Finance: Accounts Payable and Invoice Processing
Learn how companies process invoices without errors, prevent fraud, and capture early payment discounts—using automation to replace manual data entry. ---
What Is This?
Accounts Payable (AP) is the process of receiving, validating, approving, and paying invoices from vendors. It's one of the most manual, error-prone processes in most companies—and one of the highest-impact to automate.
---
Why Does It Exist?
The business problem it solves:
A vendor sends an invoice. Without a process:
- The invoice arrives in an email and sits in someone's inbox
- Someone manually types the vendor name, amount, and invoice number into the accounting system
- A typo occurs: amount is entered as $5,000 instead of $500
- The error isn't caught until reconciliation (weeks later)
- The vendor calls asking why the payment is wrong
- The company has to issue a correcting check
- Time and money are wasted
Without AP process:
- Invoices get lost or paid late (vendors get angry)
- Data entry errors cause mismatches between what vendor claims they sent and what company thinks they received
- Duplicate invoices are paid (same vendor submits the invoice twice)
- Fraud: a hacked vendor email sends a new banking account; a check is sent to that account
- Early payment discounts are missed ("pay within 10 days, get 2% off" — but payment is made on day 40)
- Financial records don't match actual payments (reconciliation nightmare)
With AP process:
- Invoices are automatically read and categorized
- Matched against purchase orders and delivery receipts (3-way matching)
- Only approved invoices are paid
- Payments are scheduled strategically (capturing discounts)
- Bank account changes are verified before paying
- Records are always accurate
---
Real-Life Example
A manufacturing company's AP process, before and after:
Without process:
Vendor sends invoice. Email sits. A week later, accounting opens it. They manually look up the PO ("was this approved?"), find the delivery receipt, type the invoice amount into QuickBooks. A typo: $5,000 instead of $50,000. The invoice gets approved by a manager. A check is issued. The vendor receives $50,000 and gets confused (they only invoice for $5,000). The company is out $45,000. It takes a month to sort out.
With process:
Vendor sends invoice. System automatically:
- Reads the invoice (OCR technology)
- Extracts: vendor name, amount ($5,000), invoice number, line items
- Finds matching PO and delivery receipt in the system
- Checks: do the numbers match? (PO said $5,000, delivery receipt confirms $5,000, invoice says $5,000 ✓)
- Approves automatically
- Schedules payment for day 8 (vendor terms are "2/10 Net 30" — 2% discount if paid within 10 days)
- Sends payment
- Updates accounting records
- Vendor receives correct amount on correct day
Result: Same transaction, massively different outcomes. One is a disaster. The other is smooth, accurate, and even captures a discount.
---
Step-by-Step Workflow
Step 1: Invoice Arrives
A vendor sends an invoice via:
- Vendor portal (some vendors host portals where they upload invoices)
- Paper mail (less common now)
Step 2: Invoice Capture
The invoice is captured into the AP system. Some systems have:
- A dedicated email address (invoices@company.com) where vendors email
- A portal where vendors upload
- Manual entry for paper invoices
Step 3: OCR Processing
Document intelligence (OCR technology) reads the invoice and extracts:
- Vendor name
- Invoice number
- Invoice date
- Invoice amount
- Tax amount
- Line items (what was purchased, how many, unit price)
- Due date
This happens automatically; no human data entry.
Step 4: Matching Against PO and GRN
The system searches for:
- PO (Purchase Order): Did we authorize this purchase? Is the amount on the PO the same as the invoice?
- GRN (Goods Receipt Note): Did we actually receive what was on the PO?
Three-way match:
- PO says: "Buy 100 units at $50 = $5,000"
- GRN says: "Received 100 units on 3/15"
- Invoice says: "100 units × $50 = $5,000"
If all three match → auto-approve. If any don't match → flag for manual review.
This matching happens automatically; no human comparison.
