Skip to content
Case Study · Finance

Email Invoice Capture to Records & a Finance Dashboard

Supplier invoices arriving by email are now read, recorded, and pushed to a live dashboard automatically. Finance and other teams stopped re-keying attachments by hand.

Duration
6 weeks
Services
Workflow AutomationAI IntegrationAPI Integration
Background

A distribution business received hundreds of supplier and partner invoices as email attachments every week. A finance clerk opened each one, typed the vendor, amount, due date, and reference into a spreadsheet, then re-entered the same data into the accounting system. Other teams, procurement and operations, kept asking finance for numbers that were always a few days stale.

Challenges

  • Invoices arrived as mixed PDFs, images, and scans with no common layout.
  • The same data was typed twice: once into a spreadsheet, once into accounting.
  • Due dates were missed because nothing tracked them until month-end.
  • Every team needed the same figures, but only finance could produce them.

Objectives

  • Capture invoice data straight from the inbox, with no manual typing.
  • Record each invoice once, into both the ledger and a shared store.
  • Give finance and other teams a live dashboard instead of a spreadsheet.
  • Flag anything the system is unsure about for a quick human check.
Solution

We built an automation that watches a dedicated finance mailbox, extracts fields from each attachment with a document-AI step, and writes a normalized record. Confident invoices post to the accounting system and appear on a shared finance dashboard; anything low-confidence or duplicated lands in a review queue with the source attachment side by side. Other teams get read-only views scoped to what they need.

Implementation

  1. 01
    Field & mailbox mapping

    One week defining the exact fields each team needs and auditing a month of real invoices for the layouts the extractor must handle.

  2. 02
    Extraction & normalization

    A document-AI step reads each attachment; results are normalized against a vendor master, with confidence scores driving the review queue.

  3. 03
    Recording & dashboard

    Idempotent posting to the accounting API plus a live dashboard for finance, with scoped read-only views for procurement and operations.

  4. 04
    Parallel run & cutover

    Two weeks running alongside the manual process to calibrate extraction accuracy, then retiring the spreadsheet.

Architecture overview

  • n8n watching a dedicated finance mailbox
  • Document-AI extraction with per-field confidence scoring
  • PostgreSQL as the invoice store and vendor master
  • Accounting-system API for idempotent posting
  • Next.js dashboard with role-scoped, read-only team views
Business impact

What changed for the business.

  • Invoice entry time cut from minutes each to seconds of review.
  • Double data entry between spreadsheet and accounting eliminated.
  • Due dates surfaced on the dashboard instead of discovered late.
  • Procurement and operations self-serve the numbers they used to request.

Lessons learned

  • Extraction confidence, not a yes/no, is what makes an invoice pipeline safe. Route the unsure ones to a human.
  • A shared dashboard removed more work than the automation itself, because it killed the internal requests for figures.
  • Recording once, then fanning out to views, ended the disagreements about which number was right.

Facing a similar problem?

Tell us about your operation and we'll show you what the same delivery method would look like for you.