Expense Approval: n8n Human-in-the-Loop Workflow Blueprint

Template prompt that converts a requirements brief into a deployable, node-by-node n8n workflow spec (OCR â†’ routing â†’ approvals â†’ escalation) with mapping tables, retries, dedup...

What You'll Get

Full node-by-node n8n workflow spec (node type, activation, credentials, full settings, expressions, inputâ†’output shapes, and explicit downstream error paths).

Webhook schema and idempotency header/field with example request/response to use as trigger.

OCR integration call spec with example HTTPS request, auth scope, and response-to-field mappings (two provider options with trade-offs).

Explicit Rule Engine / Router decision table listing each rule, condition expressions, routes and fallback.

Human approval step templates (email/Slack/web callback), expected back-channel payloads, and mapping to workflow state.

Escalation flow with SLA timers, escalation triggers, targets and example messages.

Field mapping tables (trigger â†’ OCR â†’ parsed expense â†’ accounting/CRM) with data types, required/optional flags and validations.

Retries/timeouts, node-level retry policies, exponential backoff settings, dedupe/idempotency strategy and merge policy for duplicates.

Error-handling catalogue with per-category handlers and a manual-review queue fallback node list.

At least three test payloads (valid, duplicate, low-quality image), plus example assertions for unit tests.

Implementation notes covering credential storage, rate limits, PII handling and audit logging best practices.

How to Use

1Fill the prompt inputs: trigger_event, apps_list (OCR, notification, accounting), auth_constraints, data_schema, escalation_policy. If any input is missing the template will return required fields and examples before generating the blueprint.

2Paste the completed prompt into your LLM and request the n8n blueprint output in Markdown. Review the 'missing fields' step if prompted.

3Validate credentials scopes and secret storage in n8n before implementing nodes; create credential objects rather than embedding secrets.

4Import or manually create nodes in n8n following the node-by-node spec; run the provided test payloads and assertions against each node and the end-to-end flow.

5Iterate rules and mappings using the included Field Mapping Tables and test cases; adjust retry thresholds and SLA timers to match operational policy.

Practical tweak to reduce duplicates and costs

Use a combined idempotency key formula (e.g., `sha256(source + '_' + source_id + '_' + submitted_at + '_' + image_phash)`) and a fuzzy image hash (pHash) threshold to dedupe before calling OCR; only call OCR after a quick dedupe check to reduce API calls and costs.

Works Best With

This prompt produces best results when you provide:

A complete JSON schema for the incoming webhook payload (required fields and example values).

A precise apps_list including exact target systems (e.g., Xero, QuickBooks, SAP) so field mappings can be concrete.

Auth constraints (allowed credential types and minimal scopes) to enforce fail-fast behavior on auth errors.

An explicit escalation policy (retry counts, SLA windows, escalation channels) to avoid ambiguous fallbacks.

Access to test endpoints or mock services for OCR and accounting systems to validate each node during implementation.

Prompt Template

## Goal
- Automate an expense approval human-in-the-loop workflow: receipt ingestion â†’ OCR â†’ rule-based routing â†’ approval actions â†’ escalation; include auditable mapping, idempotency, deduplication, retries, and fallbacks.

## Inputs
- Trigger: {{trigger_event}}  
- Apps/integrations: {{apps_list}}  
- Auth constraints and credential scope: {{auth_constraints}}  
- Incoming payload schema / important fields: {{data_schema}}  
- Escalation and SLA policy: {{escalation_policy}}  
- Failure policy: specify retry limits, escalation thresholds, and manual-queue rules (if not provided, the generated spec must request them)
- Idempotency key strategy and dedupe key (if not provided, the generated spec must propose a safe default)

If any of the above are missing or underspecified, the prompt below will instruct the model to return a concise list of required fields and example values before generating the workflow.

## Prompt
You are an expert n8n workflow designer. Produce a complete, operator-grade n8n workflow blueprint (node-by-node) that implements an expense approval human-in-the-loop pipeline. Use the following inputs: trigger = {{trigger_event}}, integrations = {{apps_list}}, auth constraints = {{auth_constraints}}, incoming payload schema = {{data_schema}}, escalation policy = {{escalation_policy}}. If any input above is missing, list the missing fields and give a short example for each, then stop.

Output a single copy-paste-ready specification that contains the sections listed in "Output Format". The specification must be concrete and actionable for an n8n implementation and must follow these requirements:

- Produce an ordered Node List for n8n. For each node include:
  - node type and exact n8n node name
  - trigger or activation condition
  - credentials required and minimal permission scope
  - full node settings and parameters (expressions where needed) â€” show example expressions for dynamic fields
  - input => output data shape (example JSON snippets)
  - explicit transformation steps (no abstract "transform" node without details)
  - downstream error handling path(s)
- Provide the Webhook schema or API payload expected at the trigger (content-type, example body) and the exact idempotency header/field to honour.
- Provide the OCR integration call spec (endpoint, payload, auth, response fragment mapping). If OCR is an external API, include an example HTTPS request payload and the expected response fields used downstream.
- Provide the Rule Engine / Router logic as an explicit Decision Node plan: list each rule, condition expression, target route, and fallback/default route.
- Provide the Human Approval step details: notification method(s), message template(s), approval actions (approve/reject/ask-for-more-info), expected back-channel payloads, and how the approval outcome maps back into the workflow.
- Provide Escalation flow: explicit triggers for escalation, escalation target(s), SLA timers, and example messages.
- Include Field Mapping Table(s) showing exact field-to-field mapping from trigger -> OCR -> parsed expense -> CRM/ERP/Accounting system, including data types, required/optional flag, and validation rules.
- Include Retries, Timeouts, and Dedupe/Idempotency strategy:
  - node-level retry counts, backoff, and conditions to retry vs escalate
  - global idempotency key generation formula (explicit expression using incoming fields)
  - dedupe algorithm for duplicate receipts (fuzzy image hash + fields), and merge policy
- Include Error Handling rules and Fallback Steps:
  - list error categories and per-category handler (retry, route to manual-queue, notify admin, mark as failed)
  - on OCR failures, include a fallback to manual OCR review queue with a specific n8n node list
- Provide Test Payload(s): at least three JSON test payloads covering (1) valid receipt, (2) duplicate receipt, (3) malformed/low-quality image triggering OCR fallback.
- Provide Example API/Webhook response/assertions that can be used to unit test each node.
- Provide a short Implementation Notes section calling out credential storage, rate limiting, PII handling, and audit logging requirements.

