Bot workflow examples: triggers, decisions and actions
Classify common automation claims without confusing a label with a capability.
FREE OFM FIELD KIT
Find the useful automation inside the sales pitch
Use these three maps to separate a fixed rule, an AI draft and a multi-device publishing workflow. The question is which action changes the account and who owns it.
Repeat for every distinct action bundled inside a product.
TRIGGER: schedule / account event / conversation / analysis
INPUT: approved data required
DECISION: fixed rule / generated output / human choice
ACTION: draft / message send / tag / report / publish
APPROVAL: who approves this specific action?
SUPPRESSION: what current state prevents it?
DOWNSTREAM: what other rule responds to its output?
RECOVERY: who can pause it and undo or correct its effect?
EVIDENCE: supported documentation and one recorded test
02
Audit a phone-rack proposal
Evaluate a concrete device workflow without assuming that more devices mean more organic reach.
CLAIM: [what the seller says the devices achieve]
ACTUAL TASK: filming / device QA / authorised publishing / another explicit task
WHY NATIVE TOOLS ARE INSUFFICIENT:
ONE-DEVICE PILOT AND SUCCESS CRITERION:
ACCOUNT AUTHORITY AND RECOVERY OWNER:
COST: devices + charging + connectivity + repairs + operator time
MEASURE: tasks completed, failures and paid time
REJECT AS EVIDENCE: screenshots of views without the content, period, account context and acquisition method
Avoid setups sold around fake engagement, impersonation or circumventing account restrictions.
03
Ask five questions before a demo ends
Ask for demonstrations or documentation instead of accepting ambiguous labels such as “AI powered”.
1. Show which part generates new text and which part is a fixed template.
2. Show exactly which actions execute without a person approving them.
3. Change the relevant fact after the trigger but before execution. What is rechecked?
4. Repeat the same event. What prevents a duplicate?
5. Pause and transfer ownership to a person. What happens to pending actions?
Record unknown answers as unknown. Classify a second workflow separately if the product performs another job.
These are illustrative planning scenarios, not reports of vendor tests or customer outcomes. Adapt the examples to an account you are authorised to operate. Use the field guide for the underlying method and the interactive tool for your own inputs.
Software assigns an internal category for an operator to inspect.
Work through it
Map it as analysis if the output stays a label. If the label triggers another action, document that second workflow separately.
What to check
Check who can correct an inaccurate category and what downstream actions depend on it.
02
THE SITUATION
A tool suggests a reply
The software prepares text and the operator decides whether to send it.
Work through it
Map generation and approval separately. Inspect what context the suggestion used and whether the operator can edit or discard it.
What to check
A suggestion must not silently become an autonomous send when a setting changes.
03
THE SITUATION
An event launches a sequence
A qualifying event schedules one or more predetermined actions.
Work through it
Map the event, delays, duplicate policy and cancellation conditions. Confirm the source of the event.
What to check
A stopped or changed event must have a defined effect on pending actions.
YOUR WORKING DOCUMENT
Keep the decision with the evidence.
Use the plain-text worksheet in your own approved system. It contains headings only, so you control which information is recorded and where it is stored. Nothing entered in a downloaded file is returned to this publication.
WORKFLOW CLASSIFICATION
Product feature / action name:
Trigger event:
Information read:
Generates new text?
Decision rule:
Human approval point:
Action taken:
Downstream effects:
Pause / cancellation:
Correction owner:
Platform and account-authority check:
Questions still unverified:
Common mistakes to avoid
Calling every scheduled message an AI conversation.
Assuming “assistant” means that each action requires human approval.
Using the classifier as a bot detector or a platform-permission checker.
Before implementing a workflow, confirm that it is supported by the chosen product and permitted under current platform rules. A planning example does not establish either. OnlyFans terms.