Step 5: Discrepancy Investigation
If something doesn't match:
- Amount mismatch: invoice says $5,200 but PO says $5,000
- Reason: freight charge ($200) was added
- Solution: accept and approve (common scenario)
- Quantity mismatch: invoice is for 100 units but GRN only shows 50 received
- Reason: partial shipment
- Solution: research whether second shipment is coming, or contact vendor
- Invoice is a duplicate of one already paid
- Reason: vendor submitted twice by mistake
- Solution: reject and contact vendor
A human reviews and makes the decision.
Step 6: Approval Routing
If the amount is under a threshold (e.g., <$5,000), it auto-approves. If over threshold, it's routed for manager approval via:
- Automated Slack message: "Approve this $15,000 invoice?" with a button
- Or email with approval link
Manager clicks approve or deny.
Step 7: Payment Scheduling
Once approved, the system determines when to pay based on:
- Vendor terms: "Net 30" (pay within 30 days) vs. "2/10 Net 30" (2% discount if paid within 10 days)
- Available cash
- Strategic timing: if the term is "2/10 Net 30" and we have the cash, pay on day 8 to capture the discount
Payment scheduling is automatic; no manual decision-making.
Step 8: Disbursement
Payment is executed via:
- ACH (Automated Clearing House — electronic bank transfer) — most common
- Wire transfer (faster but more expensive)
- Check (rare, for vendors who don't have bank accounts)
Multiple payments are batched and sent together (instead of one payment at a time).
Step 9: Notification and Reconciliation
- Vendor is notified (payment confirmation email)
- Payment is recorded in the accounting system (ledger is updated)
- Bank statement is monitored to confirm payment cleared
- If anything is wrong (payment failed to go through), alert is sent
---
Where Time Gets Wasted (Common Bottlenecks)
Manual Invoice Entry
Every invoice is manually typed into the accounting system. Invoices have 5-20 line items; each requires separate entry. One invoice takes 15-30 minutes.
No OCR/document intelligence; invoices come in different formats (PDF from one vendor, Excel from another, email text from a third).
Invoice team has 2 people who do nothing but data entry. They process maybe 50-100 invoices per month = 1,250-3,000 minutes = 20-50 hours per month.
---
Manual PO and GRN Lookup
For each invoice, someone must:
- Find the corresponding PO (search through a system or file)
- Find the GRN (check another system)
- Compare all three documents to verify amounts match
This comparison takes 10-15 minutes per invoice.
Systems don't talk to each other; matching must be done manually.
Same as above — 20-50 hours per month.
---
Approval Bottlenecks
An invoice lands in a manager's inbox for approval. They're in a meeting. It sits for 3 days. Meanwhile, the vendor's payment due date approaches. By the time it's approved and processed, the company is paying late fees.
Email-based approval; no visibility into approval queue; manager doesn't prioritize it.
Late fees (1-5% penalty), lost vendor goodwill, potential payment terms penalties.
---
Duplicate Payments
A vendor accidentally submits the same invoice twice. Both are in the system at the same time. Both get approved and paid. Company sends double the payment.
No duplicate detection; system processes each invoice independently.
$50,000 double payment × interest cost + time to reverse = significant financial impact.
---
Fraud
A vendor's email account is hacked. The hacker sends an invoice with a new banking account. Company pays the "hacker's" bank account instead of the real vendor.
No verification of banking details; assumption that email is secure.
$10,000-$100,000+ stolen, difficult to recover.
---
What Can Be Automated?
Automation 1: OCR-Based Invoice Data Extraction
AI reads invoices (in any format: PDF, Excel, email, photo) and extracts structured data automatically.
Someone types vendor name, amount, invoice number, line items into the system.
Invoice uploaded → OCR reads it → data extracted and validated → ready for matching.
Example:
- Invoice PDF received: "For our consulting work in Q3, we invoice $15,000 including tax. Invoice #INV-2024-5678, due by 4/15/2024"
- System extracts: vendor name, amount ($15,000), invoice number, due date
- No human touches this data
Automation 2: Three-Way Matching
System automatically matches invoice against PO and GRN; flags discrepancies; auto-approves if all match within tolerance.
