INVOICE PROCESSING

Agent-Driven Invoice Intake and Three-Way Reconciliation

An agent monitors the AP inbox, reads each incoming invoice email and attachment, locates the right PO and receipt.

CategoryInvoice Processing
Enginepaperclip
Difficultyadvanced
Triggerevent
Steps6
Setup~25 min

How it runs

The automated pipeline, trigger to output.

  • TriggerNew invoice email in AP mailboxGmailGmail
  • ActionExtract invoice and PO details from email and attachmentOpenAI
  • ActionQuery candidate POs and receipts in PostgresPostgreSQLPostgres
  • LogicClassify as match, partial, or dispute
  • ActionRecord decision and reasoning to PostgresPostgreSQLPostgres
  • OutputReply to vendor / notify AP with outcomeGmailGmail

What it does

This is an agent-led intake desk for messy, real-world invoices. It reads the email body and attached PDF, finds the purchase order even when the reference is informal, and works through the kind of ambiguity that breaks rigid matchers: split shipments, rounding, or units that differ between PO and invoice.

When to use it

Use this when invoices arrive as free-form emails and your matching rules keep dumping everything into manual review because vendors format things inconsistently. The agent absorbs that variability and only escalates genuine disputes.

How it works

  1. 1A new email arrives in the monitored Gmail AP mailbox, triggering the agent.
  2. 2The agent extracts invoice and PO details from the body and attachment using OpenAI.
  3. 3It queries Postgres for candidate POs and receipts and selects the best match.
  4. 4It reasons over quantity, price, and unit differences to classify the invoice as match, partial, or dispute.
  5. 5It records the decision and full reasoning trail to Postgres.
  6. 6It replies to the vendor or notifies AP with the outcome and any missing-document request.

Set it up

What you configure once, before turning it on.

  1. 1
    Connect GmailRead, draft, send, label.
  2. 2
    Connect OpenAIModels, embeddings, files.
  3. 3
    Connect PostgresAny Postgres URL — query, write, migrate.
  4. 4
    Set each agent's modelWe leave models unset so you pick the tier — fast + cheap, or top-quality.
  5. 5
    Tune it to your dataEdit the prompts, filters, and field mappings so it matches how your team works.
  6. 6
    Test, then turn it onRun once against a sample, confirm the output, then enable the trigger.

Run this workflow in your colony.

14-day trial. No DevOps. No Sales call. Provisioned in under a minute.