Proof
One job ad, six tests, scored blind.
We made up a company: Kroonmeer Installatie, 16 people, installing heat pumps in Groningen and hiring an admin assistant. Then we gave our AI the routine part of that job: over 2,000 documents, emails, bank lines and job requests, plus 26,000 lines of sales history, with more than 150 problems hidden in them on purpose. Every result was scored by a script against an answer key the AI never saw.
Kroonmeer is made up, and so is every name, number and document in these tests. Real work runs on your own data, and you check everything before it counts.
Test 1
Three months of books
163 supplier invoices and receipts (some of them phone photos), 485 bank lines, 96 sales invoices.
- 485 of 485 bank lines matched
- 29 of 31 problem documents caught
- 0 invoices passed that should have been stopped
15 minutes, then a person checks it.
Missed: a US-dollar invoice from a new supplier and a reminder for an invoice already paid. Neither was booked wrong, but both should have gone to a person for a check.
Full results ›
Test 2
Two weeks of the shared inbox
300 emails: orders, complaints, suppliers, job applications, spam and fakes.
- 298 of 300 emails sorted right
- 89 of 89 order lines entered exactly
- 0 of 8 scams and fake invoices answered
24 minutes.
Missed: two spam emails marked as suspicious instead, and two emails sent to a different colleague.
Full results ›
Test 3
Deadlines and paperwork
114 contracts, policies, letters and permits.
- 105 of 105 deadlines found with the exact date
- 93 of 93 form fields filled right, nothing made up, nothing signed
- 0 moved or cancelled deadlines still on the old date
12 minutes.
Missed: nothing. One deadline was listed twice.
Full results ›
Test 4
Stock and reordering
119 stock items, 26,297 sales lines over 18 months, supplier price lists and emails.
- 139 of 139 order lines in the 8-week plan, every price and delivery date right
- 11 of 11 stock-outs seen coming
- 0 discontinued or wrong items ordered
8 minutes.
Missed: nothing. One order quantity was 178 where the answer key said 172, inside the margin we allow.
Full results ›
Test 5
Chasing unpaid invoices
215 open invoices, 374 bank lines, 45 customer emails.
- 320 of 320 payments matched
- 118 of 120 customers given the right next step
- 0 customers chased who had paid, disputed or weren't due yet
11 minutes.
Missed: two leftovers under €10 not proposed for write-off. It asked the owner about them instead.
Full results ›
Test 6
Planning the technicians
8 technicians, 140 job requests and 20 maintenance visits over two weeks, with changes and cancellations.
- 151 of 151 jobs planned that could be
- 0 rules broken (skills, hours, travel, days off)
- 9 of 9 jobs that couldn't fit, flagged with the reason
17 minutes.
Missed: nothing.
Full results ›
What these tests don't show yet. Your documents, your systems and your exceptions. That's what the first month on your own accounts is for: we start with one task, check every result with you, and only then add the next.
Send us your job ad See how each test works