Yagnesh RamalingamTopic · part of a larger one
Semantic layer
Start here if two dashboards show different numbers for the same thing.
When two dashboards disagree, the meeting becomes an argument about definitions. That argument is the semantic layer failing in public. Write the contract before you write the chart.
The explainer
Semantic layer. In practice.
Vendors will sell you a semantic layer as a place to store joins so analysts stop rewriting them. That convenience is real. It is not the job. The job is an agreement that survives the people who wrote it: what the metric means, who owns that meaning, and which source is allowed to compute it.
What the contract has to say
- The grain. Deal, account, order line, or whatever level the number is actually counted at.
- The date that counts, and the timezone.
- What is excluded, by name: returns, internal accounts, tests, renewals.
- The one source allowed to compute it, and the owner who can refuse a change.
Fail any of those and two dashboards will disagree while both remain defensible. The warehouse is doing what it was asked. The asks differ. The article on that disagreement walks the usual gaps. The twelve-check list is the same contract as a working session.
It is not a feature of the BI tool
A modelling layer inside a product is useful after the agreement exists. Before the agreement, it industrialises folklore: the same hallway definition, now with a YAML file and a refresh schedule. Write the page first. Put it in software when a second team needs to consume it without a meeting.
| Semantic layer (the contract) | Metrics inside a BI tool | |
|---|---|---|
| What it is | An agreement on what a metric means, who owns it and which source computes it | A calculation stored inside one product |
| Survives a tool change | Yes. It is written down outside the tool | No. The next tool invents its own |
| Who writes it | The person who defends the number, with the analyst who computes it | Whoever builds the dashboard |
| When it goes into software | When a second team needs it without a meeting | Usually first, before any agreement exists |
This sits under governed data. Governance says the source has an owner and a purpose. The semantic layer says the metric built on that source has one meaning. Skip the first and the second has nothing stable to point at.
One metric, before the next meeting
Choose the figure two teams already defend. Write the four lines above. Take the metric off any slide that cannot cite them. A contested slide costs more hours than the contract. That is the whole adoption plan.
Articles
Articles on the semantic layer
No article on this one yet. See every article.
Keep going
Related topics. Where to go next.
- Broader topicGoverned dataStart here if nobody trusts what comes out of the warehouse.Explore the topic
- TopicDecision latencyStart here if a simple question takes weeks to become a number anyone will act on.Explore the topic
- TopicExtended Planning & Analysis (XP&A)Start here if every planning cycle begins by reconciling last month.Explore the topic
- TopicAgentic analyticsStart here if you are being asked to put an AI assistant on your data.Explore the topic
Case studies
Case studies. What changed, and how.
- Before ZerentroMulti-brand consumer enterprise
11 days→4 hours
Decision latency on the weekly pipeline question
Three answers, one Monday question
A weekly pipeline question that used to take 11 days now returns a sourced answer the same afternoon.
Read the case study - Before ZerentroEnterprise operations
18→6 days
Monthly forecast cycle
When the forecast became a political artefact
Forecast cycle time dropped from 18 days to 6, because the argument about last month’s actuals stopped happening twice.
Read the case study
Questions about the semantic layer
What is a semantic layer?
A semantic layer is the contract that says what a metric means, who owns that meaning, and which source is allowed to compute it. It is not a visualisation feature. It is an organisational agreement written down.
What does Yagnesh Ramalingam say about the semantic layer?
When two dashboards disagree, the meeting becomes an argument about definitions. That argument is the semantic layer failing in public. Write the contract before you write the chart.
Why do two dashboards show different numbers?
Because the metric was never written down as a contract. Two teams built the same figure from two defensible definitions, with a different grain, a different source or a different treatment of returns and timing, and both are correct against their own logic. The fix is a written definition, a named owner who may refuse a change, and one source permitted to compute it.
Is a semantic layer a BI feature?
A product can host the contract. It cannot be the contract. If the meaning lives only inside one tool, the next tool will invent its own, and the meeting will be back where it started.
How many metrics need one?
The five that actually run the company, before any of the hundreds that decorate a catalogue. A contract for a metric nobody decides on is filing.
Who writes it?
The person who will defend the number, with the analyst who computes it in the room. A central team can keep the format. It cannot supply the meaning.
Work with me
Bring the number your team can’t agree on.
A free 30-minute call with Yagnesh. You'll find out if the problem is the data, the definition or who owns it.