Can AI chase unpaid invoices? We tested it on 215 open invoices.
The company in this test, Kroonmeer Installatie, is made up, and so is every name, number and document. Real work runs on your own data, and you check everything before it counts.
Matching and drafting, yes, if a person sends the letters. At a made-up company, our AI matched 320 of 320 customer payments and gave 118 of 120 customers the right next step. It chased nobody who had paid, disputed the invoice or wasn't due yet. It took 11 minutes. Every letter stayed a draft.
What it got
- 215 open invoices and a list of 120 customers
- 374 bank lines
- 45 customer emails with promises, disputes and changes
- The company's written chasing policy
Match the payments, list who owes what, choose one next step per customer, draft the letters and flag anything for a person. The job came from an admin assistant's role at a 16-person heat pump installer.
The results
| Measure | Result |
|---|---|
| Customer payments matched | 320 of 320 |
| Other money rightly left unmatched | 54 of 54 |
| Right next step per customer | 118 of 120 |
| Balances by age right, all customers | 120 of 120 |
| Letters written | 37 of 37 |
| Facts right in the letters | 168 of 168 |
| Issues flagged for a person | 18 of 18 |
| Customers wrongly chased | 0 |
A script scored every line against an answer key the AI never saw. How the tests work.
What it missed
- Two small leftovers, €0.56 and €8.40, weren't proposed for write-off as our policy said. It noted both and asked the owner to decide. We still count that as a miss.
Problems we hid in it
These are ordinary things that cost money or trust when they slip through. This is a selection.
| What we hid | What it did |
|---|---|
| One payment covering three invoices | Split right |
| Two customers called S. Vos, both owing €149.00, and a payment with no reference | Matched by bank account |
| An insurance payout of exactly an open invoice's amount | Not treated as a customer payment |
| A customer quoting another customer's invoice number | Matched to the right invoice |
| A customer who says "already paid", with nothing in the bank | Held for a person |
| Promises to pay: some still in the future, some already broken | Held the first, chased the second |
| A customer declared bankrupt | Chasing stopped |
| Private customers more than 35 days late | The 14-day letter, never a business final notice |
| A large invoice with an agreed later due date | Not chased |
What this test doesn't show
It followed a written policy with fixed steps. Knowing when to phone a good customer instead of sending a reminder isn't in this test, and that call stays with you. This was also one run, on one made-up company, and we wrote the test ourselves.
Would this work on your admin?
Send us the job ad you're hiring for, and we'll send back a free one-page breakdown of what we'd take off the role.