Master data and deduplication

Finds duplicate business records and prepares auditable merges or imports without losing relationships and history.

master-data-deduplication

Agents using this skill

AgentAxel

See how it helps

When is it useful?

Use when accounts, contacts or products accumulate duplicates across entry and import paths.

What does it help you do?

Normalizes identifiers and distinguishes exact matches from reviewable fuzzy candidates.
Previews conflicting fields, preserves dependencies and records merge lineage.
Validates imports before bounded, traceable application of accepted rows.

Try asking this

Prepare a customer duplicate review with field conflicts, linked invoices and a merge history that supports later investigation. Preview merging two similar companies with different billing details. Preserve invoice links and aliases, and send ambiguous identity matches to human review.

Detailed documentation

Read the skill’s original instructions, examples and checks in their authored language.

Details…Show less

Master Data & Deduplication

When to Use

Use when managing master data (Accounts, Contacts, Products/SKUs, Vendors), building duplicate detection and merge workflows, or implementing bulk CSV/XLSX/API data imports.

1. Master Data Normalization & Duplicate Detection

Prevent duplicates at the gate and detect fuzzy matches early:

Canonical Normalization:
- Normalize emails (lowercase, trim), phone numbers (E.164 format +36...), company tax/VAT numbers (strip spaces/hyphens), and website domains (strip www. and protocol) into indexed lookup columns.
Deterministic vs. Fuzzy Match Tiers:
- Exact Identity Match (same tenant_id + same normalized tax_id or email or sku_code) $\rightarrow$ block or link automatically according to domain rules.
- Candidate Fuzzy Match (similar legal company name Acme Ltd. vs. Acme Limited + same city/domain) $\rightarrow$ flag for review/merge rather than silently merging unrelated companies!

2. Safe, Auditable Record Merging (`Winner` $\leftarrow$ `Loser`)

Never "merge" two customer or contact records by running DELETE FROM accounts WHERE id = $loserId! Deleting or blindly overwriting loses orders, invoices, and external sync mappings:

Mandatory 5-Step Merge Transaction

Execute every merge atomically inside a database transaction:

Field Survivorship Resolution:
- Compute the merged attributes on the winner record using explicit survivorship rules (e.g., keep winner's verified tax_id and billing_address, fill blank fields on winner from loser, union tags, and allow the operator to preview/pick conflicting fields before confirming).
Re-Parent All Dependent Child Records:
- Re-point every foreign key referencing loser_id (contacts, opportunities, quotes, sales_orders, invoices, activities, tasks, files, external_entity_mappings) to winner_id.
Preserve Merge Lineage & Redirect Tombstone:
- Mark the loser record as status = 'merged', merged_into_id = winner.id, merged_at = NOW() (do not hard-delete the row!).
- If an external integration, bookmark, or old webhook references loser_id, resolve it via merged_into_id to winner_id.
Store Pre-Merge Snapshot for Unmerge / Audit:
- Record an immutable master_data_merge_logs entry containing the before snapshots of both winner and loser plus the list of re-parented child record IDs so the merge is 100% auditable and reversible if merged by mistake.

3. Two-Phase Bulk Import (Dry-Run Validation $\rightarrow$ Commit)

For CSV/XLSX or bulk data imports:

Phase 1 (Dry-Run Preview): Parse and validate every row against schema, UoM/currency codes, foreign keys, and duplicate rules, returning a structured row-by-row report (valid_new, valid_update, duplicate_warning, validation_error with exact row & column).
Phase 2 (Idempotent Commit): Import validated rows in bounded batches with an import_batch_id so the batch can be traced or rolled back cleanly.

Included files 1

SKILL.md