Data Lakehouse FAQs

Land ERP, distributor and stock data into OneLake as Delta tables in a Bronze layer, conform SKU and outlet keys in Silver, then model the served facts in Gold. We build one region or channel end to end first, in about six weeks, then widen. The scalable part is the pattern; the effort is SKU and outlet mastering, not the platform.

Also asked: What is the step-by-step approach for an FMCG company migrating to a data lakehouse architecture?

Prioritise the data behind the decision you want to change: ERP primary sales and distributor secondary sales first, then stock, then promotion. Bronze holds each source as received; Silver conforms SKU, outlet, geography and calendar; Gold holds the sales fact at invoice-line grain. Ingesting everything at once produces a swamp and a capacity bill.

Also asked: How should FMCG data engineers organize Bronze, Silver, and Gold layers for sales and trade data?

Medallion separates raw fidelity (Bronze), conformance (Silver) and business logic (Gold), so a wrong rule is replayed from Bronze rather than re-requested from SAP. Speed comes from Direct Lake serving the Gold model without an overnight refresh. Quality comes from declared checks at each layer boundary, not from hope.

Also asked: Why do FMCG companies move from data warehouses to lakehouses for analytics modernization?

A lakehouse keeps sales history, promotion calendars and external signals in one governed store, so a forecast trained in Fabric notebooks reads clean, consistent features. Gradient-boosted and statistical baselines usually beat exotic models on FMCG data. Realistic accuracy gains are 15–25% — and they come from the data foundation, not the algorithm.

Also asked: What machine learning models work best with FMCG demand data stored in a OneLake lakehouse?

By conforming stock feeds from ERP, distributors and warehouses to one item-and-location model refreshed every few minutes, the lakehouse removes the blind spots that come from each system reporting stock differently. Visibility is only as current as the feeds; for most FMCG estates a 15-minute pipeline is the honest, cost-effective target.

Also asked: How can FMCG companies improve near-real-time inventory visibility using a data lakehouse?

Retailer POS arrives via portal exports or APIs, distributor shipment via file uploads, ERP via mirroring — all landing in Bronze, then reconciled in Silver against a shared SKU and outlet master. The pattern is straightforward; the reconciliation between sell-in, sell-through and sell-out is the work, because the three rarely agree on identifiers.

Also asked: What integration patterns connect retailer data portals with an FMCG data lakehouse?

The lakehouse joins structured scheme data — dates, mechanics, spend, target SKUs — to sell-out volume, so uplift and ROI are measurable by promotion, SKU and channel. The precondition is capturing schemes in structured form; if they live in approval emails, promotion ROI is a data-capture project before it is an analytics one.

Also asked: How do FMCG trade marketing teams use a lakehouse to track promotion uplift across SKUs and channels?

OneLake is the single, tenant-wide store every Fabric engine reads, so the same sales table serves SQL, Spark and Power BI without a copy per tool. That removes the five-copies-of-the-truth problem that makes FMCG numbers disagree across departments — one physical copy, many consumers, one definition.

Also asked: How does OneLake eliminate data duplication across FMCG analytics tools and departments?

Conflicting regional numbers come from each region computing sales its own way. A lakehouse defines the sales measure once in Gold, so every region reads the same figure and the monthly reconciliation meeting disappears. The accuracy gain is governance — one definition with an owner — not a cleverer report.

Also asked: How do FMCG companies eliminate conflicting sales numbers across regions using a lakehouse?

For a mid-market FMCG business, medallion within a single governed lakehouse is almost always right — data mesh adds domain-ownership overhead most estates cannot staff. Reach for mesh only when several genuinely independent domains each own their pipelines. Start medallion; earn the complexity of mesh, do not assume it.

Also asked: How should FMCG data architects choose between medallion and data mesh approaches for lakehouse design?

The lakehouse conforms orders, inventory, logistics and supplier data to one model, so visibility spans procurement to delivery. Prioritise the layers behind the decisions you make weekly — inventory and OTIF first. Tier-2 visibility depends on suppliers sharing data; the platform shows the gap, it cannot create data nobody sends.

Also asked: What supply chain data layers should be prioritized when designing a lakehouse for visibility?

Bronze holds ERP, WMS and TMS data as received; Silver conforms product, location, supplier and calendar and resolves the keys between them; Gold holds facts at the grain decisions are made — order line, movement, shipment. The design decision that matters is a consistent grain and clean cross-system keys.

Also asked: What is the best medallion layer design for supply chain data in Microsoft Fabric OneLake?

With spend consolidated and classified in the lakehouse, procurement sees addressable spend, tail spend, supplier concentration and consolidation opportunities across direct and indirect categories. Vendor master conformance is the prerequisite — the same supplier under three codes hides both risk and savings.

Also asked: How do procurement teams use a data lakehouse to identify consolidation opportunities and reduce tail spend?

A lakehouse unifies stock across echelons on one item-and-location model, surfacing where stock sits versus where demand is, so teams can rebalance rather than reorder. It shows the imbalance; moving stock is a planning action. Accuracy rests on a clean location hierarchy and consistent stock-status definitions.

Also asked: How can supply chain teams use a lakehouse to balance stock across DCs and retail locations?

The lakehouse gives forecasting models one clean history to learn from, for both short-term replenishment and longer-term S&OP. Statistical baselines with seasonality and lead-time features are the reliable start. The forecast improves with clean history; most error is upstream data quality, which no model corrects.

