Draft an SOP for a recurring small-business process
Document a real recurring workflow with roles, controls, exceptions, evidence, and handoff criteria
9 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
Draft an executable SOP from the supplied observed process and approved controls.
Trigger, purpose, frequency, roles, steps, tools, inputs, outputs, time, and variation:
[process]
Quality, approvals, access, safety, privacy, finance, exceptions, records, retention, and escalation:
[controls]
Users, skill, format, owner, effective date, review cycle, training, systems, and unverified items:
[use]
Separate observed current practice, approved rule, proposed improvement, and unresolved decision. Define purpose, scope and exclusions, trigger, owner and backup, prerequisites, inputs, system permissions, numbered actions, expected result and evidence at each control point, decision branches, handoffs with acceptance criteria, exception paths, stop conditions, escalation, records and retention, completion definition, and process measures. Use one actor and one action per step. Never invent a policy, role, approval, legal or safety requirement, system behavior, access right, duration, retention period, or exception. Put warnings before risky work and avoid exposing secrets or customer data. Add a quick checklist without replacing the detailed procedure, a test scenario for normal and exception paths, training notes, version owner, and change log. Flag any step that cannot yet be run safely.
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 process evidence
Small online shop packs paid orders each weekday at 15:00. Operations Assistant exports paid, unfulfilled orders from ShopDesk; verifies SKU and quantity; picker retrieves items; assistant scans SKU, packs, weighs parcel, buys carrier label, photographs sealed parcel, marks fulfilled, and places it in pickup cage. Typical 20 orders, 90 minutes. Owner covers absence. Inputs: order export, inventory, packaging. Outputs: labeled parcel and fulfillment record. Variations: missing stock, damaged item, address rejected, order change request.
Rules and controls
Only paid orders may ship. A second person checks orders over CNY 1,000. Customer address must not be copied outside ShopDesk/carrier portal. Missing or damaged stock stops packing and goes to Owner; no substitution without written customer approval. Address rejection goes to Customer Support. Order changes received before label purchase may be reviewed by Support; after label purchase require Owner decision. Keep parcel photo and carrier tracking in ShopDesk; retention period not yet decided. Daily parcel count must equal carrier manifest before cage handoff.
SOP use constraints
For Operations Assistant and trained owner backup. ShopDesk and carrier portal. Plain numbered SOP plus checklist. Owner is document owner. Effective date and quarterly review date not yet approved. New assistant shadows one complete run. System permissions are assigned outside the SOP. Do not invent menu paths.
Example output
SOP: weekday paid-order packing
Owner: Shop Owner. Status: draft; effective and quarterly review dates require approval. Scope: paid, unfulfilled physical orders processed at 15:00. Excludes unpaid orders, substitutions without written approval, and system-access provisioning.
Prerequisites: Operations Assistant or trained Owner backup; access already provisioned to ShopDesk and carrier portal; packaging available; pickup cage ready. Never copy customer addresses outside the two approved systems.
Operations Assistant exports the paid, unfulfilled order list at 15:00. Evidence: export timestamp and order count. Stop if payment status is unclear.
Assistant verifies SKU and quantity against each order. For value above CNY 1,000, a second person records completion of the required check; the eligible checker is not specified and must be approved before use.
Picker retrieves the listed units. Missing or damaged stock stops that order; record the exception and route it to Owner. Do not substitute unless Customer Support records written customer approval.
Assistant scans each SKU. Expected evidence: scanned SKU and quantity match the order. A mismatch returns to step 2.
Assistant packs and weighs the parcel, then buys the carrier label in the approved portal. An address rejection goes to Customer Support. A change received before label purchase may be reviewed by Support; after purchase, stop and request Owner decision.
Assistant seals and photographs the parcel without exposing unnecessary customer data, records photo and tracking in ShopDesk, then marks fulfilled. Retention remains unresolved; do not delete records or invent a period.
Assistant places the parcel in the pickup cage. Before handoff, compare physical parcel count with carrier manifest. Any difference stops handoff until reconciled.
Completion: counts match, all parcels have labels, tracking and photos are recorded, and exceptions remain outside the cage. Daily measures: orders received, packed, exception count by type, elapsed time, count mismatch. Quick checklist: paid; SKU/quantity; high-value check; scan match; safe address handling; label; photo/tracking; manifest count. Test one normal order plus missing-stock and post-label-change scenarios before approval.
Why this works
1
Control-point evidence makes the SOP auditable instead of a narrative description.
2
Exception and stop paths keep standardization from forcing unsafe improvisation.
Check the result
Can a trained backup execute every normal step from the supplied information?
Do control points have observable evidence and named authority?
Are exceptions, stop conditions, escalation, records, and unresolved decisions visible?
ChatGPT