Someone compares three documents side-by-side, checking amounts, quantities, dates.
Invoice extracted → PO and GRN retrieved from accounting system → compared automatically → decision made (approve or flag).
Example:
- Invoice: 100 units × $50 = $5,000
- PO: 100 units × $50 = $5,000 (3% variance tolerance)
- GRN: 100 units received
- Match within tolerance → auto-approve
- No manual review needed
Automation 3: Automated Approval Routing
Invoices are routed for approval via interactive Slack messages instead of email.
Email sent to manager with invoice PDF attached. Manager has to download, review, reply.
Slack message sent: "Invoice from Adobe for $2,500. Approve or deny?" Manager clicks button. Done.
Automation 4: Payment Scheduling and Optimization
System analyzes vendor terms and available cash; schedules payments at optimal times to capture discounts.
Finance manually calculates: vendor terms are "2/10 Net 30". Due date is 3/20. If we pay by 3/10, we get 2% off. Finance manually decides when to pay.
System sees "2/10 Net 30" → checks cash position → if cash is available, schedules payment for day 8 → captures discount automatically.
Example:
- Vendor terms: 2% discount if paid within 10 days
- Invoice amount: $100,000
- Discount value: $2,000
- With this automation, the system captures $2,000 per invoice
- 50 invoices per month × $2,000 = $100,000 in annual savings
Automation 5: Duplicate Detection
System compares incoming invoices against already-paid invoices to detect duplicates.
Rare duplicates slip through and are paid twice.
New invoice arrives → system searches recent payments → if match found: "This looks like a duplicate of invoice INV-5678 paid on 3/10. Confirm or dispute?"
What AI Can Do
AI Opportunity 1: Cognitive GL Account Mapping
AI automatically assigns line items to the correct accounting categories (General Ledger accounts).
Someone reads "office supplies for Q3" and decides it belongs under "Operating Expenses > Office Supplies" (account 6200).
AI trained on past categorizations sees: "office supplies" → account 6200. "Consulting fees" → account 5100. "Travel" → account 5500. New invoices are categorized automatically.
AI Opportunity 2: Fraud Detection
AI analyzes historical invoices and detects anomalies (unusual amounts, new vendors, banking changes).
Finance team spots frauds reactively ("wait, did we really pay $50,000 to a new vendor we've never heard of?").
AI sees patterns in historical invoices: "Adobe invoices are always $50-200/month from this account. New invoice for $50,000 is a 250x increase and from a different account. Flag for verification."
AI Opportunity 3: Vendor Performance Insights
AI analyzes vendor invoices, payment records, and support tickets to generate quarterly vendor health scores.
No systematic vendor evaluation.
System tracks: invoice accuracy, payment terms compliance, support response times, SLA violations → generates score.
Beginner Project
Set up basic invoice capture and approval workflow.
Tools Required
- Accounting software: QuickBooks Online
- Automation platform: Zapier
- Email: Gmail
- Spreadsheet: Google Sheets
The setup:
- Create a dedicated email for invoices: invoices@yourcompany.com
- Create an "Invoice Tracking" Google Sheet with columns: date received, vendor name, amount, due date, status (draft, approved, paid)
- In Zapier: when email arrives at invoices@yourcompany.com → extract attachments → add row to Google Sheet
- When you manually mark "approved" in Sheet → send Slack notification to finance team to process payment
What you'll learn:
- How to capture invoices from email
- How to organize them in a central place
- How to notify people of new approvals
Success metrics:
- All invoices go to one place (no scattered emails)
- Nothing is forgotten (all tracked in Sheet)
- Approvals are visible (can see which invoices are waiting for approval)
---
What You'll Learn
- How to capture invoices from email
- How to organize them in a central place
- How to notify people of new approvals
Success Metrics
- All invoices go to one place (no scattered emails)
- Nothing is forgotten (all tracked in Sheet)
- Approvals are visible (can see which invoices are waiting for approval)
---
Step-by-Step Build Instructions
Advanced Project
Build a complete AP automation system with OCR, matching, approval routing, and payment optimization.
