Data Architecture Decision Framework
Configure a reference architecture for your advancement data platform based on your environment, scale, and analytical priorities.
Environment profile
The tool bases its recommendations in all four decision areas on these inputs. Set them to match your current environment.
The framework recommends a pattern for each of four architecture decisions and generates an implementation plan.
Choose where your analytical data is stored and processed. Your choice sets which analytical workloads your team can run and how much work it takes to keep them running.
Anti-patterns
- Business logic in the bronze zone. If you transform data during ingestion, the code becomes hard to maintain and you cannot re-process the data when a transformation turns out wrong.
- Skipping the silver zone. If you go directly from raw data to consumption marts, each mart must re-implement cleaning and normalization logic on its own.
- Gold zone sprawl. The team builds a new mart for each dashboard request, and the copies of the business logic come to differ across dozens of nearly identical tables.
- Premature lakehouse adoption. For SQL-only teams focused on reporting, the added operational work outweighs the flexibility gained.
Choose how data moves from source systems to your analytical platform. The pattern you pick sets how fresh the data is, how much work the pipeline takes to run, and how well it copes when a source system changes.
Anti-patterns
- Polling CRM APIs on tight schedules. It uses up API quota, adds load on the source system, and still delivers older data than native CDC.
- Custom ETL without data contracts. Schema changes from source systems propagate silently and break downstream models without warning.
- iPaaS for analytical pipelines. iPaaS tools are designed for point-to-point operational sync, not warehouse-load patterns with medallion-zone modeling.
Choose how your team matches constituent records across systems and measures data quality.
Anti-patterns
- Batch dedup only, never at point of entry. Duplicates build up again between dedup runs, and no one fixes the root cause (no matching on record creation).
- Merging without unmerge capability. If a false positive merge combines two different constituents' giving histories, someone must reverse it in full.
- Defining "total raised" in the BI tool. If each dashboard implements its own sum of gift amounts with its own soft-credit filtering logic, the numbers will diverge. Define canonical metrics once in a shared modeling layer.
- Ignoring soft-credit double-counting. Household soft credits and matching-gift soft credits produce inflated totals unless the aggregation explicitly filters to hard credit or uses a soft-credit-aware calculation.
Decide how fresh each data domain must be and how your team monitors pipeline health. Set each freshness target to how often the people who use the data make decisions with it. Real-time infrastructure is worth its cost only when a reporting delay causes a measurable problem.
Anti-patterns
- Streaming infrastructure for batch-consumed data. If no one acts on the data faster than a batch run delivers it, streaming adds operational cost and no analytical gain.
- No freshness SLA per data domain. Set an explicit freshness expectation for each table so that monitoring has a threshold.
- CRM schema changes treated as anomalies. CRM administrators add custom fields and change picklist values routinely. Observability tools should classify these events, not fire false alarms.
- Skipping observability entirely. Without automated checks, your team learns of a pipeline failure only when someone opens a stale report. Run freshness and volume checks after every pipeline execution.
Reference architecture
The tool builds this from your profile and selections. The diagram and summary update as you change inputs.