GUIDE · RECONCILIATION
Why shift-log and ERP totals disagree,
and how to reconcile them.
Published · ProdCrest
Short answer. The shift log and the ERP record the same output at different moments, by different people, for different purposes, so some disagreement is normal. Reconcile per work order, not per month. Agree a join key, a tolerance and a period-close rule, label every difference with a reason, and report the total of the work orders that agree, together with the list of those that do not.
Two records of the same output
The shift log is written on the floor, during or just after the shift, usually by the operator or shift lead. The ERP figure is confirmed later, often by someone who books the work order complete. Neither is automatically right. The shift log is closer to what happened; the ERP is closer to what the business invoices and values. When they differ, the useful question is not which one to trust, but why this work order differs.
A worked example on synthetic data
The figures below come from the ProdCrest demo, which uses synthetic data for an invented moulding plant over four weeks. They show how the totals move, not what any real plant should expect.
| Figure | Pieces | What it is |
|---|---|---|
| Spreadsheet, before checks | 2,819,124 | The sum of every shift-log row as typed |
| Spreadsheet, after checks | 2,920,686 | Duplicates removed, shot counts converted to pieces |
| ERP confirmed | 2,949,353 | Good output booked against the 112 work orders |
| Trusted figure | 2,863,539 | ERP total of the 106 work orders where both sources agree within 2% after checks |
The checks moved the spreadsheet total closer to the ERP, but the two still differ. The six work orders that remain apart are explained by shifts that were never recorded in the spreadsheet; their output exists only in the ERP. That is a recording gap, not a calculation error, and it belongs on the exception list rather than inside a corrected total.
The method, step by step
1. Choose the join key
Work order is the best key when both systems carry it. If the shift log does not, date and machine can stand in, provided the plant runs one work order per machine per day. Rows that match no work order are listed separately, never dropped.
2. Clean both sides first
Run the record checks before comparing: dates, product identity, duplicates, missing shifts, units and plausibility. The seven-check guide describes each one. Comparing uncleaned records mostly measures typing habits.
3. Agree a tolerance
Decide in advance how close is close enough, and write it down. The demo uses 2% per work order. A tolerance agreed after seeing the numbers tends to drift towards whatever makes the report look finished.
4. Set a period-close rule
Entries booked after the period closed change last month’s figures after they were reported. Flag them rather than rewriting history. In the synthetic data, 2 work orders were entered after the period close.
5. Label every difference
Give each differing work order the reason that explains it: duplicate, missing shift, shot count, late entry, or unexplained. In the synthetic example, 17 of 112 work orders differed before checks: 7 shot counts, 6 missing shifts and 4 duplicates. None was left unexplained, which in a real plant would be unusual and worth questioning.
6. Decide which number goes to the meeting
Report the trusted figure, the number of work orders behind it and the list of work orders that do not agree. A single “corrected” total hides the decisions that produced it. A figure with its exception list shows what is known and what is still open.
Common reasons for a gap
- Pieces versus shots or cycles. Multi-cavity tools make several parts per cycle; a counter reading entered as pieces understates output.
- Good versus total. One record counts everything made, the other only what passed inspection.
- Timing. A shift that crosses midnight, or a confirmation booked days later.
- Rework. Parts reworked and booked again can appear twice in one system and once in the other.
- Scrap units. Scrap weighed in kilograms and converted with an outdated part weight.
Limits
Reconciliation shows where two records disagree and which known pattern explains it. It does not prove that the records that agree are complete, and it does not find the cause of a loss. Those answers come from the floor, the counters and the people involved.
The interactive demo shows the reconciled figures in a dashboard. If your own totals disagree, a production data review runs this method on your exports inside your environment. Prepare a short brief.