AI Invoice Processing: Stop Typing Supplier Bills by Hand

AI invoice processing reads each supplier bill, whether it arrives as a PDF, an email or a phone photo, pulls out the supplier, invoice number, dates, amounts and tax, suggests the account code from your past bills and creates a draft bill in your accounting software. You stop typing and start approving. Setup is usually an afternoon: a forwarding address plus supplier rules.

What people underestimate is the exceptions. Standard bills from regular suppliers go through almost untouched within a couple of weeks. The time goes on credit notes, statements mistaken for bills, multi-page scans and bills that belong to a specific job or vehicle. Plan for those from day one and the savings are real; ignore them and you swap typing for fixing.

Follow me on Instagram@sagnikteaches

What happens between the email arriving and the draft bill

Every tool on the market follows roughly the same seven steps. Knowing them helps you see where something went wrong when it does.

Connect on LinkedInSagnik Bhattacharya

1. Capture. Bills reach the tool by a dedicated email address you forward to (or give suppliers directly), by upload, by a phone photo, or by the tool fetching them from supplier portals. Paper bills can be scanned with the OneDrive or Google Drive mobile app; Microsoft Lens, which many people used for this, was retired in early 2026.
2. Read. A digital PDF already contains its text. A scan or photo is just a picture, so the tool first runs optical character recognition (OCR, software that turns an image of text into text) to get words out of it.
3. Classify. The AI decides what the document is: an invoice, a credit note, a statement, a receipt, a remittance advice or a quote. This step matters more than it sounds, because each needs different handling.
4. Extract. It pulls out the fields: supplier name, invoice number, invoice date, due date, net amount, tax, total, currency and, if you've asked for them, each line item. Modern tools use AI models that understand the layout, so a new supplier's format doesn't need a template.
5. Code. It suggests the account (materials, repairs, fuel), the tax rate and any tracking category, mainly from how you coded that supplier's previous bills.
6. Check. It tests whether net plus tax equals the total, whether the invoice number has been seen before, whether the date is plausible and, in better tools, whether the supplier's bank details have changed.
7. Publish. It creates a draft bill in your ledger with the original document attached. You approve it, and it joins the queue for payment.

An illustrative extracted record for one bill looks like this before you approve it:

Subscribe on YouTube@codingliquids

The supplier name here is invented for the illustration. The column worth reading every time is the last one: it tells you why the tool thinks it's right, and a "new supplier" or "doesn't add up" note is your cue to open the PDF.

Header-only or line items: decide before you start

Most capture tools extract the header by default: one line per bill with the total. That's enough when a bill goes to one account. Line-item extraction reads every row of the bill, so you can split one merchant invoice across several accounts or jobs, or track the price of a specific part over time.

Line items cost more (in Dext they're a paid add-on, and in most tools they're slower to review) and they produce more to check. Choose them only if at least one of these is true:

• One bill regularly covers several jobs, vehicles or properties that you cost separately.
• You want to spot price creep on specific items from a regular supplier.
• You carry stock and need quantities, not just values.

A cleaning supplies bill that always goes to "cleaning" doesn't need line items. A parts supplier bill that covers four different cars in a dealership probably does.

The tools, from built-in to bolt-on

Start with what your ledger already includes. Most small businesses never need more. If you're choosing between the dedicated tools, Dext against Hubdoc compares them on time saved.

A used-car dealership's supplier bills, before and after

Consider an illustrative independent dealership selling about 30 cars a month with two sales staff, a workshop technician and one administrator.

Before. Around 260 supplier bills arrive a month: auction purchase invoices, transport and collection charges, parts, valeting and cosmetic repair, inspections, the dealer management system, listing site fees, utilities and fuel. The administrator types each into the ledger at about four minutes a bill, roughly 17 hours a month, and writes the stock number on vehicle-related bills so each car's preparation cost can be totalled before pricing it.

Setup. The dealership asks its main suppliers to email bills to the capture address and to put the stock number in their reference field. It turns on line items only for the parts supplier, whose bills cover several cars. It writes supplier rules (below) and agrees one hard rule: any vehicle-related bill without a stock number goes to an "unassigned" list the administrator clears daily.

After a month. Review drops to about a minute per bill, plus around two hours of exceptions a month, so roughly 6.5 hours in total instead of 17. The dealership's 260 bills sit just over Dext's 250-document allowance, so it checks how extra documents are priced before choosing it over the ledger's built-in capture.

What went wrong in week one. Auction invoices listed the buyer's premium and the vehicle price on separate lines, and header-only capture coded the whole total to vehicle stock. The premium belongs in stock too for this dealer, so that was fine, but the auction's separate transport charge on the same bill was a cost the dealer wanted tracked separately. The fix was a supplier rule splitting auction bills into two lines.

