Free AI quote test: send 3 requests, then 30 past quotes

Two free steps, in order, with nothing installed at your site

Both are free. The 30-quote score is the one that tells you whether to go further. If it passes the bar you set, a free 30-day pilot is the next step, and the workflow goes live only if it meets the bar. After the pilot it is $595 a month at founding sites (first 100) for as long as the box is live (draft pricing), with no onboarding fee; from unit 101 it is $695 a month plus $1,800 onboarding.

Who the test is for

Firms that send at least 8 quotes a week from a price list: distributors, fabricators, cabinet and millwork shops, and IT providers. If most of your quotes come from a catalogue or rate card and the requests arrive by email, the test fits. If every quote is a custom engineering estimate, it does not, and the 3-request test will show that quickly.

How to start

Paste three of your recurring requests into the step 1 form above, email sales@staffbox.ai, or call 404 493 6248; a person answers. IT providers who want to run the test for a client can write to partners@staffbox.ai.

You send the list, its rules, and 30 real requests with your answers

• The price list as you use it today, with the date it was last updated. A spreadsheet is easiest; ask us about other formats. We turn it into notes the worker reads.
• Your discount and fee rules: volume breaks, setup or freight, minimum order. If they live in someone's head, write them down in a few lines.
• 30 past quote requests as they arrived, forwarded threads and typos included. Pick a spread, not your best ones: some easy, some messy, and a few you would not have priced without a call.
• The answer you actually sent for each one, with the total.
• Anything customer-identifying you can remove, such as phone numbers and signatures. Leave the wording of the request alone. This is our suggestion, not a requirement.

Our guide to preparing a price list for AI goes further on the price-list side.

What your test set becomes

Each request turns into one line in a plain test file: an id, a type, the request text and the expected answer. These are the first lines of the Fieldstone IT test file (a fictional company, in the open repository):

{"id": "quote-01", "type": "quote", "prompt": "Quote 40 x business laptop 14in (LT-14) with setup. What is the line total?", "expect": 46542.8}
{"id": "quote-02", "type": "quote", "prompt": "Client wants 20 business laptop 16ins, set up and ready. How much for the line?", "expect": 28760.8}

The expected answer is the answer you actually sent. Ten of the thirty are held back and not used while we tune the brain, and your team grades Staffbox's answers to those ten blind.

What comes back is a score with the misses shown

• The share of requests priced, and the share right with no edits.
• The misses, shown rather than hidden. A refusal counts as a miss, so a tool cannot score well by declining everything.
• The 90th-percentile time to a draft, from our Mac mini.
• Your 10 blind-graded answers, with your team's marks.
• A projected return: quotes per week times minutes saved per quote times your loaded hourly rate, labelled as a projection.

Method and sample numbers are on the scorecard page. For the kinds of miss to expect, see misses and refusals.

The test is a sample on our hardware, not proof of a live install

• It does not prove anything about a customer install. No customer has run Staffbox yet. Our scores come from fictional companies. Yours will be the first real one.
• Why your own 30 quotes, not our scores. On three trades it had never seen, the worker made wrong prices on all three, and a fix tuned on two trades did not carry to the third, so only a score on your own list tells you whether it fits (the three-trade results).
• Thirty requests is a sample. It shows where the tool is strong and where it breaks. It does not guarantee the next hundred.
• It runs on our hardware. Speed on your site depends on the Mac mini or Studio you choose. Mac Studio is not measured yet.
• It does not test your mail system. How requests reach the worker is set up at install by your IT provider.

Sources

• Offer terms and test steps: Staffbox's own records, 2, 6 and 13 October 2026.
• Test file format: examples/fieldstone-it/tests.jsonl in github.com/Staffbox-ai/staffbox, read 13 Oct 2026 (fictional company).