Sample with made-up data. This is a demonstration of the kind of system we build, not a client project or a live product.
Sample · made-up data

What happens to a stack of invoices

Eight invoices came in.

The system caught a duplicate, a total that doesn't add up and a contractor who billed 19% over the purchase order.

A person decides on each one.

Example Co. Property Management is a made-up Nova Scotia business with three locations (the Harbour office, the Depot and the North lot) that checks every invoice against its purchase orders before paying it.

How this sample works
  1. 1Documents come in by email, phone photo or scanner.
  2. 2The system reads them, pulls out the key details and links each one to where it was found.
  3. 3It checks them against the company's own rules and purchase orders, and flags anything off.
  4. 4A person reviews and approves, corrects or rejects each invoice.
  5. 5Staff ask questions in plain language, and every answer links to the invoices it used.
  6. 6Only approved invoices go to the accounting software.
  • Every company, person, address, phone number and tax number here is made up. Postal codes use "Z", which Canada doesn't use, and phone numbers use 555-01xx.
  • This page doesn't call any AI service or send anything anywhere. The readings, flags and answers were worked out ahead of time from the made-up data so you can click through them. Your clicks stay in your browser and reset when you reload.
  • In a real build, the reading is done by established services, such as Azure Document Intelligence and Azure-hosted AI models. The checks, rules, review screen and connection to accounting software are what we build around them, set up for each client.
  • The "Clear", "Worth a look" and "A person must check" labels are for illustration only. They aren't measured results.
Sample · made-up data

1Documents come in

Eight made-up invoices in different formats. One is a crooked, grainy scan with a smudged PO number, one is a phone photo with a handwritten note, and one is the same invoice sent again as a reminder. Tap any invoice to open it.

Sample · made-up data

2What the system pulled out

The key details from each invoice. Tap a row for all the details, and tap any value to see exactly where on the invoice it came from.

Sample · made-up data
Clear read cleanlyWorth a lookA person must checkChecked by a person
Sample · made-up data

3What got flagged, and why

The three problems an owner would most want caught. Each flag explains itself in plain words, and flags update as you review in step 4.

More checks
Sample · made-up data

Rules this sample uses

  1. Duplicates: the same supplier and invoice number seen before.
  2. Arithmetic: the lines add up to the subtotal, HST is 15% and subtotal plus HST equals the total.
  3. Purchase orders: the PO exists, it's for the same supplier and the subtotal is within 2% of the PO amount.
  4. Unsure readings: anything the system may have misread goes to a person.
  5. Required fields: supplier, invoice number, invoice date, PO number and total.
  6. Approval limit: anything over $5,000 including HST needs the owner's sign-off.

These are example rules. A real client sets their own.

Sample · made-up data

4A person reviews before anything is used

Try it: approve, correct or reject each invoice. To approve an invoice that still has a flag, you have to write a note, and anything over the approval limit also needs a sign-off. Nothing reaches your accounting software until you approve it.

Sample · made-up data

5Ask questions in plain language

Three sample questions, answered from the made-up invoices and your review decisions in step 4.

In a real build, staff type their own questions. Answers can be wrong, so each one links to the invoices it used.

Sample · made-up data

Pick a question above.

Sample · made-up data

6Send approved invoices to your accounting software

Nothing reaches your accounting software until you approve it. Rejected and unreviewed invoices stay out.

Sample · made-up data