Six places extraction goes wrong, and how each shows up

Credit notes read as invoices

The tool extracts $120 and books it as a bill instead of a credit, so you'd pay a supplier who owes you. It shows up as a supplier balance that's higher than their statement. Check anything with "credit" or a minus sign on it.

Statements treated as bills

A monthly statement lists bills you've already entered. If it's captured as a bill, everything on it is duplicated. Good tools classify statements separately; still, forward statements to a different address or folder so they never enter the bill queue.

Supplier name variants

"Northgate Motor Factors", "Northgate MF Ltd" and "NMF Parts" become three contacts, and your spend reports split three ways. Merge them once and set the rule so future bills map to one contact.

Dates in two formats

A bill dated 03/04 from a supplier using a different date convention gets read as the wrong month. It shows up as a bill due months early or months late. Look for due dates that don't match that supplier's usual terms.

Tax lines on mixed bills

Bills with items at different tax rates, or a tax line that's printed but blurred on a scan, are where the extracted tax figure is most often wrong. Check the "net plus tax equals total" test result, and keep scans clear: flat, well lit, all four corners visible.

Bank details that change

This is less an extraction error than a fraud route. A genuine-looking bill with new bank details gets captured perfectly, and that's the problem. Any change in bank details should stop the bill until someone phones the supplier on a number from your records, never the one on the bill. Catching duplicate invoices and payment fraud covers the controls.

Supplier rules that make coding right first time

Rules tell the tool what to do with a supplier's bills without asking. Ten minutes on them saves hours. A filled-in set for the dealership:

"Auto-approve" here means the bill is published without a person opening it. Only switch that on for suppliers whose bills have come through clean for at least a month, and never for suppliers where bank details could plausibly change.

Getting suppliers to send bills the right way

Half the exceptions disappear when suppliers send bills to the right place with the right reference. A short email to your top 15 suppliers does more than any setting. Illustrative wording:

Subject: Where to send our invoices from 1 October

Hello,

From 1 October, please email all invoices and credit notes for our account
to bills@[your capture address] instead of to individual staff.

Please put our stock number (format SN-0000) in the "Your reference" field
on any invoice for work on a specific vehicle. If one invoice covers several
vehicles, please show the stock number on each line.

Statements can still go to accounts@[your domain].

Thank you, this helps us pay you on time.

The last line isn't just politeness. Suppliers respond faster when the change is framed as getting paid sooner, and it's true: bills that arrive in the right place with a reference get approved days earlier than ones sitting in a salesperson's inbox.

The same setup in a property maintenance firm

The pattern changes a little when bills belong to jobs rather than vehicles. An illustrative maintenance firm with eight operatives buys materials from two merchants on trade accounts, often several times a day. Each counter purchase produces a paper ticket and, later, an emailed invoice. Operatives are told to write the job number on the ticket and photograph it in the capture app before leaving the counter. The emailed invoice arrives a day or two later, and the tool matches it to the photo, so the job number carries across.

The failure here is different: operatives forget the photo on busy days, and the invoice arrives with no job. The firm's fix was a daily list, sent by the automation to each operative at 4pm, of that day's merchant invoices still missing a job number. Replies come back as one-word answers ("J-2217") that the office types in. Low-tech, but it keeps the job costing accurate enough to quote the next job properly.

A five-minute daily routine for whoever approves bills

1. Open the review queue and sort by the tool's confidence or "needs review" flag.
2. Approve the clean bills from known suppliers in bulk.
3. Open every bill flagged new supplier, total mismatch, possible duplicate or changed bank details.
4. Clear the "unassigned" list: bills missing a job, vehicle or property reference.
5. Check the credit notes separately so none has slipped through as a bill.

Done daily it rarely takes more than five or ten minutes. Left for a fortnight, the same work takes an hour and the context for each odd bill has gone.

Proving it works: the 50-bill audit

After three or four weeks, pick 50 approved bills at random and check each against its PDF. Score every field that matters: supplier, invoice number, date, net, tax, total, account, and reference where you need one.

Illustrative results from the dealership's audit:

50 bills, 8 fields each = 400 fields checked
Supplier          50/50
Invoice number    49/50  (one read "INV-2O4418" with a letter O)
Invoice date      50/50
Net / tax / total 148/150 (one blurred scan: net and tax misread, total right)
Account           47/50  (three new suppliers coded to General expenses)
Stock number      46/50  (four missing: suppliers not yet using the ref box)
Overall           390/400 = 97.5% of fields correct
Bills with any error: 9 of 50

Two things in that result matter more than the headline percentage. First, 9 bills in 50 had an error somewhere, which is why a human still glances at every draft. Second, the errors cluster: new suppliers and missing references. That tells you where to spend effort (supplier rules and a note to suppliers), rather than on changing tools. Repeat the audit after a supplier change or a software update.

