How do I match company card statement lines to receipts and code them to jobs and cost codes with AI before month-end?

Short answer

Use Claude or ChatGPT, then Copilot in Excel or Gemini in Google Sheets: Claude or ChatGPT reads receipt photos into a table with amounts copied exactly and job notes captured; Copilot or Gemini matches them to the card statement by amount and date window, proposes job and cost code from the receipt note and your rules, and builds an exception list per cardholder. Field card purchases are where job cost goes stale: receipts arrive late or never, the job number is scribbled on the back, and someone in accounting codes a stack of fuel and hardware charges from memory at month-end. Have Claude or ChatGPT read the receipt photos into a table, have Copilot in Excel or Gemini in Google Sheets match them to the card statement export by card, date and amount, and propose a job and cost code from the cardholder, the job number written on the receipt and the coding rules you keep. Everything it cannot match or code with confidence goes on an exception list for the cardholder or project manager. Posting stays in your accounting system and with you. A person still checks: a sample of matched lines against the receipts, every exception with the cardholder or project manager, and allowability, billing, tax and approval questions with the controller before anything is posted.

AI tie-in

Tool: Claude or ChatGPT, then Copilot in Excel or Gemini in Google Sheets

Claude or ChatGPT reads receipt photos into a table with amounts copied exactly and job notes captured; Copilot or Gemini matches them to the card statement by amount and date window, proposes job and cost code from the receipt note and your rules, and builds an exception list per cardholder.

Prompt 1

Read these receipts [upload photos or PDFs]. Build a table with one row per receipt: file name, merchant, date, subtotal, tax, total, any job number or note handwritten or printed on it, and a short description of the items. Copy amounts exactly. Write UNREADABLE for anything you cannot read and NOT SHOWN for anything missing. Do not guess a job number and do not assign cost codes.

Prompt 2

Here are the headers and a few sample rows of my Statement, Receipts, Rules and Jobs tabs [paste]. Settings!B1 is my date window in days. Write Excel formulas (or a step-by-step Copilot plan) that match statement lines to receipts by exact amount within the window, propose a job from a valid job number on the receipt or else the card default, propose a cost code from my keyword rules, record the source, and flag NO RECEIPT, NO STATEMENT LINE, POSSIBLE DUPLICATE, NOTE CONFLICTS WITH DEFAULT and NO CODE. Use only my data and rules.

Human check: A sample of matched lines against the receipts, every exception with the cardholder or project manager, and allowability, billing, tax and approval questions with the controller before anything is posted.

The problem

Match company card statement lines to receipts and code them to jobs and cost codes with AI before month-end. This is a common, time-consuming task for a construction accountant.

What this workflow does

Field card purchases are where job cost goes stale: receipts arrive late or never, the job number is scribbled on the back, and someone in accounting codes a stack of fuel and hardware charges from memory at month-end. Have Claude or ChatGPT read the receipt photos into a table, have Copilot in Excel or Gemini in Google Sheets match them to the card statement export by card, date and amount, and propose a job and cost code from the cardholder, the job number written on the receipt and the coding rules you keep. Everything it cannot match or code with confidence goes on an exception list for the cardholder or project manager. Posting stays in your accounting system and with you.

How it works: AI drafts or analyzes. A person operates the tools, checks the work, and takes action.

Your checkpoint: The workbook proposes coding from your own rules and what is written on the receipt. It does not decide whether a purchase is allowable, billable to an owner, taxable or properly approved. Every coded line is reviewed before posting, exceptions are resolved with the cardholder, and policy, tax and audit questions stay with the controller or your accountant.

Before you start

Access to the named tool; confirm license, version and cost with the vendor.
A small, non-confidential sample to try the workflow on first.
A qualified person to review the output before it is used.

Tools and access

Follow the steps

Export the card statement for the period as CSV (card last four, transaction date, posting date, merchant, amount). Collect the receipt photos or PDFs for the same period and check your company policy on uploading financial documents to an AI tool first; use a business account where required.
Have Claude or ChatGPT read the receipts into one table: file name, merchant, date, total, tax, any job number or note written on the receipt, and a short description of what was bought. It should copy amounts exactly and write UNREADABLE or NOT SHOWN instead of guessing.
Keep a Rules tab you maintain: cardholder to default job for the period, merchant or description keywords to cost code (fuel, small tools, consumables) and anything that always needs approval.
Have Copilot in Excel or Gemini in Google Sheets match each statement line to a receipt by card, amount and a date window you set, then fill a proposed job and cost code from the receipt note first and your rules second, with a column that says which one it used.
Build the exception list: statement lines with no receipt, receipts with no statement line, possible duplicates, split purchases across jobs, and lines where the receipt note and the cardholder default disagree. Send each cardholder or project manager only their own exceptions.
Spot-check a sample of matched lines against the receipts, then import or key the approved coding into your accounting system. Keep the workbook and receipts with the month's close file.

Try a small example

Replace the bracketed fields with your own project details. No tested sample run is published for this workflow yet.

Sample input

Fill in the bracketed [fields] in the text below with your project data.

Starter prompt

Here are the headers and a few sample rows of my Statement tab [paste: card last four, transaction date, merchant, amount], my Receipts tab [paste: file, merchant, date, total, note on receipt, description] and my Rules tab [cardholder card to default job; keyword to cost code]. Settings!B1 is the number of days a receipt date may differ from the transaction date. Write Excel formulas (or a step-by-step Copilot plan) that match each statement line to one receipt by exact amount and a date within my window, propose a job from the receipt note if it holds a valid job number from my Jobs list and otherwise from the card default, propose a cost code from my keyword rules, and record which source was used. Add flags for NO RECEIPT, NO STATEMENT LINE, POSSIBLE DUPLICATE (same amount and merchant within the window), NOTE CONFLICTS WITH DEFAULT and NO CODE. Use only my data and rules; do not guess a job or cost code that my data does not support.

Expected output

Not published yet. Treat any output as a draft and check it against the source documents.

Check the result

Every value, citation or quantity traces back to a source document.
Missing information is marked as missing, not filled in.

Limits and recovery

The workbook proposes coding from your own rules and what is written on the receipt. It does not decide whether a purchase is allowable, billable to an owner, taxable or properly approved. Every coded line is reviewed before posting, exceptions are resolved with the cardholder, and policy, tax and audit questions stay with the controller or your accountant.
Check your organization’s data policy before sharing project files with third-party tools.

Stop or switch when: The source cannot be verified, permissions are missing, or the result contains unsupported facts.

Frequently asked questions

How does AI help with this?

Claude or ChatGPT, then Copilot in Excel or Gemini in Google Sheets: Claude or ChatGPT reads receipt photos into a table with amounts copied exactly and job notes captured; Copilot or Gemini matches them to the card statement by amount and date window, proposes job and cost code from the receipt note and your rules, and builds an exception list per cardholder. A person still checks: A sample of matched lines against the receipts, every exception with the cardholder or project manager, and allowability, billing, tax and approval questions with the controller before anything is posted.