Case studyArchitecture, governance, and how to adapt this pattern in a pilot
Business use case
Problem
Transformation programs rely on multi-step processes, intake → triage → risk review → recommendation → approvals. In many organisations, these are spread across email and spreadsheets, and they break down when steps fail, ownership changes, or a long-running process stalls.
Who benefits
- PMO / transformation leads, consistent intake artifacts and faster triage
- Steering committee, auditable decision trail and clear recommendations
- Platform teams, a durable orchestration pattern reusable across processes
Success metrics
- Reduce intake-to-recommendation cycle from days to hours
- Make retries safe (no duplicate emails / duplicate tickets)
- Preserve workflow state through page reloads and transient failures
Solution
A durable workflow that orchestrates three steps:
- Parse an intake brief into themes and stakeholders
- Assess risk themes deterministically
- Draft a pilot recommendation
This example is intentionally deterministic so it is safe to run in a public portfolio. In production you would swap step logic for real integrations (ticketing, approvals, notifications).
Technical implementation
Stack
- Workflow DevKit (
workflow,@workflow/next) - Next.js API route triggers a workflow run and polls status
- Steps return explicit timing so orchestration is observable
Architecture
Durable orchestration: the browser starts a run, then polls until steps finish, survives retries and slow work.
Implementation highlights
"use workflow"defines durable orchestration boundaries"use step"isolates cached, retryable units (safe replays)- The UI renders a timeline of step execution so reviewers can see orchestration as a first-class output
Outcomes and learnings
- Durable orchestration is ideal for approval chains, intake processes, and long-running tasks
- Steps should be idempotent and return structured outputs
- Observability (step timing + state) is what makes orchestration usable in real programs
Delivery playbookDiscovery → pilot → scale
- 1Discovery2–4 wks
Identify the multi-step process that currently relies on spreadsheets/email (intake, approvals, escalations) and define a durable run boundary.
- 2Pilot6–8 wks
Implement 3 to 5 reliable steps with observable run status; validate retry behaviour on transient failures and reloads.
- 3Scaleongoing
Add human-in-the-loop hooks, SLAs, and durable event triggers; instrument run outcomes for program reporting.
Where else this appliesDurable steps matter whenever work spans minutes or days, retries are expected, and you cannot hold an HTTP request open.
Transformation intake
Parse briefs, score risks, and assemble steering-committee packs across multiple review cycles.
Vendor security questionnaires
Extract questions, map to control library answers, flag gaps for human edit, then export PDF.
Loan or claims adjudication prep
Gather documents, run policy checks, and pause for underwriter review before decision APIs fire.
M&A data-room summarisation
Fan out per-folder summaries with resumable runs when uploads trickle in over weeks.
Workflow DevKit in production gives you checkpointed runs without standing up Celery, Step Functions, or a always-on orchestrator for early pilots.