First Decision Case¶
Reliable daily sales analytics for a growing retail business¶
The first Decision Case is deliberately narrow. It tests the complete decision method through one coherent business problem instead of presenting several unrelated technical labs.
Business context¶
A fictional retail business has outgrown spreadsheet-based daily reporting. Finance and regional operations need trustworthy previous-day sales figures before the start of the next business day.
Source files arrive from multiple regions. Some can be late, duplicated, invalid, or affected by schema change. The data team is small and has a constrained cloud budget.
Business outcome¶
Deliver complete and explainable daily sales data by the agreed reporting deadline, with visible failure and recovery behaviour and an attributable operating cost.
Core architecture question¶
What is the smallest data platform architecture that meets the reporting requirement and can evolve without forcing the team to operate unnecessary infrastructure?
Decisions the case must resolve¶
- Batch, micro-batch, or streaming
- Ingestion trigger and buffering
- Storage and table format
- Transformation execution model
- Query and consumption approach
- Duplicate, late-arriving, and invalid-data handling
- Failure detection and replay
- Minimum viable observability
- Cost guardrails
The daily reporting requirement suggests a managed batch-oriented baseline. That is a hypothesis to test, not a service selection made in advance.
Three modules¶
1. Frame and Decide¶
Define stakeholders, outcomes, constraints, workload characteristics, viable options, and the Architecture Decision Record before writing Terraform.
2. Build the Thin Slice¶
Implement only the components required to test the selected architecture's key claims. AWS is the first implementation environment and Terraform is the deployment method.
3. Validate and Review¶
Test normal processing, duplicate and invalid inputs, late arrival, one recoverable failure, operational visibility, and cost. Use the evidence to confirm or revise the decision.
Evidence the case must produce¶
- Reporting deadline met or missed
- Accepted and rejected records
- Duplicate handling
- Freshness and completeness
- Workflow success and failure
- Recovery or replay
- Estimated and observed cost
- Remaining production gaps
Current status¶
The case is in definition and user-validation. The business brief and decision method come before the complete technical implementation. This prevents the lab from becoming a technology demo looking for a problem.