```
Invoice Arrives (Email/Portal)
↓
OCR Processing (Extract invoice data)
↓
PO & GRN Lookup
↓
Three-Way Match Check
├─ All match? → Auto-approve
├─ Minor variance? → Flag for review
└─ Major discrepancy? → Hold for investigation
↓
If Flagged: Manual Review
├─ AP specialist investigates
├─ Contacts vendor if needed
└─ Approves or rejects
↓
If Approved: Payment Calculation
├─ Check vendor terms (2/10 Net 30?)
├─ Check available cash
└─ Schedule optimal payment date
↓
Payment Execution
├─ ACH or wire transfer
├─ Update ledger
└─ Notify vendor
↓
Reconciliation
└─ Bank statement matches payment
```Tools Required
- Document intelligence: Tungsten Intelligent Document Processing or Amazon Textract
- AP automation: Tipalti or Bill.com
- Accounting software: NetSuite or QuickBooks
- Automation platform: Make or Workato
- Communication: Slack
- Analytics: Power BI or Looker (for reporting)
What You'll Learn
- Complex document processing
- Multi-system data matching
- Financial calculations (discount optimization, cash flow analysis)
- Fraud prevention
- End-to-end workflow automation
- Compliance and audit trails (important in finance)
Success Metrics
- 95%+ of invoices auto-approved (no manual intervention needed)
- Processing time per invoice: <5 minutes (vs. previous 30-45 minutes)
- Zero duplicate payments
- 100% of early payment discounts captured (~2-3% savings on total AP spend)
- Error rate: <0.1% (vs. previous 3-5%)
- Bank reconciliation: automatic (vs. previous 4-hour monthly manual process)
---
Step-by-Step Build Instructions
- Set up invoice capture:
- Create dedicated email: invoices@company.com
- Point invoice portal (if vendor uses one) to same place
- Configure system to process all incoming emails
- Configure OCR engine:
- Connect document intelligence tool to invoice capture
- Train it on your company's common invoice formats
- Test with 10 sample invoices to ensure extraction is accurate
- Build three-way matching logic:
- When invoice is captured → system queries accounting system for matching PO
- Query GRN system for delivery confirmation
- Create decision logic:
- If amount matches within 2% tolerance AND quantities match: auto-approve
- If amount within tolerance but quantity mismatch: flag for review (partial shipment?)
- If amount/quantity don't match: hold for investigation
- If invoice is duplicate of recent payment: reject and notify vendor
- Set up approval routing:
- For invoices flagged for review: send Slack message to AP specialist with all context
- Specialist can approve, deny, or request more info
- Approval updates the invoice status in the system
- Configure payment optimization:
- Define vendor terms in the system (Vendor A: Net 30, Vendor B: 2/10 Net 30, etc.)
- When invoice is approved, system calculates: "If terms are 2/10, pay on day 8 to save 2%. Can we afford that?"
- If yes: schedule payment for day 8
- If no: schedule for day 25 (before late fees kick in)
- Set up payment execution:
- Batch all approved invoices scheduled for today
- Execute ACH transfer (or multiple transfers if batching by vendor)
- Update accounting ledger automatically
- Send confirmation email to vendor with payment details and reference number
- Create reconciliation logic:
- Download bank statement daily
- Match payments to invoices automatically
- Flag any discrepancies (payment didn't go through, wrong amount sent, etc.)
- Alert finance team to investigate
- Build reporting:
- Daily: invoices processed, approved, awaiting review, payment scheduled
- Weekly: payment performance, discrepancy rate, avg processing time
- Monthly: total paid, discounts captured, duplicate prevention ($ saved)