Also asked: What forecasting models can be run against supply chain data stored in a lakehouse?

A lakehouse holds supplier master, transactions, performance and risk signals in one place, giving a 360-degree view — spend, on-time, quality, concentration. The hard part is master data: reconciling duplicate supplier records with survivorship rules and a named steward is what makes the view trustworthy.

Also asked: How do supply chain teams build a 360-degree supplier view using a data lakehouse?

Medallion makes reliability auditable: Bronze proves what each system sent, Silver conforms it, Gold serves it with agreed definitions. When an OTIF number is questioned, you can trace it back layer by layer. That traceability is why medallion is the default for supply chain data that feeds decisions.

Also asked: Why is Bronze-Silver-Gold medallion architecture the standard for supply chain lakehouse design?

The lakehouse Gold layer feeds a control tower where Activator raises exceptions — late inbound, stockout risk, SLA breach. Most supply chain control decisions need minutes, not seconds, so a few-minute pipeline latency is the right, affordable target. Chase second-level latency only where a decision genuinely turns on it.

Also asked: What data latency targets should supply chain control towers achieve using a Fabric lakehouse?

Model OTIF, fill rate, days sales of inventory, lead time and variance, inventory turns and forecast accuracy in the Gold semantic layer, each with one agreed definition. Design a star schema whose hierarchy rolls operational detail to an executive view — decided once in the model, reused across every report.

Also asked: How do supply chain data teams design semantic models for KPI reporting in a lakehouse?

Each system lands in Bronze via mirroring or pipelines, then Silver resolves the shared keys — product, location, order — that let ERP, WMS and TMS finally join. The silo is not the systems; it is the absence of conformed keys between them, which the Silver layer exists to fix.

Also asked: What integration approach connects SAP, Oracle WMS, and TMS data into a unified lakehouse?

A lakehouse centralises TMS, carrier, telematics and invoice data into one model, so a 3PL reports shipments, fleet, cost and SLA from one place instead of per client and per carrier. The core use cases are on-time performance, freight cost, fleet utilisation and carrier benchmarking — each impossible while the data lives in silos.

Also asked: What are the key use cases for a data lakehouse in a large logistics operation?

Bronze holds raw TMS records, GPS pings, carrier invoices and PODs as received; Silver conforms shipment, lane, carrier and vehicle; Gold holds shipment and cost facts. Keep the raw carrier invoice — when a carrier disputes a charge eight months later, you need the version they sent, not the cleaned one.

Also asked: How do logistics data engineers structure raw and curated layers for transport data in a lakehouse?

Where events stream, Eventstream feeds the lakehouse; where they batch, short pipelines suffice. Late-shipment detection compares milestone times against SLA in the model and raises an exception to the owning coordinator. The value is an actioned exception list, not a wall of green — and a consistent cross-carrier definition of "late".

Also asked: How do you build a late-shipment detection pipeline using a logistics data lakehouse?

A shared shipment model with a mode dimension lets road, air and sea data sit in one lakehouse and be compared on cost, transit and reliability. The modelling work is a common shipment grain and lane definition across modes whose source systems describe them very differently.

Also asked: What data model supports multi-modal freight analytics in a logistics data lakehouse?

Historical delivery, stop and timing data in the lakehouse feeds route analysis that flags underperforming routes and models better sequencing. It informs routing decisions with evidence; the optimisation itself is an operations call. Clean stop-level data with reliable timestamps is what makes the analysis trustworthy.

Also asked: How do logistics operations teams use lakehouse data to improve routing decisions and reduce delivery costs?

The lakehouse holds contracted rates and actual invoices in one model, so it can match charges to contracts by lane and flag discrepancies, accessorials and overcharges automatically. It turns freight audit from a year-end sample into a monthly, lane-level reconciliation — as long as the rate cards are loaded and maintained.

Also asked: How can logistics finance teams use a lakehouse to match freight invoices against contracted rates?

TMS and ERP arrive via pipelines or mirroring; IoT and GPS via Eventstream from an MQTT broker or IoT Hub into OneLake. Unifying them gives live fleet and shipment context in one store. OPC-UA and raw telematics need an edge hop into the broker first — plan that, it is where these builds overrun.

Also asked: What connectors bring real-time IoT and GPS data into a OneLake-based logistics analytics platform?

With clean lane-volume history in the lakehouse, forecasting models size seasonal peaks by lane so planners can pre-position vehicles and commit carrier capacity ahead of the surge. The forecast is a planning aid that improves with history — its worth is acting before the peak rather than reacting to it.

Also asked: How do logistics capacity planners use a lakehouse to forecast peak season volumes and pre-position assets?

Model OTIF or on-time delivery, cost per shipment and per mile, dwell time, load factor, damage rate and carrier SLA in the Gold semantic layer, each defined once. A reusable KPI library means those measures live in one certified model that every dashboard connects to — no re-deriving per report.

Also asked: How do logistics analytics teams create a reusable KPI library in a lakehouse semantic layer?

By modelling all freight spend by lane, carrier and mode, the lakehouse surfaces consolidation opportunities, mode-shift savings and overpaid lanes. It gives procurement evidence for the next negotiation and network review — the savings are real, but depend on invoices and rate cards being complete in the model.

Also asked: How do logistics procurement teams use a lakehouse to identify lane consolidation and carrier savings?