Draft a new process from a description or revise a map by conversation, then apply, configure, save and approve the result.
Product location: /qms/processes
Use AI to turn a description of work into a process people can execute. Explain the trigger, desired outcome, responsible roles, decisions, evidence and exceptions. Identify any existing procedure, form or requirement that defines the work.
For example: “Our supplier review starts when the annual review is due. Procurement gathers performance data, Quality reviews significant findings, and the supplier owner follows up actions. Help identify the decision points and records we need.”
Use that description to draft a new process, or to revise an existing one. The process editor holds the actual map and configuration.
Product location: /qms/processes. Choose New process, keep Describe it, AI drafts it selected and choose Draft with AI.
Open Processes and choose New process. The dialog opens on Describe it, AI drafts it.
In What does this process do?, describe the steps, who does them and the decision points, in up to 5,000 characters. Name the trigger and every outcome, including what happens when a check fails.
Optionally, under Link clauses from (optional), select the enabled standards whose requirement clauses the steps should reference.
Choose Draft with AI.
The assistant builds a swimlane map, checks that it is valid and saves it as a new Draft process. The process opens with an AI-drafted badge and Not yet submitted for approval. Its name comes from the draft, and its code is derived from that name. Clause references on the steps become the process's requirement links.
A draft is a starting point. Open Edit, then check each lane, step, decision label and connection against the real work. Bind the lanes to roles or people and configure completion before submitting it for approval.
If the assistant cannot produce a valid map, the dialog asks you to describe the steps and roles more concretely. Add the missing roles, decisions or outcomes and try again. If you draft many processes in a short time, wait a moment before the next attempt.
To build the map yourself, choose Start from scratch instead, enter a Process name and Code, and choose Create draft. The editor opens with an empty map.
Product location: /qms/processes/{processId}/edit (fallback: /qms/processes). Choose Edit with AI above the map builder.
Open the process and choose Edit. On Edit process, choose Edit with AI above the map builder. The button is available to people who can manage processes.
The Edit with AI panel starts a new conversation each time you open it, based on the map currently shown in the editor, including unsaved changes. The assistant opens the conversation. Inspect the current working map before requesting a revision.
Describe a focused change, for example: “Add a decision after the evidence review. Satisfactory evidence goes to completion; missing evidence returns to the existing collection step. Keep the current roles and retained steps.” Press Enter to send, or Shift+Enter for a new line.
Provide the reason for the change. If the request is ambiguous, the assistant asks one clarifying question before it proposes a map. Keep revisions small enough to review in the map preview.
Product location: /qms/processes/{processId}/edit (fallback: /qms/processes). In Edit with AI, review Proposed change, then choose Apply to editor or Dismiss.
Each proposal appears as Proposed change, with a short summary and a read-only preview of the complete revised map. Check the lanes, step names, connections, decision labels and requirement references. Follow every path from Start to an End, including any rework loop.
Choose Apply to editor when the proposal matches the intended work. The conversation shows ✓ Applied to the editor. and the editor and Live preview now hold the revised map. Nothing is saved yet: you can keep editing the draft, close the panel, or discard the changes by leaving without saving. Choose Dismiss when you want a different revision.
If the proposal needs adjustment, explain the change using the actual step or lane names. For example: “Keep the same evidence-collection step for rework, and require a reason on the return path.” After applying, check that the steps you kept still have their completion bindings, gates and clause references.
Product location: /qms/processes/{processId}/edit (fallback: /qms/processes)
Review Who does each lane and bind the actual organisation roles or people. A lane named Procurement describes a responsibility; its bindings determine who can be proposed for a run.
Inspect each task's completion settings. Choose manual completion or the appropriate supported binding to a form, stakeholder questionnaire, risk assessment, document approval, training course or task. Add the required note, evidence or signature gates and due-date settings.
Ask AI to help explain a configuration choice, then verify it in the editor. Use process configuration for the field-level reference. A revised drawing and a configured operational workflow require both map and settings review.
Product location: /qms/processes/{processId} (fallback: /qms/processes). After Save, check Definition state and choose Submit for approval.
Choose Save. The process saves, confirms Process saved. and opens its detail page. Inspect the persisted map and related settings there.
On an approved process, Definition state still shows the approved revision, such as Approved v1, and Submit for approval is available for the saved changes. New runs keep using the approved revision until the new one is approved.
Submit the saved definition for approval when ready. The reviewer checks the map, responsibilities, clauses and completion behavior, then approves it or requests changes. See definition approval for reapproval and existing-run behavior.
Applying an AI proposal updates the working draft. Saving records that draft. Definition approval determines the revision used by new runs.
Product location: /qms/processes/{processId}/edit (fallback: /qms/processes)
Select Offer AI "Guide me" on this step in a step's settings where participants would benefit from help. Use its guidance text to explain the purpose, expected evidence and useful questions for that step.
During a run, participants choose Guide me on that step to open the guidance conversation, then complete the actual task. For example, a participant can ask what evidence to check before recording a document-review conclusion.
Inspect the bound record and completion gates. Advice supports the participant's work; the run still records its actual actions, evidence and completion. See executing process runs.
Product location: /qms/processes/{processId}/edit (fallback: /qms/processes)
Execute a controlled example and note repeated manual work, unclear handoffs and missing information. Ask AI for targeted improvements using those observations.
For example: “The reviewer repeatedly copies the form's recommendation into a task. Help identify an output rule and the process completion binding that would make the handoff explicit. Show the configuration decisions I must make.”
Review the form rule, process binding and destination record together. Use connected workflows for patterns involving stakeholders, risks, controls and Vault approvals.
