Stop Running Six Azure Services.
Build One Platform
That Does All of It.

OneLake is the single, tenant-wide data lake built into Microsoft Fabric — one Delta Parquet storage layer every workload reads from, so you stop copying data between Azure Data Lake Storage, Synapse, and separate Power BI datasets. One copy of the data serves engineering, analytics, and AI, which removes the duplicate pipelines most industrial estates run today. For a manufacturer, ERP, MES, and quality data sit in one governed lake instead of four.

Mirroring continuously replicates an operational database into OneLake as Delta tables, near real time, without building extract pipelines. For SAP S/4HANA or Dynamics 365, you get a live analytical copy in Fabric while the ERP keeps running the transactions — no overnight batch, no load on the production system. Source data lands in OneLake ready for Power BI in minutes rather than the next morning.

Medallion is a three-layer pattern: bronze holds raw source data as ingested, silver holds cleaned and conformed tables, gold holds the business-ready model the dashboards read. In Fabric each layer lives in OneLake as Delta tables, so a plant's raw SCADA feed (bronze) becomes validated downtime records (silver) and then a governed OEE model (gold). It keeps lineage auditable — you can trace any number on a dashboard back to its source.

Direct Lake reads Delta tables in OneLake directly with no data copy, so a Power BI report reflects new data as it lands and there are no refresh windows to schedule or fail. Import mode loads a separate copy into the Power BI model, so the report is only as fresh as the last scheduled refresh and large models can be slow to refresh or time out. Direct Lake requires the data in OneLake Delta format; Import works with any supported source. For live operational reporting on Fabric, Direct Lake is the default; Import suits smaller models or sources not yet in OneLake.

Eventstream ingests high-velocity event data — IoT sensors, SCADA tags, MQTT/OPC-UA feeds — into Fabric in real time and routes it to a lakehouse or KQL database. Use it when you need to act on plant-floor signals as they happen: line-speed drops, temperature excursions, downtime events. It is the ingestion layer behind a real-time OEE or condition-monitoring dashboard.

Fabric folds the Synapse engines — data warehouse, Spark, and pipelines — into one SaaS platform on OneLake, so you no longer provision and tune separate Synapse resources. The warehouse, lakehouse, and Power BI all read the same Delta tables, removing the data movement Synapse required between stages. For most mid-market estates it means one capacity to manage instead of a Synapse workspace plus a separate Power BI Premium capacity.

A Fabric Data Agent is a natural-language layer that answers questions over your governed OneLake data using Azure OpenAI, grounded in the gold-layer model. Once your operational data is modelled, an operations manager can ask which lines missed target last shift and get an answer from live data rather than waiting for a report. It is deployed on top of the semantic model, so answers respect the same definitions as the dashboards.

Land each source into OneLake — ERP via Mirroring or Azure Data Factory, MES via its database or API, IoT via Eventstream — then conform them in the silver layer to a shared plant model. Fabric holds all three as Delta tables, so production orders (ERP), actual output (MES), and machine signals (IoT) join in one place. That join is what turns three disconnected systems into a single source of truth for the plant.

It starts with one governed data product — typically OEE or production reporting — built on OneLake with the medallion architecture, live in Power BI by week 6 and in production by week 8. Sources are connected in priority order: ERP first, then MES and quality, then IoT for real-time. The first release proves the foundation; later phases add forecasting and automated alerting.

Direct Lake and Mirroring let Power BI read live Delta tables in OneLake, so reports reflect the current shift instead of last night's batch. The overnight ETL window disappears because data is replicated continuously rather than loaded once a day. A shift handover then runs on numbers that are minutes old, not twelve hours old.

Stream machine signals through Eventstream into OneLake, model availability, performance, and quality in the gold layer, and read it in Power BI with Direct Lake. The dashboard updates as the line runs, so a drop in availability shows within minutes, not at the next morning's report. We build a working OEE data product inside the 6-week first-value window.

First value lands in 6 weeks — a governed data product live in Power BI — with full production reached at 8 weeks. A 2-3 week Fabric Starter Sprint is the faster entry point if you want one dashboard proven before committing to the full build. Real-time and predictive workloads are added in a later phase on top of that foundation.

