Five Teams, One Scheduling API, and a Clinical Safety Rule

Start with a value-stream map of the scheduling API and the end-to-end booking process. The 32% blocked stories are a symptom: teams are slicing by component, and the platform team's 11-week backlog is a queue. Analyse the interfaces and shared services, then change the slicing.

Actions:

• Build a lightweight interface catalogue and API contract tests so teams can integrate against a stable contract.
• Re-slice stories vertically around clinical scenarios such as book, reschedule, and cancel rather than by layer.
• Use a visible dependency board and a weekly cross-team sync focused on upcoming dependencies, not status.
• Treat the platform team as an enabling team: pair with product teams and provide self-service capabilities.
• Build a walking skeleton early to prove the integration path.

Reasoning: dependencies are created by architecture and by late, component-level analysis. The analysis response is to make interfaces explicit, slice end-to-end, and shift platform work from bespoke requests to reusable services. This reduces queue time without stopping delivery; it changes what gets delivered first.

Non-functional requirements are analysis items, not appendices. Bring them into the backlog as constraints and acceptance criteria with measurable thresholds. Run a quality-attribute workshop with clinicians, security, and operations to identify the scenarios that matter: audit logging for every booking change, two-second response at peak, availability during clinic hours, retention periods, and access control.

Then:

• Add a non-functional acceptance criterion to every relevant story, for example audit record written within 200 ms.
• Put them in the definition of done so no increment is done without them.
• Create a small set of enabler stories for cross-cutting work, with explicit value and risk reduction.
• Trace each to the clinical safety rule it protects.

Reasoning: audit logging and response time cannot be retrofitted cheaply, and clinical safety rules require evidence in every increment. Making non-functional requirements visible in the backlog also helps the steering group see why 'just add the screen' estimates are unrealistic.

Do contextual inquiry: observe clinicians booking and rescheduling real appointments, including the workarounds they use when the pharmacist review is required. Build or refine a journey map and a process model for the safety-critical path. Use personas for receptionist, nurse, pharmacist, and scheduler, because their needs differ.

Do this now, not in a phase. Bring two clinicians into the next sprint review and use low-fidelity prototypes before committing to build. Rewrite acceptance criteria in clinical language with examples: given an anticoagulant flag, when the scheduler selects a procedure slot, then the booking is held and a pharmacist task is created. Use story mapping to sequence around real clinical scenarios.

Reasoning: the mismatch is a discovery failure, not a UI defect. Agile analysis treats user research as continuous and collaborative. The cost of finding this in a demo is a fraction of finding it after rollout, especially where clinical safety is involved.