Industries · Cabinet and millwork
AI quoting for cabinet shops: price lists, not drawing takeoffs
39 of 40 on a fictional cabinet shop's price list (6 Oct 2026); the one miss was a wrong queue, not a wrong price.
Half of a cabinet quote is a price list: a box in a width and style, plus door, finish and hardware options. The other half comes from drawings and a site visit, and a tool that cannot tell the halves apart will guess. Staffbox drafts only the list half; on Peachtree Cabinet Works, a fictional shop built for testing, it scored 39 of 40. Send up to three requests your shop already answered; drafts come back within 2 business days.
Updated 14 October 2026
Free 30-day pilot, then $595 a month at founding sites (first 100) for as long as the box is live (draft pricing). No customer install yet.
Stop any time. Shops after the first 100 pay $695 a month plus $1,800 onboarding. The scores below are from a fictional shop; your free 30-quote score decides whether a pilot starts.
Staffbox prices the list half of a cabinet quote, never the drawing half
The list half is mechanical. A base cabinet in a given width and style has a price, and door style, species, finish, soft-close hardware and filler panels add to it by a rule. The drawing half is judgement: what fits the room, what has to be built custom, how install will go. Staffbox does not read drawings or take quantities off plans, so if your quotes start from a drawing, it starts after the takeoff.
Customer wants 60 slab door 24x30s, cherry, unfinished. How much for the line?
| 60 × SL-2430 at $48.75 (cherry +25%, unfinished +0%) | $2,925.00 |
| less 8% (50 or more on the line) | −$234.00 |
| Line total | $2,691.00 |
- price-list · row SL-2430 slab door 24x30 $39.00 · updated 30 Sep 2026
- surcharges · cherry +25%, unfinished 0% · updated 30 Sep 2026
- volume-discounts · 50 or more: 8% · updated 30 Sep 2026
Things a cabinet shop should not expect from it:
- Drawing takeoff and room layout. Tools that read drawings exist; Staffbox is not one of them.
- Install labour and travel estimates, which are not a lookup.
- Design advice or matching a style from a photo.
- Sending. The unit cannot send or delete anything; a person approves each draft.
On the fictional Peachtree shop, the list half scored 39 of 40
Every figure here is machine-graded on a 16 GB Mac mini M4 running Qwen3 8B, one request at a time with the model warm. Peachtree is not a real company; its price list was written for the test.
| Run | Result | Date |
|---|---|---|
| Peachtree, 40 hand-written requests | 39 of 40 | 2 Oct 2026 |
| Generated regression set, before the unstocked-item fix | 147 of 156 | 5 Oct 2026 |
| Generated regression set, after the fix | 150 of 156 | 6 Oct 2026 |
| Average / 90th percentile seconds per answer, hand-written set | 2.5 / 4.4 | 6 Oct 2026 |
The bottom bar is there for scale. With no notes, the same small model got 1 of 40 right on the fictional IT provider's tests; notes alone lifted it to 33 of 40, and notes plus the quote action to 40 of 40 (Dell, 30 Sep 2026). The notes and the price list carry the score, which is why the shop's own list is what has to be tested.
The misses were a routing slip and declines, never a wrong total
The one miss in the Peachtree 40 was a sorting error: request triage-05 went to the office manager instead of the queue called "other." No price was involved. We name it because a shop deciding whether to trust a tool should see the miss that actually happened.
The generated set tells the rest. On 5 Oct, before a fix to the quote action, six wrong totals across both demo companies all came from one pattern: a customer asked about an item not on the list, and five times the tool priced the nearest stocked item instead (once it priced the stocked lines and skipped the question). The 6 Oct version flags unstocked items instead of substituting (by design; 0 wrong totals in 475 test answers). That brought Peachtree's generated score from 147 to 150 of 156. Of the 11 misses that remained across both companies, 10 were declines that asked the customer a question. In our view that is right for a cabinet quote with no finish or width given; a refusal still counts as a miss on the scorecard. On a new trade it had never seen, an electrical distributor (14 Oct, Dell test box), the default Qwen3 8B setup scored 30 of 40 and 2 of 15 on messy requests, with 8 wrong prices in 55, because per-foot units, price levels and trade shorthand are not handled yet; that is why we score your own list first (the electrical result).
Your own 30 past quotes decide whether a pilot starts
- The price list as the shop uses it, with the option, species and finish tables. If they sit in a separate sheet, send both.
- 30 past requests with a spread: a homeowner after a rough number, a designer with a width list, a builder repeating a style.
- A mark on the quotes that needed a site visit or drawings. Those should come back as questions, not prices.
- The rules that live in someone's head: the minimum order, the surcharge for a painted finish, lead time by door style.
For a shop, this step costs nothing and puts nothing on your network. Your requests run on our own Mac mini in Atlanta; your team grades 10 held-back answers blind against a pass bar agreed before day 0, and the files are deleted within 14 days of sending your results (per our draft data policy). Only a passing score leads to a pilot.
Questions
Does Staffbox read cabinet drawings?
No. It drafts from a price list. If your quotes start from drawings, Staffbox starts after the takeoff.
Is Peachtree a real shop?
No. Peachtree Cabinet Works is a fictional company with a price list written for testing. No real cabinet shop has run Staffbox yet.
What happens when the customer leaves out the finish?
The draft is built to ask rather than guess. Of the 11 misses in the 6 Oct runs, 10 were declines that asked the customer a question; we did not count how many of those involved a missing finish.
Sources
- How we measured: two fictional companies (Fieldstone IT, Peachtree Cabinet Works), graded by code, Qwen3 8B on a 16 GB Mac mini M4 (Dell RTX 3090 for the 30 Sep runs), 30 Sep to 6 Oct 2026. No customer has run Staffbox yet.
- Scores, test sets and every graded answer: Staffbox scorecard on Hugging Face and github.com/Staffbox-ai/staffbox (evals/results). Staffbox claims on this page come from our public scorecards (huggingface.co/spaces/staffbox/scorecard) and the open repo (github.com/Staffbox-ai/staffbox).
- Peachtree results: evals/results 2026-10-02, 2026-10-05 and 2026-10-06 in github.com/Staffbox-ai/staffbox.
- No-brain contrast: Fieldstone IT, Qwen3 8B on a Dell RTX 3090, evals/results 2026-09-30.
Or call us: 404 493 6248 · Email sales