name: mdm-planning description: Plan master data management strategies including golden record identification, matching, and merging approaches. allowed-tools: Read, Write, Glob, Grep, Task

Master Data Management Planning

When to Use This Skill

Use this skill when:

Mdm Planning tasks - Working on plan master data management strategies including golden record identification, matching, and merging approaches
Planning or design - Need guidance on Mdm Planning approaches
Best practices - Want to follow established patterns and standards

Overview

Master Data Management (MDM) ensures critical business entities (customers, products, locations) are consistent, accurate, and governed across the enterprise. MDM creates "golden records" - the single source of truth.

MDM Architecture Styles

Style Comparison

Style	Description	Best For
Registry	Index to source systems	Quick wins, low investment
Consolidation	Read-only golden copy	Reporting, analytics
Coexistence	Bidirectional sync	Complex enterprises
Centralized	Single authoring source	Greenfield, tight control
Transaction Hub	Real-time master	High-volume transactions

Architecture Diagram

CONSOLIDATION STYLE
┌─────────────────────────────────────────────────────────────────┐
│                         MDM Hub                                  │
│  ┌─────────────────────────────────────────────────────────┐    │
│  │              Golden Records (Read-Only)                  │    │
│  │   Customer Master │ Product Master │ Location Master     │    │
│  └─────────────────────────────────────────────────────────┘    │
│                              ▲                                   │
│         ┌────────────────────┼────────────────────┐             │
│         │                    │                    │             │
│    ┌────┴────┐          ┌────┴────┐          ┌────┴────┐       │
│    │  Match  │          │  Match  │          │  Match  │       │
│    │ & Merge │          │ & Merge │          │ & Merge │       │
│    └────┬────┘          └────┬────┘          └────┬────┘       │
└─────────┼───────────────────┼───────────────────┼───────────────┘
          │                    │                    │
     ┌────┴────┐          ┌────┴────┐          ┌────┴────┐
     │   CRM   │          │   ERP   │          │E-Commerce│
     └─────────┘          └─────────┘          └──────────┘

Data Domains

Common Master Data Domains

Domain	Examples	Typical Attributes
Party	Customer, Vendor, Employee	Name, Address, Contact
Product	SKU, Service, Asset	Description, Category, Price
Location	Site, Store, Warehouse	Address, Coordinates, Type
Financial	Account, Cost Center	Code, Name, Hierarchy
Reference	Country, Currency, Status	Code, Description

Domain Assessment Template

# Master Data Domain: Customer

## Domain Scope
- B2B customers
- B2C customers
- Prospects
- Leads

## Source Systems
| System | Records | Quality Score | Authority |
|--------|---------|---------------|-----------|
| CRM | 150,000 | 85% | High |
| ERP | 120,000 | 75% | Medium |
| E-Commerce | 200,000 | 60% | Low |

## Key Challenges
- Duplicate customers across systems
- Inconsistent address formats
- Missing email validation
- No unique identifier across systems

## Business Impact
- $X million in duplicate mailings
- Y% order errors from wrong customer
- Z hours/week manual reconciliation

Matching Strategies

Matching Approaches

Approach	Description	Accuracy	Speed
Exact	Identical values	High	Fast
Fuzzy	Similar values (Levenshtein, Soundex)	Medium	Medium
Probabilistic	Statistical likelihood	Highest	Slow
Rule-Based	Business logic	Configurable	Fast

Match Rules Template

# Match Rules: Customer Domain

## Rule 1: Exact Email Match
- Field: email
- Type: Exact (case-insensitive)
- Confidence: 95%
- Auto-merge: Yes

## Rule 2: Name + Address
- Fields: last_name, address_line1, postal_code
- Type: Fuzzy (threshold 85%)
- Confidence: 80%
- Auto-merge: No (manual review)

## Rule 3: Phone + Last Name
- Fields: phone, last_name
- Type: Exact phone, Fuzzy name
- Confidence: 75%
- Auto-merge: No

## Rule 4: Company Match
- Fields: company_name, city
- Type: Fuzzy (threshold 80%)
- Confidence: 70%
- Auto-merge: No

C# Matching Example

public class CustomerMatcher
{
    public MatchResult Match(Customer source, Customer target)
    {
        var scores = new List<MatchScore>();

        // Exact email match (high confidence)
        if (!string.IsNullOrEmpty(source.Email) &&
            source.Email.Equals(target.Email, StringComparison.OrdinalIgnoreCase))
        {
            scores.Add(new MatchScore("Email", 0.95m));
        }

        // Fuzzy name match
        var nameScore = CalculateLevenshteinSimilarity(
            source.FullName, target.FullName);
        if (nameScore > 0.85m)
        {
            scores.Add(new MatchScore("Name", nameScore * 0.7m));
        }

        // Address match
        var addressScore = CalculateAddressSimilarity(
            source.Address, target.Address);
        if (addressScore > 0.80m)
        {
            scores.Add(new MatchScore("Address", addressScore * 0.6m));
        }

        var overallScore = scores.Sum(s => s.Weight) / scores.Count;

        return new MatchResult
        {
            IsMatch = overallScore > 0.75m,
            Confidence = overallScore,
            Scores = scores,
            RequiresReview = overallScore is > 0.6m and < 0.85m
        };
    }
}

Survivorship Rules

Survivorship Strategies

Strategy	Description	Use Case
Most Recent	Latest timestamp wins	Frequently updated data
Most Trusted	Source priority	Known data quality
Most Complete	Non-null preferred	Sparse data
Longest	Longest string wins	Descriptions, names
Aggregate	Combine values	Lists, tags
Custom	Business logic	Complex rules

