze faster, see more possibilities, test ideas, improve decisions, automate repetitive work, and spend more time on the problems that actually require judgment.
AI brings extraordinary capability.
Domain experts bring context.
The real advantage comes from combining the two.
Applying AI to Manufacturing or Operations?
RyteTek helps organizations connect AI, ERP/MES/MOM, supply chain, process improvement, and operational expertise to real business problems.
Contact RyteTek to discuss your manufacturing or operations challenges.