Doing it without new software

If you get 20 or 30 bills a month and no ledger capture, a chat assistant on a business plan can turn a batch of PDFs into a table you paste into a spreadsheet or import. It's slower than a proper tool, but it beats typing.

You are extracting data from supplier invoices. For each attached PDF, return
one row with these columns, in this order:
file_name | document_type (invoice / credit note / statement / other) |
supplier | invoice_number | invoice_date (YYYY-MM-DD) | due_date (YYYY-MM-DD) |
net | tax | total | currency | our_reference | notes

Rules:
- Copy invoice numbers exactly as printed.
- If a field isn't on the document, write MISSING. Never estimate it.
- If net + tax doesn't equal total, write CHECK TOTAL in notes.
- If it's a credit note, show amounts as negative numbers.
Return a table only.

Illustrative output for three files:

file_name | document_type | supplier | invoice_number | invoice_date | due_date | net | tax | total | currency | our_reference | notes
bill_0912.pdf | invoice | Northgate Motor Factors | INV-204418 | 2026-09-12 | 2026-10-12 | 386.00 | 77.20 | 463.20 | USD | SN-1043 |
bill_0913.pdf | credit note | Northgate Motor Factors | CN-3371 | 2026-09-13 | MISSING | -45.00 | -9.00 | -54.00 | USD | SN-1038 |
scan_0915.jpg | invoice | Bright Valet Services | 0877 | 2026-09-15 | MISSING | 240.00 | 48.00 | 280.00 | USD | MISSING | CHECK TOTAL

What you'd fix: the third row's total doesn't add up (240 plus 48 is 288), so open the scan, which is the point of the CHECK TOTAL rule. The missing stock number means chasing the valeter. And the missing due dates need filling from your usual terms with each supplier. Keep the prompt saved so every batch comes back in the same columns, and don't paste anything with bank account numbers into a plan that trains on your data.

Once bills flow in cleanly, the next jobs are matching them against orders and getting approvals and payments in order; matching supplier invoices to purchase orders automatically is the natural next step, and the same extraction ideas apply to other paperwork in AI document extraction against traditional OCR.

Further reads

• What Finance Tasks Can AI Automate in a Small Business? — Where bill capture fits among fifteen finance jobs.
• AI Accounts Payable: Approvals and Payment Runs Without Chaos — What happens after capture: approvals and payment runs.
• Can AI Sort and File Incoming Documents for Your Business? — Sort every incoming document, not just bills, before filing.
• How to Stop Retyping Data Between Apps With AI Automation — Stop retyping between other apps, not just the ledger.
• AI Document Processing: Turn PDFs and Scans Into Usable Data — The same approach for forms, contracts and scans.
• How to Set Spending Rules and Approvals for Business Purchases — Approval limits to pair with automated bill capture.
• Go Paperless Before You Add AI: A Step-by-Step Plan — Which paper to kill, which to scan and which to leave in the cabinet, so AI tools have clean files to work from. Six steps, about eight weeks.
• AI Use Cases by Department for Small Businesses (With Examples) — 21 practical AI use cases across seven departments, each shown in a small insurance brokerage, with how to start and what to watch.
• Why Is AI Bad at Maths? What to Check in Quotes and Invoices — Why chat assistants get sums wrong, the eight errors that turn up in quotes and invoices, a 10-point check, and a set-up where AI never does the arithmetic.
• AI Receipt Capture and Bank Categorisation: A Bookkeeper's Setup — Set up receipt capture and bank coding in six stages per client, with supplier rules, ledger AI suggestions and a weekly review that catches duplicates.
• Can AI Read Emailed Orders Into a Wholesaler's System? — When AI can reliably turn emailed orders into sales orders for a wholesaler, what it depends on, and three different wholesalers' answers.
• How to Speed Up Month-End Close in a Small Business With AI — A day-by-day close calendar, four AI prompts with sample outputs, and a farm shop that went from a nine-day close to four without skipping checks.
• How to Turn Handwritten Forms Into Spreadsheet Data With AI — Convert paper forms into checked spreadsheet rows while preserving unclear handwriting, original references and leading zeros.
• AI Supplier Management: Track Prices, Lead Times, and Risk — A supplier register AI can keep up to date: price histories from invoices, measured lead times, delivery performance and a simple risk score, reviewed monthly.
• AI Expense Management: Receipts, Mileage, and Approvals — How AI reads receipts, logs mileage and approves in-policy claims, with an estate agency's month of expenses and a policy you can adapt.
• Xero vs QuickBooks AI: Which Saves More Bookkeeping Time? — A costed month in both products, the plan-by-plan AI features at list price, two wrong matches to learn from, and a two-week trial scorecard.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Xero announcements on smart document capture (July 2026); Xero Central on Hubdoc; QuickBooks Online help on Accounting AI; Dext business pricing page (checked September 2026).