For IT providers · the offer
A scored AI offer your clients can say yes to
Short answer: score it free on 30 of a client's past quotes; only a passing client gets a Mac mini, at $595 a month at founding sites (first 100) for as long as the box is live (draft pricing).
Your clients are asking about AI and expect you to have an answer. Most AI products ask an owner to trust a vendor's claim and pay before seeing anything. This offer reverses that order: the client reads a score on its own quotes, then you install a box that only writes drafts. Ask for the draft partner terms below, or run the free score on one client's quotes first.
Updated 14 October 2026
Or start with one client: run the free score on one client.
The pitch is one sentence, and the client can check every part of it
Staffbox drafts your quotes from your own price list, and a person checks each one before it goes out. Before the client pays, we run it on 30 of their past quotes and show the score.
Each part of that sentence is a promise the client can test. Drafts: the worker never sends. Your own price list: every priced line cites the row and the list's date. A person checks: control stays with the client's staff. The score first: nothing is installed until the number is good enough. The wording dates from 2 October 2026; the table shows what stands behind each line when the client pushes back.
| Element | What you tell the client | What backs it |
|---|---|---|
| The job | Quote requests arrive by email. A draft reply lands in the drafts folder, every priced line citing a price-list row and that list's date. | The quote action in the open-source repo (MIT) |
| The proof | You will see a score on your own past quotes before you pay anything. | The 30-past-quotes score is free. We run it on our own Mac mini in Atlanta; nothing is installed at the client for this step |
| The control | A person at your company sends every reply. The unit cannot send or delete. | Staffbox has no sending code; sending and deleting are off in the install |
| The data | Models run on the box in your office. Cloud AI is off unless you add your own key in writing. | Ollama on 127.0.0.1; no cloud fallback configured on the demo unit |
| The price | $595 a month at founding sites (first 100) for as long as the box is live, free 30-day pilot, no onboarding fee. | From unit 101: $695 a month plus $1,800 onboarding. |
| The exit | End any time. The unit is returned or wiped in front of you; Staffbox deletes its copies within 14 days and confirms in writing. | Security sheet, exit row |
| The honesty | No customer has run it yet. No SOC 2. No insurance confirmed. | Company status in the Trust Center |
Qualify on one number: 8 or more quotes a week from a price list
Before you spend a meeting on it, ask the client one question: do you send at least 8 quotes a week from a price list? Distributors, fabricators, cabinet and millwork shops and IT providers usually do. That threshold is written into the offer. Without a written list there is nothing for the worker to cite.
| Good first client | Skip for now |
|---|---|
| Sends 8 or more quotes a week from a price list | Prices live in one person's head and nowhere on paper |
| Has a year of past quotes in email | Fewer than 30 past quotes to test on |
| Wants a draft in the drafts folder and is happy to press send | Wants the AI to answer customers on its own (Staffbox cannot send) |
| The owner or office manager will grade 10 answers | Nobody has time to grade them |
| Ordinary business pricing data | Data covered by CJIS, FERPA or HIPAA, unless their compliance office clears it |
The free score on 30 past quotes decides, before anything is installed
The client conversation has six steps, and the first install happens only after the last one.
- Pick the client with the one-number test above.
- Ask for two files. Thirty past quote requests and the current price list. Tell them to leave out passwords and personal data.
- We run the test. On our Mac mini in Atlanta. The client sees drafts for the requests they already answered, with the price-list row cited on each line.
- The client grades blind. Ten of the 30 are held back and graded by the client's team without knowing which answers are ours. See how to run an AI pilot for the grading sheet.
- Agree the pass bar before day 0. Write it down before anyone sees a result. A refusal counts as a miss.
- Decide. If it passes, you and the client pick a founding site and schedule the install. If it fails, nothing was installed and the files are deleted.
The client never has to believe a claim from us or from you. They read their own numbers. In our view that is the only sales motion that survives a skeptical owner.
Numbers you can quote, and their limits
On 6 October 2026, on a 16 GB Mac mini M4 with Qwen3 8B, our fictional demo companies scored 15 of 15 on unseen messy requests, Fieldstone IT 40 of 40 and Peachtree Cabinet Works 39 of 40. A generated regression set (requests written by a model, answers checked by code) scored 351 of 360. Across 475 answers there were 0 wrong totals. Of the 11 misses, 10 declined and asked the customer, and 1 was a routing miss. The notes do most of the lifting: on Fieldstone (Dell, 30 Sep) the model alone scored 1 of 40, with notes 33 of 40, and with notes plus the quote action 40 of 40. 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).
Say the limits out loud. Both companies are fictional. The tests were graded by machine. No customer has run Staffbox. Fifteen requests is a small sample for the held-out sets. A client who hears the limits from you trusts the rest more.
After a pass, you install and support the unit, and Staffbox runs every score
You install the unit, apply the monthly updates and take first-line calls, as you do for any other device you manage. Staffbox runs the score at every stage: before the sale, before day 0 and before each monthly update. The client's staff send every reply. The figure shows the split across the life of one client.
Who handles what beyond first line is not written down yet. The step-by-step is in the install and support checklist, and the answers your client's security reviewer will ask for are in the security questionnaire page.
What a partner gets is in the draft terms, which are private and not listed on this site. Email partners@staffbox.ai for the current draft.
Questions
What does the client pay?
$595 a month at founding sites (first 100) for as long as the box is live, with a free 30-day pilot and no onboarding fee. From unit 101 it is $695 a month plus $1,800 onboarding.
What if the client's score fails?
Nothing is installed. The pass bar is agreed before day 0, the unit goes live only if the score passes, and Staffbox deletes the test files afterwards.
Does Staffbox have customers I can call?
No. No customer has run Staffbox yet. The first sites are founding sites, and the scores on this site come from fictional demo companies.
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.
- Staffbox-ai/staffbox on GitHub (MIT); evals/results for the 6 Oct 2026 runs.
- Scorecard Space: every graded answer.
Or call us: 404 493 6248 · How the free test works