Schedules rewritten on the floor? APS must land in dispatch and kitting—not just a pretty Gantt
APS ran out of a beautiful Gantt chart, and the squad leader changed it on the whiteboard with chalk - not rebellion on the spot, but the plan did not catch everything and abnormalities. If the plan cannot be executed and rewarded by the dispatch, it is just a diagram.
What you'll take away
- There is no complete set of constraints for scheduling, the essence is the ideal job sorting, and the on-site change is a rational choice
- Dispatch is the execution of the plan; reassignment must leave a mark, with a reason code, and write back the credibility of the plan
- APS must chain kit checks, dispatch issue, and shop-floor feedback—any broken segment makes the whole loop idle.
- Plug-in sheets should have rules: who can plug in, which layer to plug in, who to squeeze out, who must be visible
- Do a closed-loop day/week dispatch with limited capacity first, then pursue advanced optimization algorithms
The Gantt chart is still flashing, but the whiteboard has already been updated.
APS calculates the 'optimal' sequence at night, and the morning meeting team leader rearranges a version with chalk. Planners feel rebellious on the shop floor, while the workshop feels that key materials are far from complete and that molds can only be used one set at a time. Both sides are acting rationally: if the system lacks completeness checkpoints, reassignment reasons, or feedback loops, the Gantt chart is just a wish list. What's even more dangerous is that after changes are made, the system still considers the original plan valid—the planner continues to push materials according to the old chart, and sales continue to respond to customers based on the old delivery dates. Changing plans on-site is not scary; what's scary is thinking the original plan is still valid after changes—once this duality forms, no matter how good the achievement rate looks, it's self-deception.
The value of a production scheduling system does not lie in 'squeezing a few more points for changeover time,' but in making changes visible, being able to block incomplete sets, and allowing execution to be written back. First, ask whether the rescheduling discussed in the morning meeting has a reason code, or if material shortages can be blocked before dispatching work orders, then ask about algorithm parameters. When a beautiful Gantt chart cannot handle dispatching and material readiness, the optimization engine is just an expensive drawing tool — this is one of the most costly misconceptions in workshop digitization.
A closed loop with four sections; if any one section is broken, it will run idle.
| Stage | Input | Failure symptoms | The point to be welded shut |
|---|---|---|---|
| Plan | Order/Work Order Process Route Resource Calendar Standard Working Hours | The Gantt chart looks good, but the key constraints were not included in the model. | Limited production capacity; equipment/mold/skills occupation upon entry |
| complete set | Materials, molds, tooling, and personnel qualifications when necessary | Only after dispatching did we realize there was a material shortage; a large-scale rescheduling in the morning. | Pre-dispatch inspection; shortage interception or risk release |
| Dispatch work | Task assignment to person/machine and executable time window | Paper-based dispatch, verbal reassignment, delayed status | Dispatch is in execution state; change dispatch reason code authorization |
| Return | Start/Complete/Pause/Abnormal | The credibility of the plan has collapsed; the achievement rate depends on luck. | Scan the code to report update progress and remaining load |
Production scheduling without complete constraints is essentially an ideal task sequence—changing it on site is a rational choice. If unreported dispatches still occupy capacity, next week's schedule will continue to assume 'resources are fully available,' reinforcing errors on its own. When standard times are not maintained for a long time, the limited capacity model is fundamentally skewed—it cannot be fixed by buying a more expensive solver. If changeover times are hidden within 'flexibility' and not counted as resource usage, the optimization engine will favor frequent changeovers, while the site will use chalk to revert to concentrating by product type—yet another split between 'system optimal' and 'site optimal.'
The kit-up strategy should be layered according to materials: Class A critical parts default to hard blocking; Class B can have substitutes configured or allow production to start despite shortages, but the risk must be highlighted in red on the work order; Class C auxiliary materials should not block the whole order. Always using 'start only when fully complete' will almost certainly be circumvented; always using 'dispatch even if incomplete' returns to the ideal sequencing. The resource calendar should at least cover equipment shifts, planned maintenance, mold occupancy, and key skilled personnel—without molds and skills, the system will mistakenly treat 'machine idle' as 'able to work.' When inventory and work-in-progress records do not match reality, the kit-up check conclusions are unreliable, and the gate approval will turn into another verbal negotiation; first reconcile the records, then discuss automatic interception.
