Data Architecture & Modelling · Cobalt Giraffe Insights
What can data architecture still change when the platform is already chosen?

The platform has been bought. The delivery team has a date to work towards, and the migration workbook is open. One column asks whether each customer is active. Sales, Billing and Support can each supply an answer. On the same day, for the same customer, those answers can differ.
Consider a fictional subscription business. A customer signs a contract in November, pays a setup charge in December and becomes entitled to support when the service starts in January. What should each team see on 15 December?
| Team | Decision | Expected answer |
|---|---|---|
| Sales | Has the customer signed? | Yes. The contract was signed in November. |
| Billing | Can we raise the setup charge? | Yes. The charge is due in December. |
| Support | Is the customer entitled to help today? | No. Support starts in January. |
The migration team could map a source field into the platform’s single Active flag and pass every technical reconciliation check. If Support then uses that flag to grant access, a successful load can still produce the wrong business decision.
Start with what people do with the data
Ask what someone will do differently when a field says yes. That takes the design conversation from comparing column names to agreeing business rules. A logical model can distinguish the contracting organisation, the account that pays, the service being supplied and the entitlement that governs access to support. Those distinctions matter when a parent company pays for services used by several subsidiaries, or when billing and support end on different dates.
Next, establish where each fact comes from and who owns its definition. Sales may be authoritative for contract signature, Billing for charge status and the service agreement for the dates that determine support eligibility. The model and mapping need to preserve the links between those facts.
Work within the platform’s constraints
If the platform can represent dated entitlements, map them explicitly. If it exposes only a customer-level flag, agree which rule the flag represents and how Support will obtain the entitlement decision. That might mean using an existing service or keeping entitlement records in a linked system. The delivery team can then assess the cost and risk of each design against the cutover date.
The practical outputs are a model that preserves the relevant relationships, a mapping that records definitions and ownership, and acceptance examples the business can check. Test the early signature case. Test a payer covering two subsidiaries. Test a customer whose billing ends before support does. Each example should tell a named team what answer it expects and why.
When time is tight, start with the meanings that affect money, access or obligations. A reporting label can often be fixed later. Recovering relationships lost during migration can mean going back to the original systems. Before signing off the load, ask a team to make one real decision from the migrated data and trace how it got its answer.