Connecting a product to another system is more than adding an API call. The team needs to agree which system owns the information, when it changes and what should happen if the connection cannot complete its work.
Identify the source of truth.
For each shared record, name the system that owns it and the systems allowed to update it. A customer record, a subscription and an access permission may be related without belonging to the same source.
Write down how identifiers are matched across systems. If that relationship is assumed rather than defined, a working demonstration can conceal problems that appear only when records are edited, duplicated or deleted.
Check access before promising a date.
Obtain the relevant documentation, test environment and permissions early. Confirm which capabilities are available on the actual account or subscription the product will use.
A prototype integration can answer a focused question before the full application is built around it. Keep provider restrictions and unanswered questions in the delivery plan, with someone responsible for resolving them.
Describe delayed and repeated work.
Decide how the product responds to a slow request, a rejected update or an event delivered more than once. The user-facing message and the server-side processing rules need to describe the same outcome.
Some work can be retried or processed later; other operations need additional checks before repetition. The appropriate behaviour depends on the action and the guarantees the connected systems provide.
Include the people who operate the connection.
Agree how a failed integration becomes visible, who investigates it and what information they need. Logs and administrative views are useful when they support an actual operating responsibility.
The handover should explain configuration, ownership and recovery steps within the agreed scope. External systems will change over time, so ongoing maintenance belongs in the product plan.
A dependable integration has clear ownership and a plan for the moments when the normal exchange does not complete.
Questions for the working session.
- Which system owns each record?
- Are the required account capabilities and test access confirmed?
- What happens to delayed or repeated operations?
- Who investigates a failure after launch?