Skip to main content
Databricks semantic layer guide

Manufacturing semantic layer on Azure Databricks

Create one governed business language across production, inventory, quality, delivery and margin so Power BI, SQL and AI use the same grain, dimensions and KPI definitions.

Read the semantic guide
Business grain Metric Views Governed definitions
Manufacturing semantic model
Governed

Shared manufacturing context

Site
Product
Time
Customer

One business language

Keep facts separate by grain, then standardize the business meaning shared across them.

Production
OEE · yield
Inventory
stock · shelf life
Delivery
OTIF · fill rate
Power BI
SQL
Analytics
Ask Titan

Different interfaces can present the answer differently. The governed business meaning should stay the same.

Semantic answer alignment
Governed definition

Same question

How much stock is available for production?

WMS report

Physical stock before exclusions

5,440 kg

Spreadsheet

Manual exclusions for blocked and quarantine stock

4,980 kg

Power BI

Report logic with its own filter interpretation

4,820 kg

Apply shared business definition

Governed answer

4,820 kg available

exclude quarantine exclude blocked exclude expired

Semantic controls

  • grain: site × location × batch × SKU
  • unit: base UOM
  • owner: supply chain

Illustrative values. The goal is not one dashboard; it is one governed definition that every consumer can reuse.

Why semantics

The same business question can still produce different answers

Reports can use the same source data and still disagree when grain, exclusions, quality status, units or effective dates are defined independently in every consumer.

Grain determines what one row or measure actually represents.

Exclusions decide which states, lots or transactions count.

Units and time must be consistent before aggregation.

Ownership makes the definition reviewable and auditable.

Architecture boundary

A semantic layer defines business meaning. It does not fix the underlying data.

Identity, history, units and business process grain need to be reliable before data reaches the semantic layer. The semantic layer then gives teams a consistent way to interpret that data and calculate shared business metrics.

Governed data products

Reliable business data

Identity stable business keys
Grain a clear meaning for each row
History effective time and event time
Conformance units, status and mappings

Semantic layer

Shared business meaning

Business terms agreed definitions
Measures reusable KPI logic
Relationships defined ways to slice the data
Ownership accountable business owners
Manufacturing grain map

Separate facts by business process; reuse dimensions across them.

Process
Natural grain
Typical measures
Production
run × line
OEE · yield · throughput
Inventory
batch × location
available stock · shelf life
Quality
sample × test
pass rate · holds · defects
Delivery
order line × shipment
OTIF · fill rate
Margin
order line × product
revenue · cost · margin

Manufacturing context

Define the grain before you define the metric

Manufacturing systems describe different business processes at different grains. The semantic model should make those grains explicit before facts are connected through shared dimensions.

Production run is not the same grain as a production order or machine event.

Inventory batch is not the same grain as an inventory movement.

Quality sample is not the same grain as a batch release decision.

Shipment is not the same grain as a customer order line.

Business model

Reuse dimensions without mixing business grains

A good manufacturing semantic model does not become one giant factory table. Keep facts separate by process and connect them through governed shared dimensions.

Shared dimensions

site · product · time · customer

Separate facts

production · inventory · delivery

Explicit paths

only valid business relationships

A Product dimension can be reused across production, inventory, delivery and margin.

Production and inventory facts keep their own natural grain.

Cross-process analysis uses governed dimensions instead of fact-to-fact joins.

Shared-dimension model

One set of dimensions, multiple process-specific data products.

Shared dimensions

Site
Product
Time
Customer

Production performance

run × line

Yield & material loss

run × material

Inventory & shelf life

batch × location

Delivery & OTIF

order line × shipment

Margin

order line × product/customer

Do not connect facts directly just because they share a date or product code. Reuse governed dimensions and explicit relationships.

Production Performance
Metric View

Governed source

production_run_fact

Slice governed measures

The measure stays the same while the business view changes.

Fields

Good output

Yield

Downtime

OEE

Site
Line
Product
Shift
Date

Measure logic lives in one place

Reports and analytics can group the same governed measure by site, line, product, shift or date without rebuilding the KPI logic.

Unity Catalog Metric Views

Define measures once, then slice them by governed fields

Metric Views centralize reusable measures while keeping fields such as site, line, product, recipe, shift and date available for grouping and filtering.

