Key takeaways
- A useful workflow map has an observable trigger and finish state, not just boxes connected by arrows.
- Every step names an owner or executing system, required input, observable output and exception destination.
- The mapper distinguishes fixed rules, AI judgment, human action and decisions so reviewers can challenge the operating boundary.
- The tool runs locally in your browser and exports Markdown without an account.
Map one workflow
Step 1
Step 2
Step 3
Step 4
Map check · 4 steps
Boundary is complete enough to review
The map has a trigger, finish state, owners, inputs, outputs and destinations for judgment exceptions. Review the sequence with the people who run it.
What this mapper records
Process maps are useful because they make the sequence and handoffs visible. AHRQ's process-mapping job aid begins by defining the start and end points, then identifying steps and decision points. This tool adds implementation fields that decorative flowcharts often omit: the actor, data contract, produced state and failure destination of every step.
The step types are deliberately plain. A trigger starts the run. A fixed rule is code with an inspectable condition. AI judgment handles bounded interpretation. Human action retains authority or physical work. A decision branches. A finish state proves the intended outcome exists. These labels are not a replacement for formal BPMN when your organization requires it; they are a fast review vocabulary.
| Field | Test | Weak answer |
|---|---|---|
| Action or state | Uses a specific verb or names an observable state. | Handle request. |
| Owner/system | Names who acts or which service executes. | The team. |
| Input | Names the record or fields consumed. | Customer data. |
| Output | Names the changed record, event or decision. | Processed. |
| Exception destination | Names a safe state, owner and route. | Ask someone. |
Map the real path before the proposed automation
- 01Set one start and one intended finishIf the map has several unrelated starts or finishes, split it into workflows or define the join rule.
- 02Walk a recent caseUse evidence from a real run to capture systems, fields, handoffs and rework that interviews forget.
- 03Add likely failuresFor each judgment or decision, name refusal, ambiguity, timeout and downstream rejection destinations.
- 04Challenge ownershipDistinguish the executing system from the person accountable for the process and the person authorized to approve effects.
- 05Copy and review the specificationAsk operators to annotate the exported Markdown before choosing tools or architecture.
From map to an automation candidate
A complete map is not automatically worth automating. Use the Automation Readiness Assessment to test recurrence, stability, authoritative inputs, owned exceptions and finish evidence. Then apply the risk assessment to each effect—send, change, grant, publish or pay—rather than scoring the workflow as one undifferentiated object.
If the starting material is a procedure, first use the translation ledger. One SOP sentence can contain several steps, and one workflow control can implement several sentences.
Limitations and when not to use this
- The mapper does not execute, import or validate a workflow against live systems. All fields are user-supplied.
- Its diagnostics check structural completeness, not correctness, feasibility, permissions, security or regulatory obligations.
- Inputs stay in the browser, but do not enter customer records, credentials or sensitive operational data into planning tools.
Sources
- Job Aid: Process Mapping — Agency for Healthcare Research and Quality Accessed 1 August 2026
- Business Process Model and Notation 2.0.2 — Object Management Group Accessed 1 August 2026
Assess readiness
Test whether the mapped process has stable, owned and observable operating boundaries.
Assess readinessUli Prantz
Builds and operates all-agents
Uli Prantz builds all-agents, the process-automation platform this site documents. He writes about the operational side of automating recurring business work: where deterministic code beats model judgment, where it does not, and where a human still has to approve.