A planning application such as Kinaxis, o9, Blue Yonder, SAP IBP or OMP is bought for its planning engine. The project then spends most of its time somewhere else: getting items, bills of material, stock, open orders and forecasts out of the ERP and into the shape the engine expects, on the day the planners need them. In Lima's delivery programs since 2020 that data work has been about 70% of the effort. This guide names the seven data problems that cause it, shows each with an example, and gives the order in which to fix them.
A planning engine does not query the ERP. It loads a complete, consistent dataset in its own model inside a time window, runs its calculations, and produces a plan. Three things follow. The dataset has to be complete: an item without a plant cannot be planned. It has to be consistent: a bill of material whose item has not arrived is rejected. And it has to arrive on time: a load that misses the window means a planning run on yesterday's data.
The ERP holds most of the raw material, in its own shapes. The rest, such as hierarchies, planning parameters and sourcing rules, often lives in spreadsheets. Turning all of that into the engine's model, every cycle, is the work.
| Problem | What it looks like | Example from a first load |
|---|---|---|
| 1 · Many ways of supply, one item | The ERP holds production versions, info records and source lists; the engine wants one source per way of supply, each with its own bill of material. | One material with three production versions and two suppliers becomes five PartSources in Kinaxis; get the mapping wrong and the plan uses the wrong bill. |
| 2 · Referential order | The engine rejects a child whose parent has not arrived. | 58,900 PartSources and their bills of material must load after the parts they belong to; a load in the wrong order fails at row one. |
| 3 · Hierarchies from several ERPs | The same code means different things in each ERP; members are missing parents. | 2,140 orphan members found in a product hierarchy assembled from two ERPs before the first o9 load. |
| 4 · Open quantities that are not open | Delivered and closed order lines still carry quantities. | Order lines with delivered quantity equal to ordered quantity have to be excluded or the engine plans for demand already met. |
| 5 · Units, currencies and calendars | Components in units with no conversion; due dates on non-working days. | A component in litres against a bill in kilograms with no density on file; a due date on a plant holiday. |
| 6 · Planning master data outside the ERP | Parameters, families and roll-ups live in spreadsheets with no owner or version. | Safety-stock policies in a workbook that three planners edit; the engine receives a different value each cycle. |
| 7 · No report at the end | A load finishes and nobody can say how many rows were dropped or why. | 3 purchase-order lines held for a missing supplier; without a report they simply vanish from the plan. |
| Check | Kind |
|---|---|
| Every item has a plant and a unit of measure | completeness |
| Every child finds its parent | referential order |
| Each way of supply has its own structure | consistency |
| Open quantities are open | consistency |
| Dates are inside the horizon and the calendar | range |
| Units and currencies convert | consistency |
| Hierarchies are strict | completeness |
| Weekly equals the sum of daily | consistency |
| Counts and control totals match | reconciliation |
A row that fails a check becomes an exception record with an owner and a deadline; the load continues with the valid rows and the report says what was held. How exceptions work →
It is Lima's field figure from its own delivery programs since 2020 and is labelled as such. Your own programs may differ; the seven problems are the same.
The first: many ways of supply for one item. It is where ERP structures and planning structures differ most, and where a wrong mapping is hardest to see afterwards.
Vendors' connectors load their model; the work in this guide is producing the model from the ERP and keeping it correct every cycle. The checks and the report are the part most programs do by hand.
A first planning dataset in four weeks at a fixed price, with the nine checks and the reconciliation report.