Reusable measures

Standardize KPI logic once instead of recreating it in every report.

Governed fields

Expose the dimensions people need to analyse the metric consistently.

Explicit relationships

Model relationships according to the actual business cardinality.

Read the Databricks Metric Views guide

Business definitions

A KPI formula is not the complete business definition

Reusable semantics need scope, grain, exclusions, units, ownership and effective dates in addition to the calculation itself.

Business concept

Available Stock

Certified
DefinitionQuantity physically present and available for production.
GrainSite × location × batch × SKU
ExcludesBlocked · quarantine · expired
UnitBase inventory unit of measure
OwnerSupply Chain
SynonymsFree stock · usable stock
Effective from2026-01-01

Metrics and terminology are related, but different

Metric Views

Govern reusable measures, fields and semantic relationships.

Business definitions

Add the context people need to interpret a concept correctly.

Certification

Signals which shared assets the organization has approved for reuse.

Operating model

Govern semantic definitions like data products

A shared definition needs ownership, reconciliation and change control. Otherwise the semantic layer becomes another uncontrolled metadata catalog.

Business owner approves meaning, scope and exclusions.

Known operational cases validate the definition before reuse.

Effective dates and versioning preserve historical meaning.

Semantic definition lifecycle

Treat shared meaning as something that is reviewed and maintained.

Define

Agree formula, grain, scope, exclusions, unit and owner.

Validate

Reconcile against known operational cases.

Certify

Mark the approved shared definition as trusted for reuse.

Publish

Expose the data product, metric or governed business concept.

Change

Review version and effective-date impact before changing logic.

Deprecate

Retire outdated definitions instead of leaving competing versions.

Common mistakes

Avoid semantic models that hide business ambiguity

Most semantic-layer failures come from mixing grains, duplicating KPI logic or trying to solve structural data-quality problems too late.

Avoid

KPI logic inside every Power BI model

Better default

Define reusable business measures upstream

Avoid

One giant factory semantic model

Better default

Separate business processes by grain

Avoid

Joining facts directly to other facts

Better default

Use shared dimensions and explicit relationships

Avoid

Using semantics to repair dirty data

Better default

Fix identity and conformance in Silver or Gold

Avoid

A KPI without a business owner

Better default

Assign an owner and technical steward

Avoid

Changing formulas without effective dates

Better default

Version definitions and manage history

FAQ

Frequently asked questions

Practical answers about manufacturing semantic layers on Azure Databricks.

What is a manufacturing semantic layer?

A manufacturing semantic layer is the governed business-meaning layer above trusted data products. It standardizes business terms, relationships, fields and measures so concepts such as OEE, yield, available stock, OTIF and margin are interpreted consistently across reporting, analytics and AI.

Why is grain important in a manufacturing semantic model?

Manufacturing processes operate at different grains. Production runs, machine events, inventory movements, batches, sales order lines and shipments describe different events. The grain should be explicit before metrics and relationships are defined, otherwise joins and aggregations can produce plausible but incorrect results.

Is a semantic layer the same as a Power BI semantic model?

Not necessarily. Power BI semantic models remain useful for report-specific presentation and modeling, but a platform semantic layer can centralize shared business meaning upstream so Power BI, SQL and AI consumers start from the same governed data products and definitions.

Where do Databricks Metric Views fit?

Unity Catalog Metric Views provide a centralized way to define reusable measures, fields and semantic relationships on top of governed data products. They are an important implementation component of Databricks semantics, but they do not replace data quality, conformance or good fact-and-dimension design underneath.

Should all business logic move to Metric Views?

No. Identity resolution, row-level transformations, historical corrections, conformance and data-quality logic should remain in the appropriate Silver or Gold data products. Metric Views should standardize reusable business measures and fields rather than repair structural data problems.

Next step

Make manufacturing definitions reusable across BI and AI

Start with one business process, define its grain and critical measures, then publish a governed semantic contract that reporting, analytics and AI can reuse.

Explore Titan

1. Pick the process

Production, inventory, quality, delivery or margin.

2. Define the grain

Agree what one row and one measure represent.

3. Govern the meaning

Definition, exclusions, units, owner and effective date.

4. Reuse it

Power BI, SQL, analytics and Ask Titan.