From ERP data to commercial intelligence — warehouse, semantic layer and a single version of truth

The observation. Most industrial mid-caps already hold, inside their ERP, almost everything a good commercial decision would need : every order, every invoice line, every stock movement is recorded. And yet the same question keeps coming back in meetings : which products are accelerating, which are slowing, and how many weeks of cover do we have left ? The information exists ; it is simply not made legible. Turning transactional data into commercial intelligence is not a tooling project : it is a matter of modelling, governance and discipline.

Why the ERP alone is not enough

An ERP is built to execute transactions, not to analyse them. Its schema is normalised for write integrity : avoiding duplicates, ensuring an order does not contradict itself. What is excellent for data entry becomes hostile to analytical reading : a question as simple as “what is this product's velocity by market over twelve months?” can require joining five to ten tables, with a real risk of double-counting. Bill Inmon framed the founding distinction back in the 1990s : transactional systems (OLTP) and analytical systems (OLAP) have neither the same shape nor the same purpose.

The warehouse and semantic-layer pattern

The classic — and still sound — answer is to separate the two worlds. Data is extracted from the ERP into a data warehouse (or a modern lakehouse), modelled for analysis, then exposed through a semantic layer — a shared dictionary in which “net revenue”, “margin” or “velocity” have a single definition, computed once, for everyone.

• Dimensional modelling — Ralph Kimball's method organises data into fact tables (orders, invoice lines) and dimension tables (product, customer, market, time). An analytical question then becomes a combination of axes rather than a fragile query.
• Semantic layer — metrics are defined once, upstream of the dashboards, so that one word never carries two different calculations depending on who says it.
• Per-manager dashboards — each market manager sees only their scope, with the same definitions as their peers : comparable, not siloed.

Velocity and cover, two sober metrics

Two measures are often enough to inform a supply decision. Velocity is the selling pace of a SKU — how many units per week, smoothed to absorb noise. Cover translates available stock into weeks of sales at the current velocity. A high-velocity, low-cover product is a stock-out in the making ; a low-velocity, high-cover product ties up cash. Neither can be read off the raw order book ; both emerge as soon as the data is properly modelled.

Governed data, or the single version of truth

The real obstacle is almost never technical. It is governance : who owns a metric's definition, who signs off on a dimension's quality, who decides when two numbers disagree. Without that discipline, every team rebuilds its own spreadsheet and you fall back into the well-documented syndrome of multiple versions of the truth. A single version of truth cannot be decreed : it is built through a named data owner, written definitions, and a read-only principle on source systems — you read the ERP, you do not write back to it to analyse.

Where we stand

Montandor Andorra is a young house, and we build our commercial dashboards directly on the live data of our ERP — read-only, with shared definitions and an identified owner for each metric. We do not chase the most spectacular tool ; we want a market manager and the leadership to read the same number, understand it the same way, and decide faster. It is less a matter of technology than of rigour.

“Data has value only if it leads to a better decision, sooner. A dashboard that impresses but that no one uses to decide is a cost, not an asset. Our requirement is simple : one definition per metric, legible to the person acting in the market and to the one arbitrating.”
— Wouter Meijboom, CEO, Montandor Andorra.

Sources

• Ralph Kimball & Margy Ross — The Data Warehouse Toolkit, 3rd edition (Wiley) : the reference on dimensional modelling (facts and dimensions).
• William H. Inmon — Building the Data Warehouse : the founding OLTP / OLAP distinction.
• DAMA International — DMBOK (Data Management Body of Knowledge) : framework for data governance and quality.
• McKinsey & Company — work on the data-driven enterprise and the value of data-based decision-making, 2018-2023.
• Harvard Business Review — A. McAfee & E. Brynjolfsson, Big Data: The Management Revolution, 2012.
• GS1 — product identification standards (GTIN) and product-data quality.

Research led by Hélène Vincent (Growth & Analytics Lead), in collaboration with Lev Marchuk (Data Scientist) and Chloé Garnier (Head of Architecture).