---
title: "Master data and authoritative sources: how to achieve a single version of the truth without a major MDM project"
description: "What master, reference, transactional and analytical data are, how to declare an authoritative source using seven criteria and how to manage its lifecycle without a full MDM rollout."
url: https://bidatia.com/en/ideas/datos-maestros-y-fuentes-de-autoridad
lang: en
date: 2026-08-04
updated: 2026-11-04
author: "Santos Vicente"
keywords: ["master data", "master data management", "mdm", "authoritative source", "single source of truth", "golden record", "reference data", "data governance"]
publisher: Bidatia
---

# Master data and authoritative sources: how to achieve a single version of the truth without a major MDM project

> What master, reference, transactional and analytical data are, how to declare an authoritative source using seven criteria and how to manage its lifecycle without a full MDM rollout.

## Executive summary

- Without a single reference for customers, suppliers or addresses, each department keeps its own version and the indicators do not add up.
- Types of data: master, reference, transactional and analytical.
- An authoritative source is the dataset recognised as the single reference for an entity; it is declared using seven criteria (quality, traceability, owners, single definition, effective use, maintenance and formal approval).
- Master data is distributed by reference, not by copy, and is logically deactivated rather than deleted.
- Includes an example of a single version of customer data, source integration patterns, the data access agreement and when an MDM tool is worth it.

When each department keeps its own version of customers, suppliers or addresses, indicators stop being comparable and nobody trusts the data. The classic answer is a master data management (MDM) project, which is long and expensive. There is a more gradual path: **declaring authoritative sources** for the most important entities and governing them.

## Four types of data

| Type | What it is | Examples | Volatility |
|---|---|---|---|
| Master | Entities shared by several processes, with their own lifecycle | People, customers, properties, suppliers | Low |
| Reference | Stable values that classify or encode other data, often from external standards | Street directory, INE codes, NACE codes, product catalogue | Very low |
| Transactional | Facts generated when processes are executed | Orders, invoices, applications, incidents | High |
| Analytical | Derived data for analysis and decision-making | Indicators, dashboards | Medium |

## What an authoritative source is

It is the dataset recognised as the **single corporate reference** for an entity: it has proven quality and a traceable origin, and users genuinely use it instead of copies. It is not a new system, but a status granted to an existing dataset.

## The seven criteria

| Criterion | What it means | How it is checked |
|---|---|---|
| Proven quality | Meets acceptable levels across the data quality dimensions | Active quality rules and indicators |
| Traceable origin | The source system, process, frequency and transformations are known | Lineage published in the catalogue |
| Assigned owners | It has a data owner and a data steward | Asset record in the catalogue |
| Single definition | Each concept is in the glossary and has been validated by those who use it | Approved term in the glossary |
| Effective use | It is used instead of copies or parallel sources | Uses documented in the lineage |
| Active maintenance | There is a live update and correction process | Frequency and incident history |
| Formal approval | Its declaration has been approved by the governance body | Register of authoritative sources |

## How it is declared

1. **Candidates**: datasets with the same entity in several systems, or that need to be shared.
2. **Comparative analysis**: coverage, quality, timeliness, governance and users of each source.
3. **Assessment** against the criteria. If the best candidate does not meet all of them, a remediation plan with milestones is defined.
4. **Proposal**: entity, dataset, justification, alternatives ruled out and impact on current uses.
5. **Approval** by the data governance committee.
6. **Publication**: it is flagged in the catalogue as an authoritative source and the glossary is updated.
7. **Transition**: those who used parallel sources are supported, to a schedule.
8. **Annual review**, or whenever there are significant changes.

## The lifecycle of master data

Master data needs ongoing maintenance and is distributed **by reference, not by copy**: those who use it access the source (shared views, internal services) instead of replicating it.

| Phase | What is done |
|---|---|
| Definition | Meaning, mandatory attributes, validation rules and uniqueness criteria |
| Creation | Create records after checking they do not already exist |
| Modification | Update while keeping a trail of every change |
| Quality | Ongoing rules for completeness, uniqueness and referential integrity |
| Use | Publish for use by reference |
| Logical deactivation | Mark as inactive, without deleting |
| Deletion | Physical deletion only when required by data protection or the retention policy |

## Where it lives

A domain's master data stays in that domain, under its owner. Cross-cutting data with no clear domain, and external reference data, is managed in a data governance domain. In the catalogue, all you need is an “Authoritative source” attribute (with the date and the approving body), a “Data type” attribute, the glossary term linked to the source and the lineage to see who uses it.

