Document triggers, inputs, decisions, controls, exceptions, and handoffs before considering automation
7 min setupTested with:ChatGPTReviewed: 2026-08-28
1
Add your context
Your text stays in this browser. AILesson Prompts does not send it to a model or server.
2
Your prompt
Unfilled fields remain visible as placeholders, so you can still copy and edit the prompt
Turn the recurring work into a reusable, human-runnable process.
Trigger, frequency, purpose, people, current steps, tools, time, and outputs:
[work]
Input standards, decisions, exceptions, errors, approvals, sensitive data, and escalation:
[variation]
Quality, turnaround, ownership, audit, allowed tools, and improvement target:
[target]
Map the current process before proposing changes. Separate deterministic steps, judgment, approval, waiting, rework, and exceptions. Return: purpose and trigger; entry criteria; inputs and source of truth; numbered steps with owner, tool, output, control, and expected time only when supplied; decision table; exception/escalation paths; completion definition; handoff and audit record; compact checklist; measures for quality, time, and failure demand; and a small pilot. Identify duplication, bottlenecks, unsafe manual copying, and candidates for templating or automation, but do not automate judgment, approval, credentials, or sensitive-data movement without explicit authorization and controls.
Private by defaultPrompt assembly happens locally in your browser. Avoid placing confidential information into any AI service unless your organization allows it.
From input to outcome
A worked example
See how concrete context turns this recipe into a usable result
Actual input
Current recurring work
Every Monday a coordinator prepares the prior week's training attendance report for six department heads. They export CSVs from the learning system, copy rows into a workbook, remove test accounts, map department codes, calculate completion, email a PDF, and archive it. It takes 2–3 hours; only one coordinator knows all steps.
Rules and variation
EmployeeID is sensitive. The learning system is source of truth for attendance; HR directory is source for department. Duplicate enrollments occur. Test accounts have AccountType=Test. Completion means attended required minutes, but late corrections arrive through Wednesday. Department heads may dispute records. Only HR may change department mapping. The coordinator may correct obvious CSV formatting but not attendance facts.
Target process
Report by Monday 14:00 with totals reconciling to source exports and a visible late-correction note. A backup coordinator should run it. Use approved Excel and shared drive only; no new external automation. Keep export filenames and an exception log for audit. Pilot with one backup next week.
Example output
Trigger: Monday after the learning-system export is available. Entry criteria: prior-week date range confirmed; learning and HR exports saved read-only with timestamped filenames in the restricted shared folder.
Coordinator records filenames, extraction time, date range, and row counts in RunLog. 2. Import both files into separate Excel Tables; never paste EmployeeID into email/PDF. 3. filter AccountType=Test and record excluded count. 4. Identify duplicate enrollment keys; use the approved rule—currently missing—rather than deleting arbitrarily. 5. Look up department from the HR source; unmatched IDs go to ExceptionLog, not a guessed department. 6. Calculate completion from required minutes; reconcile included+test+exceptions to source rows and record duplicates separately. 7. Produce department summary and late-correction notice. 8. Second-person check verifies period, totals, exception count, and no EmployeeID in PDF. 9. Send PDF by 14:00 and archive PDF, workbook, RunLog, sources, and email evidence.
Decision table: department dispute→log and refer mapping to HR; attendance dispute→retain source fact and request learning-system correction; formatting-only issue→coordinator may normalize and record it; post-Monday corrections→reissue or Wednesday addendum requires owner decision. Pilot: backup runs one week from the checklist while primary observes; measure elapsed time, reconciliation difference, questions, rework, and exceptions. Blocker: approve duplicate rule and correction publication rule before claiming the process is complete.
Why this works
1
Mapping judgment and exceptions prevents a happy-path checklist from failing in ordinary cases.
2
Controls and source-of-truth fields make reuse safer than copying an experienced person's undocumented habits.
Check the result
Can a trained colleague run the process using the stated inputs, decisions, and exception paths?
Are approvals, judgment, sensitive data, and automation boundaries explicit?
Do completion evidence and measures reveal errors and rework?
ChatGPT