Designing an Enterprise Master Data Management (MDM) Platform: A Complete Guide

Overall Design Principles

The platform serves as the single source of truth for core master data such as customers, suppliers, products, organizations, employees, financial accounts, warehouses, and channels. All downstream systems (ERP, CRM, OA, WMS, e‑commerce) must read, create, modify, and distribute data exclusively through the MDM, ensuring consistent data across the group.

Unique source : Only the MDM can create core data; other systems are prohibited from creating their own records.

Unified standards : Group‑wide coding rules, field definitions, classifications, and formats are defined and enforced.

Full lifecycle control : Data creation → review → release → distribution → change archiving → deactivation/deletion.

Multi‑system synchronization : Bidirectional integration with ERP, CRM, OA, e‑commerce, WMS.

Layered permission control : Central administrators set standards; subsidiaries maintain only their own data via RBAC3.

Data cleansing : Automatic deduplication, missing‑field completion, and error correction for historical dirty data.

High availability : Supports multi‑level organization, multi‑industry, and tenant isolation.

Extensibility : New master‑data types (e.g., assets, projects) can be added by configuration only.

Implementation Process (Five Phases)

Phase 1 – Requirement Gathering & Master‑Data Inventory

Identify all business systems and list the basic data each stores.

Select core master‑data types (customer, supplier, product, organization, employee, warehouse, financial account).

Define coding rules, mandatory fields, dropdown enumerations, and classification hierarchies for each type.

Catalog dirty‑data issues such as duplicate customers, inconsistent naming, missing IDs, and chaotic classifications.

Phase 2 – Architecture & Module Design

Divide the platform into four hierarchical modules (detailed later).

Design database schema, synchronization interfaces, message middleware, and caching layer.

Define RBAC3 roles, organization isolation, and data‑masking policies.

Phase 3 – Historical Data Cleansing & Initialization

Extract historical data from each source system.

Run automatic cleaning: deduplication, field completion, format standardization, error flagging.

Human reviewers correct remaining issues, then bulk‑load the cleaned data into the MDM as the authoritative repository.

Phase 4 – System Integration & Synchronization Configuration

Configure integration methods: API calls, direct database sync, or message queues (JMQ / RocketMQ).

Set distribution rules that specify which downstream systems receive which master‑data types.

Define back‑flow rules: downstream systems can only query; any create/modify actions must be routed back to the MDM for review.

Phase 5 – Pilot Run & Ongoing Operations

Dual‑track operation: keep legacy data alongside MDM data for one month.

Train subsidiary data administrators on the new processes.

Schedule regular data inspections, duplicate‑data clean‑ups, and synchronization optimizations.

Enable full audit logging for every data change.

Four‑Level Functional Modules (Six Primary Modules)

1. Data Standard Management

Master‑data type management – add, enable/disable, tenant isolation.

Encoding rule configuration – custom formats (e.g., customer code: 01<branch>00001), auto‑generation, duplicate checks.

Field standard management – templates, mandatory/optional flags, format validation (e.g., 11‑digit phone, 18‑digit ID).

Data dictionary – common enumerations, hierarchical categories, deprecation handling.

Uniqueness and value‑range rules – single‑field and composite uniqueness, numeric ranges, date constraints.

2. Master‑Data Lifecycle Management

Manual single‑record entry with real‑time duplicate alerts and draft saving.

Batch import via standardized Excel templates, automatic format validation, and bulk submission for review.

Multi‑level approval workflow configuration, role binding, and rejection‑reason templates.

Publishing controls: immediate release, scheduled low‑peak release, and automatic retry on failure.

Change management with version comparison and downstream propagation.

Archiving, logical deletion, physical deletion with permission checks and approval.

Advanced search (multi‑condition, hierarchical tree, fuzzy global) and customizable export (field selection, pagination, audit logging).

3. Data Integration & Synchronization

Downstream system onboarding: API endpoint, direct DB connection, or message‑queue configuration.

Upstream back‑flow settings with filtering and exception alerts.

Distribution rules based on data type, tenant, and field sensitivity.

Sync frequency policies – real‑time, scheduled incremental, full re‑sync.

Monitoring panel showing success/failure counts, detailed failure logs, and manual re‑push capability.

Alert channels (email, enterprise WeChat, SMS) for timeouts, bulk failures, and latency breaches.

4. Data Cleansing & Governance

Quality rules: mandatory‑field detection, classification completeness, null‑value filling.

Duplicate detection (e.g., same name + phone) and merge policies (keep latest record).

Format normalization and sensitive‑field masking.

Scheduled cleaning tasks (daily, weekly, monthly) with automatic report generation.

Manual ad‑hoc cleaning for specific data subsets.

Governance dashboard showing total dirty‑data volume, subsidiary rankings, and trend curves.

One‑click auto‑repair for simple issues and closure logging.

5. Permission, Tenant & Organization Management (RBAC3)

Multi‑tenant organization tree (group → subsidiary → department) with isolation.

Tenant isolation for partners, independent data standards, and restricted data distribution.

Role inheritance, mutual exclusion, time‑bound assignments, and bulk user‑role operations.

Fine‑grained UI, button, and field permissions, including data‑masking for sensitive fields.

User account lifecycle, password policies, SSO integration (WeChat, Feishu, OA), IP whitelist, and single‑device login enforcement.

6. Operations Monitoring, Logging & Auditing

Full‑link operation logs for data CRUD, approvals, batch imports/exports, and permission changes.

System health monitoring: request volume, DB storage, scheduled‑task execution.

Alert center with multi‑channel notifications and severity levels.

Business reports: master‑data totals, monthly additions/retirements, subsidiary comparisons, sync success/failure statistics, and back‑flow volumes.

Common Production Issues and Recommended Solutions

Resistance to switch to MDM : dual‑track migration, administrative policies forbidding unsanctioned data creation, simplified entry UI, dedicated data admins per subsidiary.

Massive historical dirty data : automated cleaning plus manual review, split tasks by subsidiary, prioritize financial‑critical data, full backup before cleaning.

Frequent sync failures : automatic retry, monitoring dashboard with real‑time alerts, nightly full‑incremental sync, one‑click full re‑sync after downstream recovery.

Cross‑subsidiary data leakage : organization‑bound permissions, physical table partitioning per tenant, field‑level masking, audit logs for cross‑org queries.

Downstream systems modifying data without back‑flow : read‑only UI, enforced back‑flow validation, daily diff reports, financial reconciliation based on MDM data.

Lengthy approval workflow : separate simple vs. core data paths, batch approval, mobile approval via enterprise WeChat, fast‑track role for temporary items.

Performance bottlenecks with millions of records : horizontal sharding by subsidiary/type, Redis cache for hot master data, low‑peak batch jobs, pagination for queries and exports.

Permission abuse (self‑review) : RBAC2 mutual‑exclusion roles, full audit trails, periodic admin spot‑checks.

Integration overhead for new systems : standardized open APIs returning JSON, no‑code DB sync or MQ adapters, pre‑built templates for common ERP/e‑commerce/WMS.

Employee turnover data leakage : immediate account lock, export logging with alerts, field‑level masking, IP whitelist for backend access.