## An example: a single version of customer data

A company has customers in the ERP, in the CRM, in the billing tool and in a spreadsheet kept by the sales team. When management asks “how many active customers do we have?”, each department gives a different figure.

1. **Candidates**: ERP and CRM.
2. **Comparative analysis**: the ERP has all invoiced customers, with validated tax IDs; the CRM has more contacts, but many are prospects and there are duplicates.
3. **Assessment**: the ERP meets almost all the criteria, but it lacks a formal owner and the definition of “active customer” is not in the glossary.
4. **Remediation plan**: appoint an owner (the finance department) and a steward, agree the definition (“a customer with at least one invoice in the last 12 months”) and activate uniqueness rules for the tax ID.
5. **Approval and publication**: the committee declares the ERP the authoritative source for customers.
6. **Transition**: the CRM links its accounts to the ERP customer code, and the spreadsheet is retired.

Result: a single figure for active customers, traceable and with someone accountable for it.

## How sources are integrated

Each integration uses an agreed pattern, and exceptions must be justified:

| Pattern | Latency | Typical use |
|---|---|---|
| Batch (ETL/ELT) | Hours or daily | Business applications and files; the default pattern |
| Event streaming | Seconds or minutes | Sensors and real-time platforms |
| On-demand API | On demand | Ad hoc queries to third-party services or other public bodies |
| Change data capture (CDC) | Minutes | Databases that must stay in sync |
| Virtualisation | On demand | One-off cases, without copying the data |
| Document ingestion | Batch or event-driven | Document management systems and knowledge bases for AI |

And no integration goes live without a **data access agreement** with the owner of the source system: which data, with what pattern and frequency, which schema and version, what sensitivity and restrictions, what availability and how changes are communicated.

## When an MDM is worth it

Declaring authoritative sources solves most cases. A master data management tool makes sense when:

- the same entity is **created** in several systems at once and records need to be merged (identifying that two customers are the same);
- an **approval workflow** is needed to create or modify master records;
- the volume of duplicates and manual corrections is so high that it cannot be managed with rules;
- several applications need to receive changes to the master data in near real time.

Even then, it is advisable to start with one entity and expand.

## Common mistakes

- **Copying the master data into every system** “so it runs faster”, and ending up with different versions again within a month.
- **Declaring the largest system the authoritative source** without checking the criteria.
- **Deleting records** that are no longer current, breaking history and reports.
- **Not supporting the transition**: if nobody helps users of the parallel source to migrate, they will keep using it.
- **Forgetting external reference data** (postcodes, NACE codes, country codes): it also needs an owner and updates.

## Why it matters for AI

An assistant or model fed by divergent copies gives divergent answers. Declaring authoritative sources for the most widely used entities, and prioritising their quality, is one of the highest-return steps before any AI project.

At Bidatia we help identify and declare authoritative sources and set up their governance on Microsoft Fabric, Purview or OpenMetadata. To learn more, see [data quality](/en/ideas/calidad-del-dato-dimensiones-metricas) and [metadata management](/en/ideas/gestion-de-metadatos-catalogo-de-datos).

## Frequently asked questions

### What is the difference between master data and reference data?
Master data describes essential entities with their own lifecycle, such as customers or suppliers. Reference data consists of stable values that classify other data, such as country codes or the street directory, and often comes from external standards.

### Do you need an MDM to have a single version of the truth?
Not necessarily. You can start by declaring authoritative sources for the most relevant entities, with criteria, owners and formal approval, and consider an MDM later if needed.

### What is an authoritative source?
The dataset formally recognised as the single reference for an entity, with proven quality, traceable origin, owners and effective use, approved by the data governance body.

### Why distribute master data by reference rather than by copy?
Because every copy goes out of date and diverges. By accessing the source through views or services, everyone uses the same data and changes reach everyone at the same time.

### Can master data be deleted?
The usual approach is logical deactivation: marking records as inactive. Physical deletion is only carried out when required by data protection or the retention policy.