Fabric ingests OPC-UA and MQTT feeds through Eventstream and stores them in a KQL database or lakehouse for sub-second query. SCADA historian data can be batched in via Azure Data Factory where real-time is not required. Both land in OneLake alongside ERP and MES data, so machine signals sit in the same governed model as the rest of the plant.

Fabric brings sales, shipments, and inventory into one OneLake model, then runs machine-learning forecasts natively in the same platform — no export to a separate tool. Forecasting on conformed, current data typically reduces forecast error by 15-25% across SKU and regional planning cycles. The forecast feeds straight back into Power BI and replenishment, so planners act on it rather than re-keying it.

Eventstream ingests telematics, WMS, and TMS events into OneLake, and Direct Lake powers a Power BI control tower that updates as shipments move. Dispatchers see OTIF risk and dwell time live instead of in a next-day report. Fabric Activator can then raise an alert the moment a shipment breaches its window.

Land ERP, WMS, and order data in OneLake, model OTIF, fill rate, and inventory cover in the gold layer, and build the dashboard in Power BI on Direct Lake. Because every metric reads one governed model, the same OTIF number appears everywhere it is used. We deliver a first working supply chain dashboard inside the 6-week first-value window.

Eventstream ingests sensor data, Fabric's native machine learning trains failure-prediction models on the equipment history in OneLake, and Activator triggers a work order when risk crosses a threshold. The whole loop — signal, prediction, action — runs in one platform, so there is no hand-off between an analytics tool and a separate alerting system. It moves maintenance from fixed schedules to condition-based intervention.

Azure OpenAI is built into Fabric and runs against your governed OneLake model, so answers and Copilot results are grounded in your operational data, not generic text. A manager can query production or supply data in natural language with the same definitions the dashboards use. Because it stays inside the Fabric tenant, the data does not leave your governance boundary.

Activator watches your Fabric data and fires an action when a condition is met — an email, a Teams alert, a Power Automate flow, or a logged event. For operations, a stockout risk or an OEE drop triggers a response automatically instead of waiting for someone to spot it on a dashboard. It is how analytics turns into action without a person watching the screen.

Fabric runs Spark ML and Azure OpenAI inside the platform, trained and scored against Delta tables in OneLake, so models read the same governed data as the reports. A demand forecast or a defect-prediction model can be built, scheduled, and surfaced in Power BI without moving data to a separate ML stack. Keeping the model next to the data is what makes the output repeatable and trustworthy.

Train a forecasting model on conformed sales and shipment history in OneLake, score it on the schedule that matches your planning cycle, and surface results in Power BI. Moving from spreadsheet averages to machine learning on current data typically cuts forecast error by 15-25%. The gain comes as much from one clean, current dataset as from the algorithm.

MyData Insights builds the full Foundation implementation for USD 8,000 — first value in 6 weeks, production at 8 — covering OneLake, the medallion architecture, a governed model, and core dashboards. A 2-3 week Fabric Starter Sprint that proves one dashboard starts at USD 1,500. Microsoft Fabric capacity is separate and starts at an F2 SKU (about USD 262/month pay-as-you-go); most plants begin on a small capacity and scale up.

F32 provides 32 capacity units at about USD 4,205/month pay-as-you-go; F64 provides 64 capacity units at about USD 8,410/month. The deciding difference is licensing: on F32 every Power BI viewer needs a Pro licence, while F64 is the threshold at which free-licensed users with the Viewer role can consume content. Once you have many report consumers, F64 is often cheaper than F32 plus per-user Pro licences. Reserved one-year capacity is about 40% below these pay-as-you-go rates.

8 weeks to production, with the first governed data product live in Power BI at 6 weeks. The 2-3 week Starter Sprint is available if you want a single dashboard proven first. Real-time ingestion and predictive models are added in a later phase, not the initial 8 weeks.

The Foundation tier is the data platform baseline — OneLake, the medallion architecture, a governed semantic model, and core Power BI dashboards across your ERP/SAP plus one operational source. First value lands in 6 weeks, production at 8, for USD 8,000. The Operational Analytics tier builds on Foundation, adding real-time ingestion (Fabric Eventstream, IoT/SCADA/MES), predictive models, and automated alerting with Fabric Activator across multiple live sources — typically a further 4-6 weeks, scoped per requirement. Start with Foundation to get one trustworthy source of truth live, then add Operational Analytics when you need real-time and prediction on top.