How to Manage KPI Definitions and a Glossary in a Data Catalog

---

To manage KPI definitions and a business glossary in a data catalog, store each KPI as a governed glossary term that carries a formula, an owner, a certification status, and links to the exact columns and dashboards that compute it. Approve changes through named stewards, and review critical KPIs every quarter. The catalog then shows every user the same definition wherever the number appears.

The problem is common. In the Precisely and Drexel University LeBow 2025 Outlook report, a survey of more than 550 data professionals, 45% named inconsistent data definitions and formats as an obstacle, and 67% said they do not completely trust their organization’s data.

---

What is the difference between a glossary term, a KPI definition, a data dictionary and a data catalog?

A business glossary holds the meaning of business concepts in plain language. A KPI definition is a glossary term with a calculation attached. A data dictionary describes the physical fields. The data catalog is the inventory that holds all of them, with discovery, lineage and usage, and it is the surface terms attach to.

	Glossary term	KPI definition	Data dictionary entry	Data catalog
Answers	What does this word mean here?	How is this number calculated, and who signs off?	What is this column, and what type is it?	Where is the data, where did it come from, and who uses it?
Example	“Customer”	“Net revenue retention”	fct_orders.net_amount_usd	The fct_orders table, its lineage and the dashboards it feeds
Written by	Business owners and stewards	Metric owner, often in finance or ops	Engineers, or ingested automatically	Mostly ingested by connectors, curated by stewards
Must include	Definition, synonyms, owner	Formula, filters, grain, source columns, owner, status	Name, data type, constraints	Assets, lineage, usage, and links to terms
Typical failure	Two teams use one word for different things	Two dashboards show different numbers for one KPI	Field changes break downstream reports	Users find a table but not what it means

The glossary failure is called semantic drift, and revenue is the classic case. One widely cited guide to startup metrics separates bookings, the value of a signed contract, from revenue, which GAAP recognizes as the service is delivered. A sales deck and a finance report can both say “revenue” and show two different numbers. Each is valid in its own system. The glossary gives each one its own name and owner.

For a deeper split of these concepts, see data dictionary vs business glossary and business glossary vs data catalog.

---

What fields should every KPI definition and glossary term carry?

A definition sentence alone tells a reader what someone meant once. It does not say which table to query, who to ask, or whether the definition still holds.

Guides from the pages buyers read most agree on a short core. A glossary build guide lists term name, definition, owner, related terms, data sources and formula. A public KPI template from the UN World Food Programme adds a calculation formula, calculation steps, a data source, a responsible division, a review frequency, and the KPI’s assumptions and limitations.

Field	Glossary term	KPI definition	Why it matters
Name and unique ID	Required	Required	Stops duplicates and lets tools reference the term
Plain-language definition	Required	Required	The sentence a business user reads
Synonyms and aliases	Required	Required	Routes “client” and “account” to “customer”
Owner and steward	Required	Required	One person accountable, one person maintaining
Formula	Optional	Required	The calculation, written once
Filters and grain	Optional	Required	Excludes test accounts; sets daily or monthly
Source columns	Recommended	Required	The certified table and columns behind the number
Status	Required	Required	Draft, approved or deprecated
Review date	Recommended	Required	When someone last confirmed it is still true

For synonyms, borrow a rule from the W3C SKOS standard, a Recommendation since 18 August 2009: a concept has no more than one preferred label per language, and any number of alternative labels. Apply that rule and “MRR”, “monthly recurring revenue” and “recurring revenue (monthly)” resolve to one term with one owner.

---

How do you link glossary terms and KPIs to catalog assets?

A glossary that floats above the data gets ignored. The value appears when a user opens a dashboard or a column and sees the approved definition right there, with its owner.

Link at the column level wherever you can. A link on a whole report says “this report uses net revenue somewhere”. A link on net_amount_usd says exactly which field computes it, and lineage can carry that meaning to every downstream table and chart. Cloud catalogs support both levels. One cloud catalog’s documentation allows up to 10 glossary terms per asset and 5 per column during its preview, which also hints at good practice: an asset that needs ten terms probably holds ten ideas.

Before you map, decide the mapping rules. One alignment guide sorts links by cardinality: one-to-one, one-to-many, many-to-one and many-to-many. It also warns about over-linking. A “net revenue” term mapped to a “gross sales” column is a wrong link, and a wrong link is worse than no link because users trust it.

Plan for scale limits too. Another cloud catalog caps a glossary at 5,000 terms and 200 categories, nested up to three levels. Few teams reach those caps, but they push you toward a flat structure organized by domain, such as finance, sales and product, instead of deep trees.

Automation can suggest matches between terms and columns based on names and usage. Treat those as drafts. A steward confirms each link against lineage and the formula before it goes live.

---

Who owns a KPI definition, and how is a change approved?

Every KPI needs one accountable business owner. Committees can review a definition, but a committee cannot answer a Slack message at 9 a.m. when the board deck shows two numbers.

Most programs split the work into three roles. The business owner approves the meaning and is accountable for it. The data steward maintains the entry, validates links and watches for drift. A technical validator confirms that the linked column really computes what the definition says. A guide to aligning a glossary with a catalog uses this split.