Insertion rules should be written earlier than solver parameters.
| Rules | Explanation |
|---|---|
| Authorization | Who can be inserted at the process level/work order level/cross-production line; skipping levels requires dual confirmation |
| Visible effect | Automatically list which work orders are being removed or postponed, and notify the follow-up/sales staff |
| Leave a mark | Reason code (customer urgent, quality rework, equipment failure insertion, etc.) Applicant Time |
| Review | Daily/weekly rush order rate and on-time rate of displaced work orders are displayed on the dashboard; if there are too many rush orders, prioritize improving demand management |
Inserting orders by itself is not a sin. A lack of rules, lack of visibility of impact, and still allowing sales to sell according to old promises externally—that is the real breeding ground for delivery failures. If shop floor order insertions occupy capacity, the minimal closed loop is: save the insertion → list the affected work orders/order lines → confirm with the planner whether to change the customer promised date. If the shop floor absorbs the delay internally while the sales commitment remains unchanged, it is the classic path to unfulfilled on-time delivery. Sharing the 'impact visibility' logic with customer-side delivery commitment management is more important than each side having its own exception channel. Daily execution can be adjusted within the team leader’s authorized scope; adjustments across shifts, across production lines, and affecting delivery commitments go back to the planner—the authority should be written into the system so that there is a basis for dispute resolution. Retaining a verbal order insertion channel while having fully automated scheduling will empty the automation through exceptions—exceptions must be made explicit and included in statistics.
How can the system handle a 'command' rather than just a 'plan'?
Compare with the APS scheduling, work order dispatching, and barcode reporting in the manufacturing execution plan:
| Ability | Key points for implementation |
|---|---|
| Planning layer | Work orders are expanded to operations, binding the process route and standard working hours; the resource calendar includes maintenance and already occupied times |
| Assembly layer | Aligned with inventory/in transit; constraints on mold availability status; default not dispatchable or strong alert if not complete |
| Dispatch layer | Task assigned to team/equipment; the board displays work in progress, pending start, and reasons for blockage |
| Reward layer | Scan code to start and complete work, pause reasons, and return of defective quantities; capacity continues to be occupied if not reported |
| Quality Gate | For the first piece that fails or does not meet standards in the isolated process, the downstream dispatch will be automatically frozen. |
| Delivery risk | Plan changes that lead to committed date risks are exposed to the order follow-up, not just handled internally within the workshop. |
JEDP-MES puts it plainly: when completeness is not met, interception or explicit exceptions should occur, and inserting or changing orders must be authorized and traceable—without completeness constraints, a Gantt chart is just wishful thinking, and the plan achievement rate will not match the site. If the progress board does not display the reasons for blocking (material shortage, waiting for mold, quality gate), the shift leader will still return to chalk and verbal scheduling, and the system dispatch board will be virtually useless.
Don't use a single number to cover up chaos in achievement rate
| Indicator | Definition proposal | What will be missed when looked at individually |
|---|---|---|
| Work order on-time completion rate | Promise or plan to close by the completion date | Repeated reassignment and waiting during the process |
| On-time start rate of processes | Actual start of work in the dispatch window | Saving the completion by working overtime |
| Completion Rate | Proportion of first-pass inspections before job dispatch | Forcing changes on the spot |
| Proportion of reassignment without reason | Change to the missing reason code or select 'Other' | Seemingly achieved on the surface, actually out of control |
| Order Insertion Rate | Number of last-minute work orders / Total number of dispatched work orders | Demand management issues are treated as workshop problems |
Plan reliability can be operationalized: if the consistency between the 'system planning sequence and the morning meeting execution sequence' is below the threshold for two consecutive weeks, do not move on to heavier optimization—first fix completeness and returns. Consistency can be measured by the proportion of overlap in the sequence of operations, without needing complex algorithms. Quality gates should also be included in dispatching: if the first piece fails or operations in quarantine are nonconforming, downstream dispatching should be automatically frozen; otherwise, plan achievement will be falsely inflated by 'starting work with defects.'
