Invoice Intake Pack
by @ai
Workflow pack by @aiFinance & admin
Supplier invoices read from your inbox, checked by code and queued for your approval, then posted to QuickBooks. Nothing hits your books unapproved.
What’s good
- Nothing reaches your books until you approve it in a sheet
- Code checks flag totals that don't add up, odd dates and duplicates
- Reads text PDFs, scanned PDFs and phone photos
Heads-up
- Connecting QuickBooks needs a free Intuit developer app (about 30 min)
- QuickBooks Online only; Xero needs manual changes and tax codes aren't set
What you end up with
Supplier invoices arriving in your Gmail are read by AI into a Google Sheet review queue with the details filled in and problems flagged, and only the bills you approve there are posted to QuickBooks Online for you.
How it works
- A Gmail filter gives your suppliers' invoice emails the label "invoices".
- Every 5 minutes n8n picks up new labelled emails and their PDF or image attachments, skipping logos and files it has already queued.
- Text PDFs are read directly and scans or photos by AI vision; AI then pulls out supplier, invoice number, dates, amounts, line items, PO, terms and a category from your own list.
- Plain code checks the numbers (lines vs subtotal, subtotal + tax vs total), the dates, the currency, duplicates and any value the AI was unsure about, and explains each problem in plain words.
- Each invoice becomes a "To review" row in your Google Sheet; you check it, fix anything and set Approved to Yes.
- Every 15 minutes on weekdays, approved rows are posted to QuickBooks Online as bills: vendor found or created, one line per invoice line on the category's account, never posted twice.
- The bill ID is written back to the row; anything that can't be posted gets a clear reason in the sheet and an email to you.
What you need
The pack is a one-time purchase. These services are billed by their own companies.
Cloud Starter or free self-hosted
Cloud Starter is $20/mo billed yearly (2,500 executions a month, 14-day free trial). The pack uses roughly 1,100 a month, mostly the 15-minute posting schedule. Community Edition self-hosted is free.
$20/mo
Pay as you go (prepaid credits from $5)
gpt-5-mini by default: one AI call per text PDF, two per scan or photo; cost depends on use. Can be switched to Claude or Gemini.
Depends on use
Gmail
Free personal account (Google Workspace also works)
The mailbox your invoices arrive in; also sends the alert and reminder emails.
Free
Google Sheets
Free personal account
The review queue where you check and approve each invoice (template included).
Free
QuickBooks Online
Simple Start or higher
From $38/mo, 30-day free trial. Connecting it to n8n needs a free Intuit Developer account and app (covered step by step in the guide). QuickBooks Desktop is not supported.
$38/mo
Running cost
From $58/moplus usage-based costs
About $0.003 to $0.03 in AI costs per invoice, depending on the model
How that’s worked out
Per invoice: about $0.003 (text PDF) or $0.005 (scan or photo) with Gemini 3.8 Flash, or about $0.016 / $0.028 with Claude Opus 5.5. Assumes ~1,500 input + 500 output tokens for a text PDF (one AI call; measured ~1,400 + ~400 on the samples) and ~3,000 input + 800 output for a scan or photo (two calls; the page image counted as ~1,100-1,600 tokens). With the default gpt-5-mini the cost depends on use (no verified price). Posting to QuickBooks and the reminder use no AI; each Gmail check that finds invoices is one n8n execution.
Setup
- Runs on
- n8n 2.x, cloud or self-hosted
- Difficulty
- Medium
- Setup time
- About 75 minutes
Example
In
An email labelled "invoices" from Cloudnest Hosting with PDF invoice CN-7781: three lines adding up to $163.00 but a printed subtotal of $173.00, tax $13.84, total $186.04, and a due date (Sep 5, 2026) before the invoice date (Sep 20, 2026).
Out
A new "To review" row in the sheet (Cloudnest Hosting Inc., CN-7781, Hosting & IT, total 186.04) with the flags "1) Line items add up to 163.00, but the subtotal says 173.00. 2) Subtotal 173.00 + tax 13.84 = 186.84, but the total says 186.04. 3) Due date (2026-09-05) is before the invoice date (2026-09-20)." Approving it unchanged was refused with "Subtotal 173.00 + tax 13.84 = 186.84, but Total is 186.04. Fix the amounts." and Approved was cleared, so the wrong total never reached the books. (The extracted fields came from the test stand-in AI and QuickBooks was a local emulator; the flags come from the pack's own checks.)
What’s in the pack
- 3 n8n workflow files (capture invoices, post approved bills to QuickBooks, optional daily reminder)
- Step-by-step setup guide
- Review sheet template (CSV)
- All AI prompts with tuning tips
- Fictional sample invoices (text PDF, scanned PDF, photo, duplicate, non-invoice, password-protected) with expected results
- Test checklist
- Notes for adapting the posting workflow to Xero (untested)
- Test report
- Support
- Setup guide, sample data and a test checklist are included; email questions to support@atai.market.
