Can AI read customer orders and quotes? Yes, straight into a table

Yes. Give an AI assistant such as ChatGPT, Claude or Gemini the order a customer sent (a PDF or Excel file attached to an email, or a photo or screenshot sent on LINE, the messaging app most Taiwanese businesses use), attach your part number lookup table and ask for a table. In minutes you'll have a list with part numbers, quantities and delivery dates laid out, ready to key into your ERP or paste into an import spreadsheet. The same works for a customer's request for quotation.

Every customer's order format is different, and entering them reliably takes experience. A seasoned assistant knows that "Customer A's part numbers need the prefix removed" and "when Customer B says a case, it means 12 units." This article isn't about replacing that. It's about letting AI speed up the reading and part number lookups: first, prompts you can copy; then what to check, how a lookup table helps, and what to do as volume grows.

Ask AI directly: 3 prompts to copy

Copy these and swap in your own columns, item numbers and dates (the ones below are examples).

Turn an order into a table

Attached are an order from a customer (PDF) and our part number lookup table (customer item number, internal part number, unit conversion). Please turn the order into a table: customer item number, internal part number, product name, quantity (converted to pcs), unit price, delivery date. If an item number isn't in the lookup table or something is hard to read, don't guess: explain why in a Notes column. Finally, add up the amounts and compare the total with the one on the order. Tell me if they don't match.

A handwritten photo or fax

This is a photo of a handwritten order a customer sent on LINE. Please read the product name, quantity and delivery date on each line and put them in a table. Mark anything you can't read clearly as "To confirm" instead of guessing. Then write a short LINE message asking the customer to confirm those lines.

A request for quotation

Attached are a customer's request for quotation and our quote history for this year (Excel). For each item, list the unit price we last quoted this customer and the date. List items we've never quoted separately, so I can prepare the quote.

After AI reads the order, check these 3 things

1. Quantities and totals: the converted quantities and the total should match the original document. Ask the AI to show the original figures side by side, then check them yourself.
2. Are the part numbers right? New item numbers and products with similar names are the easiest to get wrong, and every line in the Notes column needs a person's eyes. Accuracy drops noticeably for blurry or skewed photos and faxes, so it's better to ask the customer to send it again.
3. What isn't on the document: a price agreed by phone, this customer's special delivery terms, whether you have the stock. It doesn't know these, so check them before you create the order.

Orders include customer names and prices. For Excel orders, replace customer names with codes and delete columns you don't need before you upload. PDFs and photos are hard to edit, so use your company's enterprise plan for those.

Make it easier: your lookup table and AI working together

1. Write down each customer's conventions in a lookup table

Set up a shared Excel or Google Sheets file with one rule per row: customer, customer item number, internal part number, unit conversion (for example, "1 case = 12 pcs") and notes (for example, "remove the prefix from the part number"), plus a "usual quantity" column. Attach it every time you ask AI to read an order, and it will convert part numbers by the same rules and flag quantities well above usual. New hires can follow the table, too.

2. Collect orders in one place

Set up a shared order inbox or cloud folder, and save order photos received on LINE there too, named "Customer_Date." Add a simple log (time received, customer, who's handling it, status), so files are easy to find when you upload them to AI and everyone can see where each order stands.

3. Offer frequent customers an order template

For customers who are happy to use one, offer an Excel order template with the item as a Data Validation drop-down list, or collect orders through Google Forms, where responses flow into a spreadsheet automatically. With fixed columns, AI reads orders most reliably and the results are easy to work with.

When to consider the next step

AI plus a lookup table can serve you for a long time. It's worth looking at the next step when:

• You download attachments, upload them to AI and key the results into your ERP every day.
• You'd like AI to check orders against your ERP's customer and part master data, order history and quotes directly, without attaching them each time.
• You'd like your assistants to spend more time talking with customers and chasing delivery dates.
• You need a record for every order: what AI read, who confirmed it and what was changed.

The next step: connect AI to your order inbox and ERP

Once it's connected, AI reads each email, PDF, Excel file and LINE photo as it arrives, checks it against your existing data and drafts the order. This uses multimodal AI (models that understand images and text together): it reads the whole document, including the table structure and context, rather than turning the image into text the way traditional OCR (optical character recognition) does, which needs rules written for every layout.

Matching the data back to your master data

What really decides whether you can automate is mapping what was read to the master data in your system:

• Customer: the document says "Delta Precision," but in your system it's customer code C0231.
• Part number: the customer uses its own item number, XJ-220; internally, it's P-88120.
• Unit: the customer writes "3 cases," but your system needs "36 pcs."
• Price: prices on the document have to be checked against the quoted or contract price.

Your lookup tables, master data and past orders are what this step works from: the AI decides which item each line refers to, and the system checks whether the result makes sense.

Route by confidence

1. Create automatically: every field matches the master data, quantities and amounts are within a reasonable range, and the total adds up. At first, you can have a person confirm every order.
2. Human review: low-confidence fields, a new customer item number, or a quantity much larger than usual. What the AI read appears side by side with the original, and the assistant checks, corrects and confirms.
3. Send back or contact the customer: required information is missing (no delivery date, no item number), or the document contradicts itself.

Anything uncertain or that doesn't match is confirmed by a person before it's written to the ERP, and every step is logged. Track field accuracy, the share of orders created automatically, average review time, and the most important one, escaped errors: mistakes found after an order was created automatically. Set conservative thresholds at first, record every manual correction so it matches automatically next time, and spot-check a few orders every week, even when the system is confident.

The same approach works for requests for quotation, supplier invoices and statements (see reconciliation for wholesalers), shipping and customs documents (product descriptions, weights, HS codes), and business cards, application forms and inspection reports.

Rollout steps

1. Collect samples: 10–20 real documents from each major customer, including some hard-to-read ones.
2. Clean up your master data: customers, part numbers and unit conversions. Your lookup table carries straight over.
3. Define the checks: which cases must a person see? For example, amounts above a certain threshold, new customers and new part numbers.
4. Run in parallel for a while: the AI makes suggestions while people keep entering orders as usual; compare the two, and switch over once you're confident in the accuracy. The first phase typically goes live in about 4–6 weeks.

After rollout, your assistants' experience goes where it's needed most: checking exceptions, talking with customers and chasing delivery dates. To see whether your documents are a good fit for AI, start with a free AI Health Check, or take a look at our AI integration services and custom business systems. You're also welcome to send us a few sample documents.