The logical layer that lets a half-migrated roll-up still answer one question consistently — model once, federate the data, generate insights anyway.
Arvind can't wait for every mill and export desk to migrate to SAP before it gets answers. The fix isn't one warehouse — it's a shared ontology (so everyone means the same thing) over a data mesh (each division owns its data as a product), with a semantic layer that federates them. Insights generate today; they just carry a confidence flag where a division isn't on SAP yet.
Ten classes everything maps to. The Plant is the keystone: it's where division, leader, entity and geography reconcile.
70% of revenue is already plant-grain actual; the rest is read in place from legacy mill/export systems and reconciled — no big-bang migration required.
Every metric has one definition and a grain. The layer federates it across on-SAP and legacy domains, flagging where a value is allocated.
| Metric | Definition | Grain | How it federates across divisions |
|---|---|---|---|
| Revenue | Σ recognized revenue | plant · order | actuals where on SAP; allocated from area where not |
| Adjusted EBITDA | revenue − COGS − SG&A (+ add-backs) | division · entity | entity P&L normalized to one chart of accounts |
| Value-added & AMD revenue | annuity-like program revenue | program | from SAP SD / MES across all divisions |
| Value-added mix | value-added ÷ revenue | division | federated — same formula, many sources |
| DSO | AR ÷ revenue × 365 | entity · plant | legacy/export entities measured at area grain, flagged |
| Gross margin | (revenue − COGS) ÷ revenue | order · division | mapped via canonical cost categories |
| Repeat-order rate | expansion − attrition on base | brand customer | resolved across duplicate customer records |
Entity resolution matches legacy mill / division / plant codes to one canonical node — so the AMD data lines up with everything else.
Query reads each division's data product in place; the semantic layer maps native SAP/MES fields to canonical metrics.
Where a division reports at area level, allocation disaggregates to plant on learned drivers and marks it an estimate with a confidence band.
Allocated parts must tie back to the source total; anomalies and duplicate brand-customers/suppliers across divisions are surfaced.
This is not theoretical — it's how this cockpit already works. The Story, Briefing and 360 views read the same governed metrics over on-SAP and legacy divisions alike; 70% of the numbers are plant-grain actuals and the balance is SAP-allocated and labelled. As each division migrates to SAP, its data product's grain rises and estimates flip to actuals — the mesh closes itself.