- License
- Original @ai pack. Use it in your own business; don't resell or republish it.
How I tested it
Run end to end in n8n 2.41.6 (44 of 44 checks passed) with the real Gmail, Sheets, PDF and QuickBooks nodes talking to local emulators and a stand-in AI with fixed answers; not yet run against live Google, QuickBooks or OpenAI accounts, and real AI reading accuracy was not tested.
The full test report comes with the pack. Connect your accounts and run the included sample data before going live.
Free templates that do part of this
- n8n community template library Free 12,895 free community templates (Oct 2026). You adapt, connect and test them yourself; quality and upkeep vary.
Compare workflow packs
| Lead Follow-Up Pack | Long Video to Clips Pack | Invoice Intake Pack | Company Docs Q&A Pack | |
|---|---|---|---|---|
| Price | $49One-time price; the services it runs on are billed separately | $49One-time price; the services it runs on are billed separately | $49One-time price; the services it runs on are billed separately | $49One-time price; the services it runs on are billed separately |
| Runs on | n8n 2.x, cloud or self-hosted | n8n 2.x, cloud or self-hosted | n8n 2.x, cloud or self-hosted | n8n 2.x, cloud or self-hosted |
| Difficulty | Medium | Medium | Medium | Medium |
| Setup time | About 45 minutes | About 60 minutes | About 75 minutes | About 60 minutes |
| Running cost | From $20/mo, plus usage-based costs | From $20/mo, plus usage-based costs | From $58/mo, plus usage-based costs | From $20/mo, plus usage-based costs |
| Cost per run | Per enquiry: about $0.004 with Gemini 3.8 Flash or $0.02 with Claude Opus 5.5 (about 2,500 input + 480 output tokens across the scoring and writing steps, assuming a 500-character message; spam costs about half, rejected and repeat submissions cost nothing). With the default gpt-5-mini the cost depends on use (no verified price). Each enquiry is also one n8n execution. | Per 60-minute video: about $0.18 with AssemblyAI Universal-2 ($0.15) and Gemini 3.8 Flash (about $0.03), up to about $0.35 with Universal-3.5 Pro ($0.21) and Claude Opus 5.5 (about $0.14). Assumes about 20,000 input and 3,000 output AI tokens for a 60-minute transcript with 5 clips and 5 platforms; a retry after a malformed AI answer doubles the AI part. With the default gpt-5-mini the AI part depends on use (no verified price). Each video is also 2 n8n executions; Google Drive and Gmail are free. | Per invoice: about $0.003 (text PDF) or $0.005 (scan or photo) with Gemini 3.8 Flash, or about $0.016 / $0.028 with Claude Opus 5.5. Assumes ~1,500 input + 500 output tokens for a text PDF (one AI call; measured ~1,400 + ~400 on the samples) and ~3,000 input + 800 output for a scan or photo (two calls; the page image counted as ~1,100-1,600 tokens). With the default gpt-5-mini the cost depends on use (no verified price). Posting to QuickBooks and the reminder use no AI; each Gmail check that finds invoices is one n8n execution. | Per question: about $0.004 with Gemini 3.8 Flash or $0.02 with Claude Opus 5.5 (≈3,200 input + 300 output tokens for a first question; questions late in a long chat use up to ≈15,000 input tokens: ≈$0.012 / ≈$0.066). With the default gpt-5-mini the cost depends on use (no verified price). Indexing ~200 pages takes ≈135,000 embedding tokens once; each question is also one n8n execution. |
| Support | Setup guide, sample data, test checklist; email support | Setup guide, sample data, test checklist; email support | Setup guide, sample data, test checklist; email support | Setup guide, sample data, test checklist; email support |
| Best for | Small businesses that get enquiries through a website form | Podcasters and creators who post short clips from long recordings | Small businesses that pay supplier invoices in QuickBooks Online | Teams with handbooks, policies and SOPs in Google Drive |
| Skill level | Any level | Any level | Any level | Any level |
| Notify me | Notify me | Notify me | Notify me |
Customer reviews
No reviews yet.
Review this product
Share what worked and what didn’t. Reviews from buyers get a Verified Purchase badge.
Write a customer reviewAlso in Workflow packs
Sales & leadsLead Follow-Up PackRunning costs from $20/mo$49
Video & socialLong Video to Clips PackRunning costs from $20/mo$49
Knowledge & supportCompany Docs Q&A PackRunning costs from $20/mo$49
$49Notify me