Decision-Led Workflow Design
"If we did this — what should it look like?"
Architect workflows so every decision is explicit, owned, governed, and safe to automate. Available after a Stage 1 GO decision.
Most workflows are designed after the tool is chosen.
Vendors are selected before processes are understood. Platforms are configured before ownership is defined. Automations are built before failure modes are documented. The Design stage reverses this — producing a blueprint that is tool-agnostic, human-readable, and governance-ready before any build decision is made.
Five deliverables for every workflow.
Decision-gate map
A visual flowchart (Mermaid.js) of every step in the workflow, with explicit decision gates marked at every point where a human or AI must make a choice. Every gate has a named owner and defined criteria for proceeding.
RACI ownership matrix
For every decision gate: who is Responsible, who is Accountable, who is Consulted, and who is Informed. No ambiguity. No shared ownership of critical decisions.
Technical manifest (JSON)
A structured JSON document that defines the input schema, decision logic, owner, and failure mode for every node. This is what the Stage 3 Enablement Agent reads to deploy the governed digital worker.
Failure-mode documentation
For every automated step: what happens if the AI makes the wrong call? Who is notified? What authority do they have? What is the recovery path? This documentation is the difference between a robust system and a fragile one.
Automation boundary document
An explicit statement of which parts of the workflow are approved for AI involvement and which require human authority. This becomes the kill-switch logic for Stage 3 — a compliance officer can read it and understand exactly what the AI is permitted to do.
Collaborative, structured, and human-reviewed at every step.
Intake from Stage 1
The Architect Agent reads the Stage 1 friction map and Readiness Score. Every identified gap becomes a design constraint.
Decision-node identification
Every point where a human or AI makes a choice is mapped and classified — decision node or action node. Decision nodes require an owner. Action nodes require an input schema.
Ownership assignment
Every node is assigned to a named role, not a team. The RACI matrix is built collaboratively with the client — not assumed.
Failure mode design
For every decision gate, the failure path is designed alongside the success path. This is the most commonly skipped step in traditional workflow design — and the most important.
Blueprint review and STOP / ENABLE decision
The completed blueprint is reviewed by the client. An explicit ENABLE decision is required before Stage 3 proceeds. A STOP means the design needs refinement — not that the project has failed.