Workflow guide
How to turn a price list into quotes without retyping
Short answer: Only picking the row needs judgment; code does the sums and a person sends. On unseen messy test requests: 15 of 15 right (6 Oct 2026).
A quote from a price list is five small steps, repeated many times a week. Each retype is a chance to read the wrong row, and the request waits while someone does it. The fix is to let software find the row and do the sums while a person stays the sender. Send up to three requests below; drafts come back within 2 business days.
Updated 14 October 2026
Four of the five steps are arithmetic, so only one needs a model
Most quotes from a price list are not hard. They are repetitive. A request arrives by email, someone finds the row, multiplies, checks whether a volume break applies, adds a setup or freight line, and types the reply. Five small steps, done many times a week, each one a chance to read the wrong row.
Here is one request worked by hand, using the price list of Fieldstone IT. Fieldstone IT is a fictional IT provider we built to test on; no real company or customer is described. The request: 12 of the 27 inch monitors, set up.
| Step | Work |
|---|---|
| 1. Find the row | MON-27, monitor 27in, $289.00 each, $15.00 setup per unit |
| 2. Multiply | 12 x $289.00 = $3,468.00 |
| 3. Apply the break | 12 units falls in the 10 to 24 band: 4% off hardware. $3,468.00 less 4% = $3,329.28. Setup fees are never discounted. |
| 4. Add setup | 12 x $15.00 = $180.00 |
| 5. Total and write the reply | $3,329.28 + $180.00 = $3,509.28 |
We reproduced these numbers by running the staffbox quote command on the Fieldstone price list, which prices a line from the table with no language model involved. The language model's job is step 1, working out which row the customer means; the arithmetic is done in code.
Nothing in that list needs judgment except step 1: working out which row the customer means. That is the part a language model is for, and the part a person should check.
Here is the same kind of request run by the model: two dozen of the same monitor, with the quantity in words. It lands in the next discount band.
Need two dozen 27 inch monitors, set up please.
| 24 × MON-27 at $289.00 | $6,936.00 |
| hardware after 4% (10 to 24) | $6,658.56 |
| setup 24 × $15.00 | $360.00 |
| Line total | $7,018.56 |
- hardware-price-list · row MON-27 · updated 30 Sep 2026
- volume-discounts · 10 to 24: 4% · updated 30 Sep 2026
The price list needs four properties, not a new system
The list does not need to be in a special system. Fieldstone's is a Markdown table in a folder, with a header that says who owns it and when it was last updated. Four properties matter:
- One row per stocked item, with a stable code (SKU), a plain description, a unit price and any per-unit fee such as setup.
- Discount rules written down once, in their own note, saying what they apply to. In the Fieldstone brain, discounts apply to the hardware part of a line only, in three bands: 1 to 9 units none, 10 to 24 units 4%, 25 or more 7%.
- A date on the list. Every priced line in a draft cites the row and the date of the list it came from, so a stale list is visible to the person checking.
- A stated edge. The Fieldstone list says that anything not on it is a special order and goes to a person. That sentence is what lets the worker decline instead of guessing.
If your price list lives in Excel, the first job is turning it into that shape. The guide to preparing a price list for AI covers it step by step.
A person owns the last two of five steps, and the box cannot send
Staffbox runs this on a Mac mini in your building. The flow has five steps, and a person owns the last two.
| Stage | What happens | Detail page |
|---|---|---|
| Intake | A request lands in a folder or a synced mailbox, as it came in. | RFQ email intake |
| Sort | Each email goes to one queue: quote, service desk, billing and so on. | Quote request email triage |
| Price or answer | Code prices quote lines from the price table; questions are answered from the notes and the source is cited. | Price list lookup |
| Draft | A reply is written to your drafts folder with the row and list date on every line. | Draft replies that cite the row |
| Approve | A person reads the draft and sends it. Staffbox has no sending code. | Human approval for AI drafts |
Every page in this guide
| Page | What it answers | When you're ready |
|---|---|---|
| Price list lookup by email | Asking a question of the price list and policies, and getting the row and its date back. | Send up to three requests, free |
| Quote request email triage | How a shared inbox is sorted into queues, and the one routing miss we have recorded. | Send up to three requests, free |
| AI draft replies from a price list | What is inside a draft: the file, the citation footer and the reviewer note. | Send up to three requests, free |
| Human approval of AI drafts | Why a person stays the sender, and what that person checks. | Send up to three requests, free |
| AI quote accuracy scorecard | How we grade, with every score from the 2 to 6 October runs. | Send up to three requests, free |
| Free AI quote test | What to send, what you get back, what happens to your files. | Send up to three requests, free |
| RFQ email intake | The mechanics of reading an email file into a request. | Send up to three requests, free |
| AI misses and refusals | Every kind of miss we have found, by count. | Send up to three requests, free |
| What Staffbox does not do | Phone, sending, certification and the other limits, plainly. | Send up to three requests, free |
For the product overview and price, see AI quoting software. For messy requests and an item you do not stock, see RFQ automation. For what is in the box, see how it works.
It got 15 of 15 on messy requests it never saw, and none of 475 answers had a wrong total
The test that matters most is the messy set never used while fixing: 15 of 15 (6 Oct 2026). The notes do most of the lift: on the Fieldstone IT tests (Dell RTX 3090, 30 Sep 2026), Qwen3 8B scored 1 of 40 with no notes, 33 of 40 with the notes, and 40 of 40 with notes plus the quote action. 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).
How we measured: 6 October 2026, a 16 GB Mac mini M4 with Qwen3 8B, the notes and the quote action. Six sets, 475 answers: Fieldstone IT 40 of 40, Peachtree Cabinet Works (also fictional) 39 of 40, messy sets 19 of 20 and 15 of 15, and 351 of 360 on a generated regression set (requests written by a model, answers checked by code). No answer had a wrong total; of the 11 misses, 10 declined and asked the customer and 1 went to the wrong queue.
No customer has run Staffbox yet. These are our own test companies, graded by machine. The free 30-quote test is how you find out what it does on your list. The detail is on the scorecard page and in the Staffbox scorecard on Hugging Face.
Questions
Can AI turn a price list into a quote?
Yes, for requests where most lines come from a price list. Staffbox picks the item and quantity from the email, does the arithmetic in code from your price table, and writes a draft that cites the price-list row and the list date. A person checks and sends.
Does the price list have to be in a special format?
No special software. It needs one row per stocked item with a code, a description, a unit price and any fee, a date, and written discount rules. A table in a Markdown note works; a spreadsheet can be converted.
What happens when the customer asks for something not on the list?
The draft flags the item as not on the price list and names what you do stock, instead of swapping in the nearest item (by design; 0 wrong totals in 475 test answers). A person decides what to offer.
Sources
- Staffbox repository, quote action and Fieldstone IT demo brain: github.com/Staffbox-ai/staffbox, read 13 Oct 2026.
- Graded answers: huggingface.co/spaces/staffbox/scorecard.
- Test results, 6 Oct 2026: github.com/Staffbox-ai/staffbox/tree/main/evals/results.
Or call us: 404 493 6248 · How the free test works