Ninety Days: Close the Loop First, Then Talk About Intelligent Scheduling
| Order | Do it first | Sign of effectiveness |
|---|---|---|
| 1 | Limited capacity daily/weekly dispatch board Mandatory reassignment trace | The disconnect between the plan and the actual situation is immediately apparent. |
| 2 | Key components and mold set inspection assignment gate | The number of sequence changes in the morning decreased significantly |
| 3 | Scan code to start work and complete work, and the return rate meets the standard | The plan's reliability is measurable |
| 4 | Insert order authorization and impact visibility | Reduced disputes over delivery dates |
| 5 | Further introduce APS optimization (model change aggregation, delivery priority, etc.) | The algorithm is built on trusted execution |
In the first month, get the dispatch board and the reason codes for changes running, and track sequence changes in morning meetings; in the second month, get all the gates in place and ensure the yield meets the standard; in the third month, visualize the impact of rush orders, then evaluate whether to implement optimization. A more realistic goal within ninety days is: sequence changes can be tracked, completeness can be intercepted, returns are trustworthy. Don’t rush to buy 'smart scheduling' right away. When master data (processes, work hours, calendars) and yield are not up to standard, optimization results will only be quickly overturned on the floor. Algorithms are amplifiers; what they amplify is the master data and execution discipline—without discipline, what they amplify is chaos.
Anti-pattern quick notes: Only produce Gantt charts without issuing work orders; performing completeness checks after the fact to assign blame; shift leaders infinitely reassigning without reason; fully automated scheduling yet still allowing verbal rush orders; planners finishing work on behalf of the shop floor leading to false reporting rates; forcing start only when everything is complete, pushing the shop floor to bypass processes. Process routes and standard times are inaccurate or frequently manually modified without trace, inventory records don’t match physical stock, key process reporting rates are below usable thresholds, planning and workshop are adversarial and unwilling to define rush order authorizations, or the product structure is extremely simple with almost no resource constraints— these situations can be addressed first with governance rules or using a dispatch board, with limited benefit from complex APS. Even without an APS engine, one can first implement a dispatch board for limited capacity with completeness checks, processing, and reporting, which is often more valuable than a null-running algorithm.
This week, let's first tally two accounts.
Don't rush to hold an intelligent scheduling project meeting. First, collect statistics for a week on 'number of sequence changes in morning meetings' and 'proportion of unreasoned reassignments,' then use the numbers to decide whether to prioritize full-set welding or feedback welding. If the sequence changes don’t decrease, implement the optimization engine—it’s equivalent to accelerating chaos. The main cause being material shortage means give priority to full-set gates; if it’s equipment/mold conflicts, prioritize the resource calendar; if it’s urgent customer orders, prioritize how order insertion affects visibility and customer commitment. If the proportion of unreasoned reassignments doesn’t decrease for two consecutive weeks, it indicates that execution discipline hasn’t been established. Before the dispatch feedback rate meets the standard, any 'intelligent order insertion suggestions' will be based on expired loads—the system thinks resources are full, while the site is actually waiting for materials or molds. Close the loop first, then optimize: this statement is more important than any solver parameter and is the shortest path for a Gantt chart to turn from a 'diagram' into a 'command.' If the sequence is reversed, it will only crash faster. First command, then diagram—that’s the only way to stand firm.
- MES Manufacturing Solution: APS Scheduling, Work Assignment Reporting, and Progress Visualization
- Scan code to report work collection point
- OEE three rates
- Trace from batch number to workstation
Want to apply this in your plant?Resources has matching self-assessments and templates—or book a live demo.
FAQ
Quick answers on this topic—still stuck?Contact
Can we assign tasks first without an APS engine?
Yes. Using dispatch plates with limited capacity + complete inspection + return to completion is often more valuable than idling algorithms. The algorithm is an amplifier, which amplifies the master data and enforcement discipline.
How is the plan achievement rate calculated?
It is recommended to look at both the on-time completion rate of the work order and the on-time start rate of the process. Just looking at the former will cover up the repeated reassignments and waiting in the process.
How to schedule equipment and mold constraints?
as a resource calendar with occupancy. When key resources (equipment, molds, tooling, personnel skills) are not available, the system should directly prevent dispatch, rather than rely on on-site discovery after dispatch.
Whose instructions should be followed when the planner and the team leader conflict?
Daily execution can be adjusted within the scope of the shift supervisor's authorization; adjustment to planners across shifts, production lines, and delivery commitments. Permissions are written into the system, and the dispute has a basis for adjudication.
To what extent must everything be complete before work can be assigned?
At least key parts are complete with the mold/tooling; non-key parts can be configured to replace or allow shortage to start but must expose risks. In multi-variety scenarios, it is often too strict to start work before it is complete, and if it is not uniform, it is equal to no constraints.