Data architecture covers: model, quality, integration, retention, governance.
1. Data model.
- Standard objects first — use Account, Contact, Opportunity, Case unless they don't fit.
- Custom objects for genuinely custom entities.
- Relationships:
- Lookup for loose links (Contact -> Account).
- Master-Detail for tight couples (Quote Line Item -> Quote).
- Junction for many-to-many.
- External Id fields on every business record — stable foreign keys for integrations.
2. Naming & conventions.
- API name conventions: prefix custom fields by purpose, suffix by type.
- Picklist value sets: global value sets for shared lists (Country, State, Industry).
- Field standards: every field has Description, help text, owner.
3. Sharing & security architecture.
- OWD per object based on data sensitivity.
- Role hierarchy mirroring org chart.
- Sharing rules for cross-team access.
- FLS for sensitive fields.
- Encryption for Restricted / Mission Critical.
4. Data quality.
- Validation rules at every save.
- Duplicate Rules + Matching Rules on Lead/Contact/Account.
- Required fields and picklists instead of free text.
- Auto-generation where possible (Auto-Number, formula fields).
5. Integration architecture.
- System of record per object clearly defined.
- External Id fields for upserts.
- Mulesoft / iPaaS as middleware for complex flows.
- CDC / Pub/Sub API for real-time outbound.
- Bulk API for batch loads.
6. Data retention & archival.
- Per object: how long to keep?
- Big Objects for compliance archival.
- External warehousing (Snowflake / BigQuery) for analytics on historical data.
- Auto-delete or anonymise after expiry.
7. Governance.
- Data stewards per domain (Sales data, Service data, etc.).
- Field ownership — each field has an accountable owner.
- Change requests for new fields / objects.
- Data quality dashboards — completeness, dupes, age.
Common pitfalls:
- Schema sprawl — unchecked custom field creation.
- No External Id strategy — integrations create dupes.
- Wrong relationship choice — lookup where master-detail needed (or vice versa).
- No data retention policy — old data piles up indefinitely.
Senior architects own data architecture as a durable artefact — a diagram + supporting docs that govern decisions for years.
The most senior insight: data architecture decisions are expensive to change. OWD is irreversible (you can change it but at cost). Master-detail conversion is hard. Get it right early; resist the temptation to "we'll fix it later."