01 the business
Building services contractor
85 staff, 400 supplier bills a month
Bills arrive as PDFs, phone photos and email bodies. One bookkeeper types them all into the accounting system. Coding is inconsistent, duplicates slip through, and the pile grows whenever the bookkeeper is on leave.
02 before
The problem
Supplier invoices arrive as PDFs, photos and email bodies, and someone in the office types each one into the accounting system. Line items get keyed against the wrong account, duplicates slip through, and the pile grows whenever that person is on leave.
- Bills sit for days before they are entered.
- The same supplier gets coded to different accounts.
- Duplicate bills are paid now and then.
03 the build
What the AI does
- Step 1: Watches the accounts inbox in Gmail or Outlook and picks up every invoice attachment, including scanned PDFs and phone photos.
- Step 2: Reads the supplier, invoice number, dates, line items, tax and totals, and flags anything it cannot read clearly.
- Step 3: Matches the supplier to an existing contact and suggests account codes and tracking categories based on how past bills from that supplier were coded.
- Step 4: Checks for duplicates by invoice number and amount, and compares the bill against the purchase order when one exists.
- Step 5: Creates a draft bill in Xero or QuickBooks with the original file attached and a short note explaining each coding choice.
paused: waiting for a person to approve
How it was put together
- Picked up every bill from the accounts inbox, including scans and photos.
- Read supplier, number, dates, lines, tax and totals, flagging anything unclear.
- Suggested account codes from how each supplier's past bills were coded.
- Checked for duplicates and compared with the purchase order, then created a draft bill with a note on each coding choice.
04 human in the loop
Where a person stays in charge
Your bookkeeper reviews each draft bill and approves it before it is posted. Anything flagged as a possible duplicate, a new supplier or a price change against the purchase order goes to a separate queue.
- The bookkeeper approves every bill before posting.
- Possible duplicates, new suppliers and price changes go to a separate queue.
- The original file stays attached to every bill.
05 rollout
From kickoff to live in 4 weeks
- Week 1Connect the inbox and accounting system; review coding history.
- Weeks 2–3Create drafts alongside manual entry and compare coding.
- Week 4Switch to review-and-approve for all bills.
06 after
What changes
The bookkeeper stops retyping invoices and spends their time checking and approving. Bills reach the ledger sooner, so payables and supplier balances are current when you look at them.
Assume 400 bills a month at 6 minutes each to key by hand, and 1.5 minutes to review a draft.
400 × 6 minutes = 40 hours a month today. 400 × 1.5 minutes = 10 hours. 30 hours a month go back to the bookkeeper.
What to measure
- Days from bill received to bill entered.
- Coding corrections at review.
- Duplicates caught before payment.
- Bookkeeper hours per month on bill entry.
07 evidence
What the research says
In 1,921 fraud cases, expense reimbursement schemes made up 13% of cases with a median loss of $50,000, and frauds at organisations with under 100 staff had a median loss of $141,000.
In a study of 453 professionals doing writing tasks from their own jobs, access to ChatGPT cut the time taken by 40% and raised quality, as rated by evaluators, by 18%.
08 lessons
Mistakes to avoid
- Auto-posting. Keep a person's approval on every bill.
- Messy supplier records. Merge duplicate suppliers first.
- Ignoring the notes. The coding explanations are what make review fast.
09 questions
Common questions
- Does it replace Dext or Hubdoc?
- It can work alongside them or instead of them. The difference is coding suggestions based on your own history and the duplicate and purchase order checks.
- Which accounting systems are supported?
- Xero and QuickBooks most often, plus NetSuite and others with an API.