Survivorship Configuration

public class CustomerSurvivorship
{
    public Customer Merge(IEnumerable<Customer> duplicates)
    {
        var ordered = duplicates
            .OrderByDescending(d => GetSourcePriority(d.SourceSystem))
            .ThenByDescending(d => d.ModifiedDate)
            .ToList();

        return new Customer
        {
            // Most trusted source
            CustomerId = ordered.First().CustomerId,

            // Most complete (first non-null)
            FirstName = ordered.Select(d => d.FirstName)
                .FirstOrDefault(v => !string.IsNullOrEmpty(v)),

            LastName = ordered.Select(d => d.LastName)
                .FirstOrDefault(v => !string.IsNullOrEmpty(v)),

            // Most recent
            Email = ordered
                .OrderByDescending(d => d.EmailVerifiedDate)
                .Select(d => d.Email)
                .FirstOrDefault(v => !string.IsNullOrEmpty(v)),

            // Aggregated
            Tags = ordered.SelectMany(d => d.Tags ?? [])
                .Distinct()
                .ToList(),

            // Highest value
            LifetimeValue = ordered.Max(d => d.LifetimeValue),

            // Custom logic
            Segment = DetermineSegment(ordered)
        };
    }

    private int GetSourcePriority(string source) => source switch
    {
        "CRM" => 100,
        "ERP" => 80,
        "E-Commerce" => 60,
        _ => 40
    };
}

Reference Data Management

Reference Data Types

Type	Description	Examples
Code Lists	Simple lookups	Countries, Status codes
Hierarchies	Parent-child	Categories, Org structure
Cross-References	System mappings	Product codes across systems
Rules	Business logic	Tax rates, Pricing tiers

Reference Data Table

-- Reference data tables (PascalCase - SQL Server Convention)
CREATE TABLE RefCountry (
    CountryCode CHAR(2) PRIMARY KEY,
    CountryCode3 CHAR(3) NOT NULL,
    CountryName VARCHAR(100) NOT NULL,
    Region VARCHAR(50),
    IsActive BIT DEFAULT 1,
    EffectiveDate DATE NOT NULL,
    ExpirationDate DATE,
    CreatedAt DATETIME2 DEFAULT GETUTCDATE(),
    UpdatedAt DATETIME2
);

CREATE TABLE RefCrossReference (
    CrossRefId INT IDENTITY PRIMARY KEY,
    Domain VARCHAR(50) NOT NULL,
    SourceSystem VARCHAR(50) NOT NULL,
    SourceCode VARCHAR(100) NOT NULL,
    TargetSystem VARCHAR(50) NOT NULL,
    TargetCode VARCHAR(100) NOT NULL,
    IsActive BIT DEFAULT 1,

    CONSTRAINT UQ_CrossRef
        UNIQUE (Domain, SourceSystem, SourceCode, TargetSystem)
);

MDM Governance Model

RACI Matrix

# MDM Governance RACI

| Activity | Data Owner | Data Steward | IT | Business |
|----------|------------|--------------|-----|----------|
| Define data standards | A | R | C | C |
| Approve match rules | A | R | C | I |
| Review merge candidates | I | R | C | A |
| Monitor data quality | I | R | A | C |
| Resolve data issues | A | R | C | I |
| Maintain MDM system | I | C | R | I |

R = Responsible, A = Accountable, C = Consulted, I = Informed

Data Stewardship Workflow

flowchart TD
    A[New Record Ingested] --> B{Duplicate Found?}
    B -->|No| C[Create Golden Record]
    B -->|Yes| D{Auto-Merge Threshold?}
    D -->|Yes| E[Auto-Merge]
    D -->|No| F[Queue for Review]
    F --> G[Data Steward Review]
    G --> H{Decision}
    H -->|Merge| E
    H -->|Not a Match| I[Mark as Distinct]
    H -->|Needs Investigation| J[Research Queue]
    E --> K[Update Golden Record]
    C --> K
    I --> K

MDM Platform Components

┌─────────────────────────────────────────────────────────────────┐
│                      MDM Platform                                │
├─────────────────────────────────────────────────────────────────┤
│  ┌────────────┐  ┌────────────┐  ┌────────────┐  ┌───────────┐ │
│  │ Data       │  │ Match &    │  │ Workflow   │  │ Data      │ │
│  │ Integration│  │ Merge      │  │ Engine     │  │ Quality   │ │
│  └────────────┘  └────────────┘  └────────────┘  └───────────┘ │
│  ┌────────────┐  ┌────────────┐  ┌────────────┐  ┌───────────┐ │
│  │ Hierarchy  │  │ Reference  │  │ Governance │  │ API       │ │
│  │ Manager    │  │ Data Mgmt  │  │ Dashboard  │  │ Gateway   │ │
│  └────────────┘  └────────────┘  └────────────┘  └───────────┘ │
└─────────────────────────────────────────────────────────────────┘

Validation Checklist

Master data domains identified and prioritized
Source systems inventoried with quality assessment
MDM style selected (registry, consolidation, etc.)
Match rules defined with confidence thresholds
Survivorship rules documented per attribute
Governance roles assigned (owners, stewards)
Reference data management approach defined
Data quality metrics established

Integration Points

Inputs from:

conceptual-modeling skill → Entity candidates
data-governance skill → Stewardship model
data-quality-planning skill → Quality rules

Outputs to:

schema-design skill → MDM hub schema
migration-planning skill → Initial load strategy
data-lineage skill → Cross-reference tracking

mdm-planning

$ 安裝