How Do You Alert When ERP-to-Shop-Floor Handoffs Fail?
To handle a failed handoff between ERP and the shop floor, treat every production order as a tracked transaction with an expected acknowledgement. Then decide in advance how each kind of failure is detected, who is alerted, whether a retry is safe, and when a person takes over. The four failures to plan for are lost messages, duplicates, late delivery and rejection. Each one needs a different response, so a single "interface error" alarm is not enough.
Why is a failed order handoff hard to notice?
Most order interfaces are built for the normal case. ERP releases an order, middleware or the MES receives it, and the line starts work. When something breaks, the failure is often silent. Nothing crashes. The order just never shows up at the line, or it shows up twice.
The person who released the order assumes the line has it. The line can't miss an order it was never told about. By the time someone notices, it is usually because a shipment is short or a material count does not reconcile.
So handoff alerting has to watch for missing outcomes, not only explicit errors. That means each order needs a state that both sides can see: sent, received, accepted, started, completed. You also need a rule for how long each state is allowed to last.
If you are still designing the exchange itself, start with exchanging production orders between ERP and the shop floor. This post assumes the order flow exists and covers what happens when it fails.
What are the failure modes, and how should each be handled?
The table below sets out one practical pattern. Agree the specific time limits and owners with your production, planning and IT teams. The values here are not universal.
| Failure mode | How to detect it | Who is alerted | Retry rule | Human fallback |
|---|---|---|---|---|
| Message lost | Order sent but no receipt acknowledgement within the agreed window | Interface support first, then planning if unresolved | Resend with the same order identity; the receiver must ignore an identity it already holds | Planner confirms with the line by phone or terminal and enters the order manually under a controlled procedure |
| Duplicate | Receiver sees an order identity it already holds, or two active orders share one ERP reference | Interface support; production supervisor if both reached the line | No retry. Discard the second copy and log it | Supervisor checks whether any material was issued or produced against the copy and reconciles it |
| Late | Receipt acknowledged, but the order is not accepted or started before its planned start time | Production supervisor | Usually none; the message arrived | Supervisor resequences or confirms the delay with planning, who update the ERP date |
| Rejected | Receiver returns an explicit rejection with a reason, such as an unknown material, missing recipe or invalid quantity | Order owner in planning or master data | No automatic retry until the cause is fixed | Owner corrects the master data or order and re-releases it, keeping the same identity or cancelling and replacing it |
Two principles sit behind this table.
First, retries must be idempotent. A resend carries the same order identity, and the receiver treats a repeat as "already have it". Without that, a retry after a lost acknowledgement can turn into a duplicate order. The same thinking applies further down at machine level; see preventing duplicate commands between MES and PLC.
Second, retrying a rejection is pointless. A rejection means the receiver understood the order and refused it. Resending it unchanged only creates noise and hides the real cause.
Who should get the alert, and what should they do?
Route each alert to the person who can act on it, and give it to one named role rather than a shared inbox.
- Interface or IT support owns transport problems: lost messages, queues that are stuck, connections that are down. Their action is to restore the flow and confirm that pending orders have arrived.
- Planning or master data owns content problems: rejected orders, missing materials, wrong units. Their action is to correct the data and re-release the order.
- The production supervisor owns timing problems: orders that arrived but haven't started, or duplicates that reached the line. Their action is to decide what the line does now and record it.
Each alert should say which order, which state it is stuck in, how long it has been there, and the first action to take. An alert that just says "interface error" makes the recipient investigate before they can act.
Add escalation. If the first owner has not acknowledged the alert within an agreed time, it goes to the next role. If an order is close to its start time, the supervisor should hear about it whatever the cause.
Keep handoff alerts out of the operator alarm list unless the operator has to act. The problems are similar to those in why operators ignore alarms: too many alerts with no clear action train people to stop looking.
What does a worked example look like?
Illustrative example: all order numbers, times and windows below are invented to show the sequence.
- At 06:10, planning releases order 4471 in ERP for a 08:00 start on Line 2. ERP marks it as sent.
- The agreed rule is that receipt should be acknowledged within 15 minutes. By 06:25 no acknowledgement has arrived, so the order is flagged as unacknowledged and interface support gets an alert.
- Support finds that the receiving service restarted at 06:09 and the message was not stored. They resend order 4471 with the same identity. At 06:31 the MES acknowledges receipt, and because it holds no earlier copy, it accepts the order.
- Suppose the resend had also failed. The alert escalates to the planner at 06:45. At 07:15, still with no acknowledgement and within an agreed margin of the start time, the production supervisor is alerted as well.
- Under the fallback procedure, the supervisor confirms the order details with the planner, and the order is entered at the line terminal against the same ERP reference. The manual entry is flagged.
- When the interface recovers, the queued resend arrives. Because the identity already exists, the MES logs it as a duplicate and does not create a second order.
- After the shift, support records the cause and the planner confirms that ERP and MES show one order, one quantity and the correct status.
The key details are the shared identity, which makes both the resend and the manual entry safe, and the escalation timed against the start time rather than against the error itself.
How do network outages change the picture?
A short outage between the office and plant networks can look like many lost messages at once. If the sending side buffers orders and delivers them in sequence after reconnection, most of those "losses" turn out to be late messages instead. Design the buffer and the alerts together, so a recovering backlog does not set off a flood of alerts. See buffering plant data during network outages for the buffering side.
Group the alerts during an outage. One alert saying "link down, 12 orders pending" is more useful than 12 separate ones.
What mistakes should you avoid?
Alerting only on explicit errors. A lost message produces no error. Watch for acknowledgements that are missing.
Retrying without a stable identity. Every resend becomes a possible duplicate.
Retrying rejections automatically. Fix the cause, then re-release.
Sending every alert to everyone. Shared alerts get ignored. Name one owner per failure type.
Manual fallback with no reconciliation. Orders entered by hand must be matched back to ERP afterwards, or counts will drift. The same reconciliation habit helps when machine and production report counts disagree.
When does this approach not apply?
If orders reach the line on paper or through a planner reading ERP directly, there is no automated handoff to monitor. The controls there are procedural.
Where an order directly starts equipment, the handoff is part of machine control. Behaviour on communication loss belongs in the control design. Anything affecting safety should be reviewed with a qualified assessor, not handled through interface alerts.
Frequently asked questions
How long should we wait for an acknowledgement?
Base the window on the gap between release and planned start, and on how quickly someone can respond. It should be short enough to leave time for the human fallback before the line needs the order, and long enough that a normal delay doesn't set off alerts.
Should the ERP or the MES own the alert?
Whichever side can see the missing outcome. The sender knows what it sent and can detect a missing acknowledgement. The receiver knows what it rejected. In practice, a monitoring layer that tracks order states on both sides gives one consistent view and avoids two systems raising conflicting alerts.
Can we just rely on daily reconciliation?
Daily reconciliation catches mismatches, but usually after the line has already missed or repeated work. Keep it as a safety net that confirms ERP and shop-floor records agree. Use near-real-time alerts on acknowledgements and order states to catch failures while there is still time to act.
What should a manually entered order look like?
It should use the same order identity as ERP, carry a visible flag showing it was entered by hand, and record who entered it and why. That lets a late automated copy be recognised as a duplicate, and makes the order easy to find during reconciliation.