AI Expense Approval Workflow: Airtable + GPT-4 + Pinecone Auto-Audit

Airtable expense submission → n8n AI Agent analysis → structured output → Pinecone vector storage → write-back approval result

Who

Finance teams, CFOs, or ops staff who need automated expense approval and compliance auditing

You get

New expense submissions automatically trigger AI analysis, flag suspicious items, and leave a searchable vector audit trail

You need

n8n · airtable · pinecone

Steps

1. Configure Airtable Trigger

   Add an 'Airtable Trigger' node in n8n, select your base and Expenses table. Set the trigger field to Amount and authentication to airtableTokenApi. Once saved, the workflow auto-starts when a new or updated record with Status = Pending appears.

2. Configure AI Analysis Agent and Structured Output

   Add an 'AI Agent' node (CFO Expense Review Agent), connecting an OpenAI Chat Model and a Structured Output Parser. Enter the expense data template in the Agent's text field; set the CFO analysis role in systemMessage. Choose the model in the OpenAI node (default gpt-4o-mini). Provide a JSON Schema example in the Structured Output Parser to ensure outputs include decision, reason, amount, submitted_by, category, etc.

   Execution prompt · Source text。CFO Expense Review Agent node text / user prompt field

   Copy prompt

   =An employee submitted an expense:
   Amount: ${{ $json.fields.Amount }}
   Submitted by: {{ $json.fields['Submitted By'] }}
   Category: {{ $json.fields.Category }}
   Description: {{ $json.fields.Description }}
   Date Submitted: {{ $json.fields['Date Submitted'] }}
   Status: {{ $json.fields.Status }}

   Execution prompt · Source text。CFO Expense Review Agent node systemMessage field

   Copy prompt

   You are a CFO expense analysis agent. Flag suspicious expenses with the reason. When you answer try to give answer with all the given details

   Execution prompt · Source text。Parse CFO Agent Response node jsonSchemaExample field

   Copy prompt

   {
     "amount": 4500,
     "submitted_by": "Alice",
     "category": "Travel",
     "description": "Business class flight to Tokyo",
     "date_submitted": "2025-05-29",
     "status": "Pending",
     "decision": "Flagged",
     "reason": "The amount of $4500 for a business class flight appears unusually high and requires verification against standard travel policies and previous similar expenses. Additionally, the submission date is in the future (2025), which raises concerns about the legitimacy of the expense. It is advisable to confirm the travel plans and the necessity of the business class flight for this trip."
   }

3. Configure Vector Store Audit Chain

   Add nodes in sequence: Default Data Loader → Recursive Character Text Splitter → Embeddings OpenAI → Pinecone Vector Store. Feed the AI-generated reasoning text into the Data Loader. Set Text Splitter to chunkSize=100, chunkOverlap=20 (source defaults; adjust as needed). Use openAiApi credentials for Embeddings. Set Pinecone to insert mode with namespace as a dynamic value like {{ $json.output.decision }} and specify the target index.

4. Configure Airtable Write-Back and Closure

   Add an 'Airtable' update node with operation set to update. Use the record id from the trigger node as the matching field. Map Amount, Reason, Status (fixed value 'completed'), Category, decision, Description, and Submitted By from the CFO Expense Review Agent output. Save and activate the workflow.

You are setting up an n8n automated expense approval workflow. Please confirm you have: an Airtable base (with fields Amount, Submitted By, Category, Description, Date Submitted, Status), an OpenAI API key, and a Pinecone account with an index. Verify credentials before importing the template or rebuilding nodes step by step.