The factory plan is obsolete the moment reality changes

Manufacturers can spend hours constructing a production schedule that balances orders against available capacity, only for the assumptions behind it to disappear shortly after the shift begins. A machine fails, a material delivery is late, or an employee is unexpectedly absent, leaving planners with a schedule that may still look orderly on screen but no longer reflects what the factory can do.

Avner Ben-Bassat, CEO of Plataine, sees the real difficulty emerging once disruption starts to spread through the production system. A late material delivery or unavailable machine can affect several connected activities, leaving planners with multiple competing priorities to resolve at the same time. “Visibility is critical, but it is not sufficient,” he says. “It gives you data and, at best, a report or dashboard that somebody still needs to look at and make decisions from. With planning, you have a multitude of decisions that need to happen concurrently and in a coordinated way.”

Once several dependencies begin moving at the same time, planners have to decide which commitments can still be protected, and which parts of the original sequence need to change. The difficulty is no longer simply following the plan but choosing when to depart from it. “A fixed schedule is not only a constraint, but I would also say it is fiction,” he says. “There are so many dimensions intersecting that trying to operate in a completely fixed mode is not realistic. You need somebody to make the decision about what happens when reality is different from the plan.”

Rescheduling means solving the whole factory again

Detecting disruption is becoming easier as manufacturers connect more equipment and production systems, but rebuilding the plan remains difficult because an individual event rarely stays confined to the machine, person or order where it first appeared. The planner has to understand not only what has changed, but how much of the existing schedule is still usable and how far the consequences extend before a revised plan can be trusted.

“The disruption will ripple across multiple aspects very quickly,” Ben-Bassat says. “If I was planning to use a tool at two o’clock, send it to maintenance at three and use it for something else at five, that whole sequence is now disrupted. It then affects other work orders, equipment and machines, so you may have to rerun the plan for a significant portion of the factory.”

Manual scheduling struggles with that level of interdependency. Experienced planners can solve complex production puzzles, but doing so may take hours. By the time a replacement schedule is ready, much of the shift may already have passed and production teams have continued with whatever work remains possible.

Factories often compensate by building buffers into the schedule, reserving extra time between operations for disruption that may never occur. Ben-Bassat sees AI-based scheduling as a way to recalculate much faster while considering more of the production environment simultaneously. “It is not just computing power,” he says. “The AI needs to comprehend the depth and breadth of the process, evaluate millions of options and come back with a plan that is valid and optimal. We cannot create a production plan for a meaningful factory in seconds as human beings. Software can.”

Producing a new schedule quickly has little value if planners still must spend hours making it usable. The real test is whether the result can move directly into execution rather than becoming the starting point for another round of manual planning. “We use the term W6: who does what, where, when and with what,” he says. “There must be a material plan, a tooling plan and a plan for each work order and customer. If the plan only solves part of the problem, the planner still must spend hours finishing it.”

Multi-Ply Components provides a useful example of what changes when planning moves away from spreadsheets. The carbon fiber component manufacturer had been spending hours each week creating production plans manually, with individual departments then developing their own detailed schedules and problems sometimes only becoming visible during production meetings. Its work orders can extend beyond ten weeks, while a single plan may contain more than 10,000 tasks across 40 stations and 19 departments.

After deploying Plataine’s Production Scheduler, work orders transferred from SAP can be converted into a detailed three-month plan in minutes, while operators report task progress back into the system. Multi-Ply says this has improved visibility and resource utilization, reduced planning effort to as little as an hour a week and given the business a way to test scenarios such as adding shifts or investing in additional equipment before committing resources.

The best local decision can still damage production

As production planning becomes more integrated, manufacturers have to account for constraints that historically sat outside the planner’s immediate remit. Decisions made within one function can invalidate assumptions elsewhere, which makes it increasingly difficult to optimize production by looking at individual activities in isolation.

“If you create a production plan and set aside preventive maintenance, you are ignoring part of the reality,” Ben-Bassat says. “Imagine you are working on a CNC machine or press, and somebody arrives to shut it down for maintenance while another operation is waiting for its output. Ignoring those maintenance cycles when you build the plan is an invitation to trouble.”

The broader objective is to avoid what Ben-Bassat describes as a local optimum, where one part of the operation appears efficient only because other constraints have been excluded from the calculation. Bringing maintenance into the same planning process allows manufacturers to optimize around the actual availability of machines and tooling rather than an idealized production schedule.

The planning challenge becomes more complex again when manufacturers operate across multiple sites. Each plant may have its own workloads, resource constraints and priorities, while decisions made at one facility can affect what the wider network can deliver.

“We can look at the entire enterprise as if it is one big factory,” Ben-Bassat says. “If my facility is behind but another facility has capacity, the agents can communicate in the same way two planners would. You can automate the routine work and elevate the exceptions. The same concept can eventually extend across the supply chain.”

Looking across a wider manufacturing network increases the number of possible responses, but not every technically feasible decision should be made automatically. Commercial commitments, workforce implications and customer priorities can all require judgement beyond simply finding available capacity.

Autonomy needs boundaries

As AI takes on more of the scheduling workload, manufacturers need to decide where automation should stop and human authority should begin. That balance is unlikely to be fixed, because confidence in the system can change over time and different operating conditions may justify different levels of autonomy.

“Let the AI deal with the day-to-day work and the usual exceptions, but elevate the major issues to the planner,” Ben-Bassat adds. “An AI agent might not be allowed to approve overtime, but it can identify that overtime is needed and escalate that decision. Once somebody approves it, the system can optimize the plan using that additional capacity.”

Customer priorities can provide another boundary. A manufacturer may allow the system to rearrange deliveries when capacity is constrained while requiring human approval before delaying an order for a strategically important customer.

Greater scheduling speed does not mean every improvement identified by the system should immediately be pushed to the shop floor. Manufacturers still need a point beyond which stability becomes more valuable than another marginal optimization, particularly once work is close to execution.

“You cannot keep juggling the schedule just because you can, because it will drive everyone crazy,” Ben-Bassat explains. “You need to pull material, prepare machines and carry out setups. We therefore have a ‘do not touch’ window. It might be the next two hours, four hours or the next shift. The AI will not automatically change that part of the plan, although a planner can intervene if something serious happens.”

Manufacturers can also examine potential changes away from live production. Plataine’s sandbox capability lets planners test urgent orders, staffing levels or additional capacity before committing those decisions to the shop floor, while Multi-Ply uses scenario simulation to assess future resource requirements and help justify investment decisions.

For Ben-Bassat, the distinction between visibility and adaptability ultimately comes down to whether technology helps the manufacturer reach a decision. Knowing that a machine is down or an order is late is useful, but somebody still must determine what production should do next.

“If I give you a map showing roadblocks and traffic jams, but you still have to work out which roads to take, I have only given you visibility,” he says. “The useful system figures out the route and gives you the instructions. There are places where we absolutely need a human in the loop, but AI can remove much of the work involved in getting to the decision.”

A production plan will always begin with assumptions about what should happen. Adaptability comes from accepting that those assumptions will change and giving the factory a way to respond without turning every disruption into another round of manual replanning.