Industries · IT providers

AI quoting for IT providers: what the Fieldstone IT test proves

15 of 15 on unseen messy requests to a fictional IT provider, and 40 of 40 on its hand-written set (6 Oct 2026).

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).

IT providers quote all week: hardware refreshes, licence renewals, managed-service seats, each from a vendor list. A general model does not know your list, and a wrong price on a 35-seat renewal costs margin or trust. On Fieldstone IT, a fictional provider built for testing, the model alone scored 1 of 40, notes lifted it to 33 of 40, and notes plus the quote action reached 40 of 40. Ask for the draft partner terms below; a person on our team replies.

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.

End it whenever you want. Unit 101 and later: $695 per month, $1,800 onboarding. Fieldstone is fictional, so your own 30 past quotes get scored, free, before any client pilot.

Notes plus the quote action, not a bigger model, carry the Fieldstone IT score

Fieldstone IT is a fictional managed IT provider with a made-up price list and 40 test requests: 16 quotes, 10 price lookups, 8 sorting tasks and 6 requests that should be declined. Graded by code on a 16 GB Mac mini M4 with Qwen3 8B.

Share of Fieldstone IT requests answered right. Mac mini, 6 Oct: 40 of 40 hand-written, 201 of 204 generated. Dell GPU, 30 Sep: Qwen3 8B with no notes 1 of 40, notes only 33 of 40, notes plus quote action 40 of 40; Qwen3 27B with no notes 2 of 40. Hand-written set · Mac mini · 6 Oct 40/40 Generated regression set · 6 Oct 201/204 Dell GPU, 30 Sep: what lifts the score Qwen3 8B, no notes 1/40 Qwen3 8B, notes only 33/40 Qwen3 8B, notes + quote action 40/40 Qwen3 27B, no notes 2/40 0100%
Share right, from zero; filled = Staffbox as shipped (notes plus quote action), outline = baselines. Mac mini rows: 16 GB Mac mini M4, 6 Oct 2026; the generated requests were written by a model and checked by code. Dell rows: RTX 3090, 30 Sep 2026. Fieldstone IT is fictional. Source: results files on GitHub.
40/40Fieldstone hand-written setfictional · 6 Oct 2026
201/204generated regression set (requests written by a model, answers checked by code)fictional · 6 Oct 2026
2.3 saverage per answer, 90th percentile 5.3 sMac mini M4 16 GB
1/40same model with no notes (GPU test box)fictional · 30 Sep 2026
TestResultAverage / 90th percentile seconds
Fieldstone IT, 40 requests (6 Oct 2026)40 of 402.3 / 5.3
Generated messy requests, Fieldstone (6 Oct 2026)201 of 2041.6 / 2.1
No notes, Dell RTX 3090, Qwen3 8B (30 Sep 2026)1 of 40not measured
No notes, Dell RTX 3090, Qwen3 27B (30 Sep 2026)2 of 40not measured

A bigger model did not rescue the no-notes run: the 27B scored 2 of 40. Written notes on the price list, the discounts and the rules lifted Qwen3 8B to 33 of 40, and the quote action, which does the arithmetic in code, closed the rest to 40 of 40. For a provider that cuts both ways. The brain is yours and moves with you, and a demo brain proves nothing about how well your own list is written down.

Hardware, renewals and per-seat services are lookups; scoping goes to an engineer

  • A hardware list: "20 laptops, 20 docks, 5 monitors, same spec as the last order."
  • A renewal: "Quote the licence renewal for the same 35 seats."
  • A managed-service quote priced per seat or per device.
  • A short note from a client: "Can you price a replacement firewall?"
  • A request that needs an engineer: a scoping question, not a price.

Our read of the segment. The first three are lookups and arithmetic. The firewall note lacks a model and the scoping question lacks a price at all, so both should come back to a person; the six "should decline" tests in Fieldstone exist to check that the draft hands them over instead of inventing a number.

Staffbox · quotes deskfictional company · real run
Request · Fieldstone IT (fictional)

Price 24 of DOCK-C including setup please.

Draft
24 × DOCK-C at $219.00$5,256.00
less 4% on hardware (10 to 24)−$210.24
setup 24 × $10.00$240.00
Line total$5,285.76
Sources the draft cites
  • hardware-price-list · row DOCK-C USB-C dock · updated 30 Sep 2026
  • volume-discounts · 10 to 24: 4%, setup never discounted · updated 30 Sep 2026
graded right · 0.9 s · Mac mini · 6 OctNext: your review · nothing sent
A real request (quote-04) from the fictional Fieldstone IT test set, run on our 16 GB Mac mini on 6 October 2026. The figures are the model's output; the layout and the discount-amount line are ours. The sources are the notes in the fictional provider's brain the numbers come from. Results file

Note the setup line in that receipt: the discount applies to hardware only, because the brain says setup is never discounted. Rules like that are where an untrained model goes wrong.

Distributor feeds, PSA links and per-client rates are what a demo cannot show

  • Distributor price feeds. Staffbox reads a price-list file. It does not connect to a distributor feed. If your prices change daily, the draft cites the date of the file you gave it, and an old file means old prices.
  • PSA and quoting software. We make no PSA integration claim. Quoting systems with PSA integrations are compared in Staffbox vs QuoteWerks and Quoter.
  • Your own bundles and margins. The brain can hold them, but only Fieldstone's have been tested.
  • Your client list. Different clients may get different rates, and that rule has to be written down to be followed.

A provider's own 30 quotes come before any client pilot

Staffbox is sold and supported through managed IT providers; first partner conversations are under way and none is named. Running the free score on your own quotes first is the quickest way to judge the product before you put your name on it with a client. What to send:

  • Your price list or distributor export as one file, with its date.
  • 30 past quote requests and the quotes you sent, a mix of hardware, licences and services.
  • Your rules for margin, bundles and which clients get special rates.

Nothing is installed for this step. The files run on our own Mac mini in Atlanta, your team grades 10 held-back answers blind against a pass bar agreed before day 0, a refusal counts as a miss, and the files are deleted within 14 days of sending your results (per our draft data policy). The IT providers page has the who-does-what table and the install checklist; partner terms are on request from partners@staffbox.ai.

Questions

Is Fieldstone IT a real company?

No. Fieldstone IT is a fictional IT provider created to test Staffbox. No real IT provider has run it yet.

Does Staffbox integrate with ConnectWise or Autotask?

No integration is claimed. Staffbox reads a price-list file and forwarded requests, and writes drafts into a folder.

Where do IT providers email about partnering?

partners@staffbox.ai.

Sources

Or call us: 404 493 6248 · Email sales

Get partner termsTextCall