Modern Data Strategy for Industrial Operations

Your finance team closes the month with one OTIF number. Your operations director has a different one. The plant floor's shift supervisor has a third — pulled from a spreadsheet that nobody officially owns. Same period. Same business. Three answers.

That is not a reporting problem. It is a strategy problem — and no dashboard fixes it.

Mid-market manufacturers and FMCG producers are not short of data. SAP S/4HANA is logging every goods movement. SAP ByDesign is tracking purchase orders and supplier confirmations. Microsoft Dynamics 365 is capturing customer orders and AR ageing. The MES is writing machine state every thirty seconds. The WMS is timestamping every pick.

The data exists. The problem is that none of these systems agree on what a "confirmed order" means, when a "shipment" counts as dispatched, or what the denominator for OTIF actually is. Each system applies its own logic. Each team builds its own extract. By the time the numbers reach a leadership meeting, you are reconciling Excel files instead of making decisions.

This is the core failure mode — and it has nothing to do with which BI tool you purchased.

The root cause is structural: no integration layer, no governed definitions, no single foundation that every downstream report draws from. Until that exists, analytics is just decorated confusion.

The phrase gets used loosely. The sequence is not.

**Unify the data first.** Build an integration layer that pulls SAP S/4HANA production orders, SAP ByD purchase confirmations, Dynamics 365 sales orders, and MES output data into a single governed store — OneLake on Microsoft Fabric, structured as Delta Lake. Not a copy of everything. A curated, versioned, agreed-upon set of the facts that finance, operations and the plant floor all read from the same place.

Delta Lake gives you ACID transactions on that data. Time-travel lets you reconstruct what the numbers showed at month-end close, not just what they show today. That matters enormously when a CFO asks why last month's gross margin report contradicts this month's revised actuals.

**Then predict.** Once you have a clean, unified foundation, AI forecasting becomes viable. Demand forecasting on SAP ByD order history. OEE prediction from MES time-series data. Inventory DIO modelling across plant and distribution centre. None of this works reliably if the underlying data is still dirty — the model will just learn your noise and call it a trend.

**Then act.** Automation and Power BI operational dashboards sit at this layer. An OTIF alert that fires before a shipment misses its window — not after. A production schedule adjustment triggered by a supplier confirmation delay in SAP S/4HANA. Action workflows built in Power Automate that fire from governed, trusted data — not from someone's inbox.

The businesses that skip Unify and go straight to Act end up with impressive-looking dashboards built on numbers nobody trusts. Two years later, they are rebuilding.

This is the stack that MyData Insights works with on industrial data programmes:

**Source layer:** SAP S/4HANA, SAP ByDesign, Microsoft Dynamics 365, SCADA / OPC-UA feeds from the plant floor, WMS export files — ingested via Azure Data Factory pipelines into OneLake.

**Storage and governance layer:** OneLake as the single logical data lake, Delta Lake as the open table format, Microsoft Fabric as the orchestration and compute environment. Microsoft Purview for data cataloguing and lineage — so when a manager asks "where does this OTIF number come from?" the answer is a documented lineage graph, not a guess.

**Semantic layer:** A properly built Power BI dataset with defined measures — not a collection of ad-hoc calculated columns that each report author interprets differently. One definition of OTIF. One definition of forecast accuracy. One definition of DIO. Enforced at the model level.

**Reporting layer:** Power BI Direct Lake for live operational dashboards — no import lag, no stale data for shift supervisors making real-time decisions.

The semantic layer is the piece most organisations underinvest in. They build the pipes, load the data, and then let each report author define their own measures. Six months later, the CFO and the COO are in a meeting with two different gross margin figures and no way to reconcile them programmatically.

A data strategy that does not name the metrics it is supposed to shift is not a strategy — it is an architecture exercise.

For manufacturing and FMCG clients, the metrics that matter at the leadership level are: OTIF (on-time in-full), OEE (overall equipment effectiveness), forecast accuracy (demand vs actual, by SKU family), days inventory outstanding (DIO), and scrap rate by production line.

Each of these needs a governed, agreed definition in the semantic layer before you can track whether it is moving. OTIF especially — because every business calculates it slightly differently, and that difference is almost always the source of the finance-vs-operations disagreement.

Once the definitions are standardised in the Microsoft Fabric / Power BI semantic model, the dashboard becomes a governance artefact as much as a reporting tool. The number shown is the number everyone agreed on. That changes the nature of leadership conversations — from "which number is right?" to "what are we doing about it?"

A mid-market FMCG business running SAP ByDesign for procurement and a standalone WMS for warehouse operations had three separate OTIF calculations in circulation — one from the SAP ByD team, one from the logistics coordinator's Excel file, and one from the customer-facing PowerPoint the sales director sent every month.

The Unify phase took six weeks — ADF pipelines from SAP ByD and WMS into OneLake, Delta Lake tables structured by domain (orders, fulfilment, inventory), and a single Power BI semantic model with one defined OTIF measure. By week six, the monthly leadership review ran off one report. The three-way reconciliation that had consumed three hours of a senior analyst's time every month-end stopped.

Forecast accuracy and DIO instrumentation followed in the next phase. No invented numbers — the improvement ranges we see in these programmes typically run 15–30% reduction in manual reconciliation effort, and the data foundation then supports predictive work that was not viable before.

If your ERP is not the system of record for your operational data — if the plant runs primarily on spreadsheets and manual entry with no MES, no SCADA, no structured digital capture at the process level — the Unify phase will be significantly harder and longer. This programme is most effective where source systems already exist and the problem is integration and governance, not digitisation from scratch.

Similarly, if your organisation does not have an internal data owner who can define and agree on metric definitions with finance and operations, the semantic layer will stall in committee. Technical architecture alone cannot resolve a political disagreement about what OTIF means.

In the Discover phase — typically two weeks — we map your source systems (SAP S/4HANA, SAP ByD, Dynamics 365, or others), identify the one or two metrics where the finance-vs-operations disagreement is loudest, and agree a governed definition.

In the Prototype phase — weeks three to six — we build the ADF pipelines, load into OneLake as Delta Lake, and deliver one Power BI report with one agreed metric: OTIF, OEE, or forecast accuracy — your choice. That is the first artefact every stakeholder reads from the same source.

Expansion follows — but you have seen it work before you commit to it.