Changes move through statuses. A typical lifecycle looks like this:

1. Draft. A steward writes or imports the term. Only stewards and owners see it.
2. Review. The owner and affected teams comment and agree on one definition.
3. Published or certified. Every catalog user sees it, and dashboards inherit it.
4. Deprecated or expired. The term stays visible to stewards for audit but disappears from search.

Cloud catalogs build in the same flow. In one widely used catalog, a new term starts in a draft state visible only to stewards and domain owners, bulk imports accept up to 1,000 rows per CSV file, and retired terms are set to expired rather than deleted.

Do not gate every term behind approval on day one. Put the heaviest review on KPIs that reach executives, regulators or customers. A definition for “active subscriber” in a board metric deserves a sign-off. A definition for “SKU” does not.

---

How do you keep KPI definitions from drifting after launch?

Definitions decay quietly. Marketing counts an “active user” over 30 days, product counts one over 14 days, and both dashboards stay green. A guide to aligning a glossary with a catalog uses this example of semantic drift.

Three habits stop most drift.

First, set review cadences by risk. A guide to building a business glossary recommends quarterly or twice-yearly reviews, with unused terms archived. Put certified KPIs on the quarterly cycle and standard terms on an annual one.

Second, write the formula where the computation lives. If your KPI logic lives in dbt, analysts define each metric once in version-controlled YAML, and dbt’s metrics docs list five metric types: conversion, cumulative, derived, ratio and simple. When the catalog reads that definition instead of a copy, the glossary and the dashboard cannot disagree for long, and every change shows up in a pull request.

Third, measure the glossary itself. Useful health metrics include:

• term-to-asset coverage, the share of priority terms linked to validated columns;
• duplicate terms found and merged per quarter;
• KPIs past their review date;
• schema changes that triggered a mapping review.

A rising count of overdue KPIs tells you the program is losing owners before any dashboard breaks. For a step-by-step maintenance routine, see how to keep a business glossary up to date, and for how the glossary fits the wider program, data governance vs business glossary.

---

How does Atlan manage KPI definitions and a business glossary?

Atlan models a glossary in three layers: glossaries, categories and terms. Atlan’s guide to structuring glossaries and metrics places permissions at the top-level glossary, so teams that need isolation get separate glossaries, and categories organize terms by domain inside one glossary.

In Atlan, a metric is a term with extra required fields. The same guide lists them: the formula, data sources down to tables and columns, refresh frequency, benchmarks or historical trend, the accountable business owner, and certification status. It recommends one term per distinct formula, so “net revenue (USD)” and “net revenue (local currency)” become two related terms instead of one ambiguous one.

Atlan recommends linking terms at the column level after connectors are live, so the meaning follows lineage to downstream tables and dashboards. Certificates move a term from draft to verified to deprecated. The guide suggests quarterly reviews for critical metrics, annual reviews for standard terms, and archiving deprecated terms after about six months instead of deleting them. It also advises against approval workflows in the first phase: populate first, then add targeted workflows once the glossary passes roughly 100 terms.

Kiran Panja, Managing Director of Cloud and Data Engineering at CME Group, says on atlan.com that within the first year the team cataloged over 18 million assets and defined more than 1,300 glossary terms.

Through Atlan MCP, Atlan serves certified metric definitions, glossary terms and column-level lineage to AI agents, so an agent asked for net revenue retention reads the formula the finance dashboard uses.

---

FAQs about data catalogs for business glossary and KPI definitions

Should KPI definitions live in the business glossary or in a separate metrics catalog?

Keep them in the glossary as a term type with extra required fields: formula, grain, filters and source columns. A separate metrics list drifts from the glossary over time, because two teams maintain two copies. One record per KPI, linked to its columns and dashboards, gives users one place to look and stewards one place to update.

How many terms should a business glossary start with?

Start small, with the terms that appear in executive and regulatory reporting, or that two teams already define differently. Those carry the most risk and the most visible payoff. Add terms as users search for words that return nothing. A glossary that launches with thousands of unowned terms teaches users to ignore it.

Who should own a KPI definition?

One named business owner per KPI, usually the leader whose team is judged on the number, such as finance for net revenue. A data steward maintains the entry and its links, and a technical validator confirms the linked column computes the formula. Naming a team instead of a person leaves nobody to answer when two dashboards disagree.

How often should KPI definitions be reviewed?

Review certified KPIs every quarter and standard glossary terms once a year. Trigger an extra review when a source schema changes, a new dashboard reuses the KPI, or the owner changes roles. Record the review date on the term itself, so any user can see whether the definition was confirmed recently or has gone untouched for two years.

Can AI write glossary definitions for you?

AI can draft definitions from column names, SQL, dashboard titles and usage, which shortens the blank-page stage. It cannot decide which of two competing definitions the business has agreed to. Treat every AI draft as a draft: a steward checks it against the formula and lineage, and the owner approves it before it is published.

---

Sources

1. 2025 Outlook: data quality remains the top data integrity challenge, Precisely and Drexel University LeBow
2. 16 Startup Metrics, a16z
3. KPI template, UN World Food Programme indicator compendium
4. SKOS Simple Knowledge Organization System Reference, W3C
5. Metrics overview, dbt documentation
6. A glossary build guide
7. A guide to aligning a glossary with a catalog
8. One cloud catalog’s glossary documentation
9. Another cloud catalog’s glossary documentation
10. One widely used catalog’s glossary documentation