Can AI plan stock and reordering? We tested it on 18 months of sales.
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.
Yes, when the ordering rules are written down. Given a written policy and 18 months of a made-up company's sales, our AI matched all 139 lines of an 8-week order plan, with every price and delivery date right. It saw 11 of 11 stock-outs coming and ordered nothing wrong. It took 8 minutes. Purchase orders stayed drafts.
What it got
- 119 stock items and 26,297 sales lines over 18 months
- Open purchase orders, booked installation jobs and a warehouse stock export
- 9 supplier price lists and 35 purchasing emails, 7 of which changed something
- The company's written ordering policy
Plan orders for the next 8 weeks and draft today's purchase orders per supplier, without sending anything. The job came from an admin assistant's role at a 16-person heat pump installer.
The results
| Measure | Result |
|---|---|
| Order lines in the 8-week plan | 139 of 139 |
| Quantities exactly as in our answer key | 138 of 139 |
| Unit prices right on the order date | 139 of 139 |
| Delivery dates right | 139 of 139 |
| Items to order today | 29 of 29 |
| Stock-outs seen coming | 11 of 11 |
| Draft purchase orders | 7 of 7 |
| Hidden problems caught | 13 of 13 |
| Discontinued or wrong items ordered | 0 |
A script scored every line against an answer key the AI never saw. How the tests work.
What it missed
- Nothing outside the margin. Our scorer allows one pack or 20% on a quantity, and all 139 were within that.
- One quantity differed: 178 where the answer key said 172.
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 |
|---|---|
| A discontinued item | Replacement ordered |
| A longer lead time, only in an email | 14 of 14 delivery dates used it |
| A supplier closed over the holidays | Ordered earlier, around the closure |
| A price rise from a later date | Right price on each order date |
| One-off bulk sales in last year's figures | Not treated as normal demand |
| An item renamed, with old and new codes in the history | Treated as one item |
| Items counted in boxes, not units | Converted |
| Stock already allocated to booked jobs | Not counted twice |
| A negative stock line | Flagged as a data error |
What this test doesn't show
The policy here was exact, so there was one right answer. Many small businesses keep these rules in someone's head, and getting them written down is part of the work. This test is about applying rules, not about choosing better ones. 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.