Formatting and delivery rules:
- Produce the output in Markdown and structure it exactly as required in "Output Format".
- Use tables for mapping and test payloads when helpful.
- Use inline code-style for expressions and field names.
- When referencing any external API endpoints include an example HTTPS request (method, path, headers, body) and the minimal expected response fragment.
- If any claim requires a choice (e.g., which OCR provider), present 2 practical options and explain trade-offs in one sentence each.

Start by validating the provided inputs. If valid, produce the full workflow spec now.

## Output Format
Output:
1) Overview  
2) Node list (ordered)  
3) Field mapping table(s)  
4) Retries/timeouts + dedupe strategy  
5) Test payload(s)

Ensure each numbered section is present and labeled exactly as above.

## Guardrails
- No â€œmagic nodesâ€; every transformation and validation must be explicit and described.
- Enforce idempotency and dedupe at the trigger and at each downstream write.
- Require explicit credential scopes and secure storage for secrets; do not embed secrets in the spec.
- Fail-fast on auth errors, retry on transient network errors with exponential backoff, escalate after configured retry limit.
- Validate image MIME type and size at the webhook; reject and respond 400 for invalid formats.
- Keep PII handling auditable: log only allowed fields, mask sensitive fields in logs, and store full receipt images behind secured storage with access logs.
- Provide a manual-queue fallback for any step that exceeds retry limits.
- Avoid polling when webhooks are available; prefer push-first architecture.
- Limit number of external calls per receipt to reduce costs; batch where safe.

## Examples
Mini node list excerpt (format example):
- Webhook Trigger (n8n Webhook)
  - Activation: incoming POST to /webhook/expenses
  - Expected body: `{ "source":"mobile_app", "employee_id":"E123", "receipt_image_url":"https://...", "submitted_at":"2026-01-01T12:00:00Z" }`
  - Idempotency: header `Idempotency-Key` or `idempotency_id` computed as `sha256(source + "_" + source_id + "_" + submitted_at)`
  - On receive: validate `receipt_image_url`, respond 202 with `processing_id`
- HTTP Request (OCR)
  - Call: POST https://api.ocrprovider.com/v1/parse
  - Payload: `{ "image_url": {{$node["Webhook"].json["receipt_image_url"]}} }`
  - Map: `ocr.total_amount` -> `expense.amount`, `ocr.vendor` -> `expense.vendor`, `ocr.date` -> `expense.date`
- IF / Switch (Rule Router)
  - Rule: `expense.amount > 1000` -> route to `Manager Approval Path`
  - Else -> `Auto-Approve Small Expenses` or `Regular Approval Path`
- Human Approval (Email / Slack / In-app)
  - Notify manager, include `Approve` and `Request More Info` links that call back to `/webhook/expenses/approval`

Use this template prompt as a single copy-paste input to your LLM to generate the full n8n workflow blueprint.

Variables

Usage Notes

Not for

â€¢ Final production deployment in highly regulated payment/financial systems without a formal security and compliance review (this template produces an operator-grade spec, not a certification).

â€¢ Cases where the platform is not n8n or where only vendor-proprietary nodes are allowed (this is n8n-focused).

â€¢ Replacing a security architect â€” credential scope and PII handling require org-specific policies.

Do not include

â€¢ Do not embed real API keys, passwords, or PII in the prompt or generated spec; use credential placeholders and secure storage.

â€¢ Avoid platform-specific private/custom node code â€” the spec references standard n8n node types and explicit HTTP calls rather than proprietary binary blobs.

â€¢ Do not include unverifiable performance claims, cost estimates, or SLA guarantees â€” those must be decided by the operator.

Expense Approval: n8n Human-in-the-Loop Workflow Blueprint

Generate a complete, operator-grade n8n workflow blueprint for receipt ingestion â†’ OCR â†’ rule routing â†’ approvals â†’ escalation, including mappings, auth, retries, dedupe and test payloads.

Turn this into a working asset in minutes.

Import into Workspace to auto-fill variables, adapt output style, and iterate with OctiAI's prompt editor.

Auto-fill & personalize from your inputs

AI-assisted edits: tone, structure, constraints

Attach context so outputs stay consistent

Iterate and save versions

What you'll get

Output format

Markdown

Best used when

When working on template or detailed.

Avoid when

Final production deployment in highly regulated payment/financial systems without a formal security and compliance review (this template produces an operator-grade spec, not a certification).

Made for

What to fill in

Use the guidance below to complete each placeholder with the right level of detail.

Trigger Event