One of the most troublesome monthly tasks for small shop owners is paying suppliers.

Vegetable suppliers, meat vendors, rice dealers, and packaging suppliers all send invoices in different formats—some handwritten, some printed, and some sent as photos via LINE.

When there are many documents, it’s easy to encounter problems like:

  • The same batch charged twice.
  • Prices quietly increased without anyone noticing.
  • Invoice quantities that don’t match the goods actually received.

Can AI help review these first? Yes, but AI’s role is only to “spot suspicious items.” Whether to pay is still up to the owner.

This article uses a hypothetical scenario of a 4-person bento shop to break down a step-by-step process you can follow.

1. What challenges does this bento shop face?

Imagine a bento shop with 4 employees: the owner’s wife handles purchasing and accounting, there are 2 kitchen workers, and 1 front-of-house staff.

The shop regularly works with 5 suppliers, each invoicing monthly. At month-end, the owner’s wife reviews and pays all invoices at once.

Her problems include:

  • Having to flip through one invoice after another, often finishing very late.
  • Once, a supplier charged twice for goods delivered on the same day—discovered only two months later.
  • Vegetable prices fluctuate frequently, making it difficult to recall which price changes are normal and which are suspicious.

2. Why not just feed all invoices directly to AI?

Because invoices often contain sensitive data that shouldn’t be freely shared:

  • Supplier bank account numbers and business registration numbers.
  • The shop’s own bank account details.
  • Contact names and phone numbers.

These details don’t help with amount verification but pose a risk of data leaks.

So the first rule for this shop is: AI only looks at "what was purchased and for how much," not "where the money is sent."

3. Step one: Redact account and personal info first

The owner’s wife uses her phone’s built-in editing tools to block out bank account numbers, registration numbers, and phone numbers when taking pictures of the invoices.

Supplier names can be replaced with codes like “Vegetable Supplier A” or “Meat Supplier B.”

For reference, see SasaDaily’s guide on redacting personal info before using ChatGPT to organize receipts.

4. Step two: Have AI organize invoices into a standard list

Provide AI with the redacted photos and ask it to format each invoice uniformly:

Please organize these invoices into one list, with each line showing: supplier code, delivery date, item, quantity, unit, unit price, and subtotal. Mark “unclear” for any unreadable parts—don’t guess. Finally, calculate the total amount for each supplier.

That last instruction to “mark unclear instead of guessing” is important.

AI may misread handwritten or smudged figures, so it’s better if it honestly flags unclear parts.

5. Step three: Ask AI to compare and flag suspicious items

Next, give AI two additional data sources:

  • The shop’s stock receipt log (recorded daily when goods arrive).
  • The previous month’s organized invoice list.

Then ask it to compare:

Please compare this month’s invoice list, the shop’s stock receipt log, and last month’s invoice list. List any of the following: 1. Same supplier invoicing the same item on the same day more than once; 2. Items with unit prices more than 10% higher than last month; 3. Cases where invoiced quantity does not match the stock receipt log. For each, specify invoice and line number. Only list suspicious items—do not conclude who is at fault.

“Do not conclude who is at fault” reminds AI that its job is to identify issues, not judge them.

The 10% price increase threshold is specific to this shop; each business can adjust this limit as they see fit.

6. Step four: Owner confirms—this step is essential

After AI flags suspicious items, the owner’s wife reviews each one:

  • Cross-check the original invoice: AI might have misread numbers, so verify what’s really written.
  • Compare with the receipt log: Maybe the missing quantity was just an oversight during logging.
  • If there really is a problem, call the supplier: To clarify duplicate invoices, price hikes, or wrong shipments.

Often, suspicious items have reasonable explanations, like typhoon-related vegetable price surges or supplemental shipments by suppliers.

So, what AI highlights is a checklist for questions to ask, not proof that the supplier erred.

7. Step five: Final check of total amounts before payment

Once everything checks out, the owner’s wife manually adds up each supplier’s total on a calculator and compares it with AI’s calculation.

Payment itself is made entirely by the owner’s wife using online banking or at the bank counter—AI has no part in the payment process.

8. What do AI and people do in this process?

  • AI’s role: Standardize differently formatted invoices into one list, compare three datasets, and list suspicious entries.
  • Owner’s wife’s role: Redact sensitive info, record daily stock receipts, review suspicious items, communicate with suppliers, and make payment decisions.
  • Employees’ role: Count goods when receiving and assist with stock receipt logging.

In short: AI handles “finding,” while humans handle “judging” and “paying.”

9. What are the limitations of this process?

First, accurate stock receipt logging is crucial.

If received goods are not logged properly, even the best AI cannot detect discrepancies. The process relies heavily on the daily receipt logs.

Second, AI may misread handwriting.

Handwriting, blurry photos, or wrinkled invoices can cause misinterpretation by AI. That’s why the fourth step—to verify original documents—is essential.

Third, avoid making suppliers feel doubted.

When calling suppliers, use a tone like “wanting to confirm” rather than “you made a mistake.” Most issues are misunderstandings, and maintaining long-term partnerships is important.

10. How to get started?

  1. Pick the supplier with the most invoices and try the process for a month.
  2. Always record item names and quantities upon delivery—even if just on paper that you photograph.
  3. At month-end, use the provided instructions to organize and compare data, and check AI’s flagged items for accuracy.
  4. After one month, decide whether to expand the process to all suppliers.

Whether this saves time or catches issues depends on your shop’s real-world situation.

The most important takeaway today

Use AI to help you spot suspicious invoices—but whether to pay and how to discuss with suppliers is always your call.

If you want to find which step in your work is best to delegate to AI first, leave a comment saying “process.”

Recommended Reading