Machine Shop Quote Template: AI Turns RFQ Emails Into Draft Quotes

Set up a machine shop quote template with fixed fields (part, revision, quantity, material, finish, delivery), have an AI assistant extract those fields from each RFQ email into it, price the draft from your written rate card, and let an estimator check and approve every number before it leaves. Never quote from a drawing description alone.

The temptation is to let the assistant do the pricing too. Resist it. Language models read emails well and do shop arithmetic badly when the rules live in someone's head, so the shop that gets value from this keeps judgement (how hard is this part to hold?) with the estimator and hands over the clerical work: reading, sorting, looking up rates, multiplying and writing the cover note. If you have already tried the shortcut, what goes wrong when ChatGPT quotes machine shop jobs shows where the errors come from.

Follow me on Instagram@sagnikteaches

Where the hours go between an RFQ landing and a quote leaving

Before changing anything, time one ordinary RFQ from arrival to send. In most small shops the estimator spends the first ten to twenty minutes on tasks that need no estimating skill: opening attachments, finding the revision, copying quantity and material into a sheet, working out which outside processes are needed, looking up this month's bar price, and replying to ask what the customer forgot to say.

Connect on LinkedInSagnik Bhattacharya

That clerical part is what the assistant takes. The skilled part, looking at the print and deciding that the 0.0005 in bore needs a second setup and a grinding sub-contract, stays human. Splitting the job this way also tells you what to protect: the estimator's attention should go to the parts of the drawing that decide the price, not to typing.

Subscribe on YouTube@sagnikteaches

The fields every RFQ must be reduced to

An RFQ email arrives in any shape: a one-line message with a PDF, a forwarded thread, a portal notification, a spreadsheet of forty line items. The template is the fixed shape you convert them into. Keep it short enough that people fill it in, and mark which fields can be blank without stopping the quote.

Notice the third column. A template that only lists fields lets an incomplete RFQ drift through as a complete-looking draft. The "if missing" rule is what turns the template into a gate.

An extraction prompt you can paste and adapt

The first AI job is narrow: read the email, fill the fields, flag what is missing, and invent nothing. Give it the field list and tell it explicitly to write "NOT STATED" rather than guess.

You are helping the estimator at a small machine shop. Read the RFQ email
below (and the attachment names I list) and fill in this table. Rules:
- Use only what the email states. If a field is not stated, write NOT STATED.
- Never infer material, revision, tolerances or finish from the part name.
- Copy quantities exactly as written, including any break quantities.
- List every question we must ask the customer before we can quote.
- Do not give prices. Do not comment on whether the job is a good one.

Fields: customer, contact, part number, revision, quantity (and breaks),
material and condition, finish / outside process, critical tolerances or
inspection requirements, certifications, delivery date, delivery terms,
attachments received.

Attachments: BR-2210_revC.pdf, BR-2210_revC.step

EMAIL:
[paste the email text here, without signature blocks or other customers' details]

Here is the kind of output you should expect from a decent assistant, illustrated for a made-up RFQ asking for 200 aluminium brackets. Treat it as a draft to check, not as a record.

What you would fix: the assistant wrote "rev C" from the file name while the email says "latest", which is exactly the gap that causes trouble. The estimator confirms the revision on the drawing's title block and the reply to the customer quotes it back. That one check is worth the whole setup.

Your rate card, written down so the AI can read it

Most shops have a rate card in the estimator's head and in three old spreadsheets. An assistant cannot apply what is not written. Put the rules in one short document, with the numbers that change (bar prices, sub-contract rates) in a dated table the estimator updates. These figures are illustrative, so replace them with yours.

The line "price = cost divided by 0.75" matters. A 25% margin on price is not cost plus 25%. If your card is vague on this, the assistant will pick one reading and use it silently, and your margin will be wrong in a consistent direction on every quote.

A worked quote: 200 brackets, three quantity breaks

Now the estimator's input. The estimator reads the drawing and decides (these are human judgements, entered as numbers): setup 2.5 hours, cycle time 11 minutes a part, 2 minutes of deburr and inspection a part, stock of 1.1 lb a part at $4.20 a pound, anodise at $1.60 a part from the outside vendor with a $150 minimum lot.

The assistant then applies the card. For 200 parts:

• Material: 1.1 lb × $4.20 = $4.62, plus 8% waste = $4.99 a part.
• Machining: 11 minutes at $95 an hour = $17.42 a part.
• Deburr and inspection: 2 minutes at $55 an hour = $1.83 a part.
• Anodise: 200 × $1.60 = $320, above the $150 minimum, so $1.60 a part, plus 15% handling = $1.84 a part.
• Setup: 2.5 hours × $95 = $237.50, spread over 200 = $1.19 a part.
• Cost per part: $27.27. Price at 25% margin on price: $27.27 ÷ 0.75 = $36.36 a part, $7,272 for the lot.

Asked for breaks at 50 and 500, the same card gives the table below. At 50 parts the anodise lot is $80, below the $150 minimum, so the vendor cost is charged at the minimum: $3.00 a part plus handling.

(The 25-piece row is added in case the customer asks later; the quantities requested were 50, 200 and 500.) The break table tells a story the customer can use: most of the difference between 50 and 200 is setup and the vendor minimum, not cutting time. A buyer who sees that may take the 200.

Ask the assistant to show every step as a line, not just a total, and have the estimator re-add the column on a calculator or in a spreadsheet. Language models make small arithmetic slips, and a total that is $0.30 a part out on 500 parts is $150 gone. Better still, put the rate card in a spreadsheet with formulas, and use the assistant only to read the email and to write the cover note. For help structuring a sheet like that, see Google Sheets AI prompts.

The estimator's read of the print: what the AI cannot see

The numbers above rest on four estimator decisions: setup hours, cycle time, inspection minutes and stock size. None of them comes from an email. They come from looking at the drawing and knowing how your machines behave. A workable division is to let the assistant propose, from the email and the part description, a list of things worth checking, and let the estimator tick them off:

• Any tolerance tighter than the general block, and whether it needs a second setup, a fixture or grinding.
• Thin walls, deep pockets, long slender features or sharp internal corners that slow the cycle or risk scrap.
• Threads and their class, and whether they need gauges you do not own.
• Surface finish callouts finer than your machine leaves as-cut.
• Notes about deburring, marking, packaging or cleanliness that add minutes the quote forgot.
• Whether the material size on the print matches a standard stock size, or needs a special order.

Keep this as a short checklist on the template, with the estimator's initials against it. If the AI has been given only the email, it can tell you what to look for but not what is there, and the template should say so in plain words at the top: "Draft prepared from email text only. Not checked against drawing."

Writing the quote itself, with terms attached

Once the estimator has approved the numbers, the assistant's second job is the paperwork: a quote document in your house style, with the quantity breaks, lead times, validity period and terms. Give it your standard terms as fixed text it must copy and not rewrite, since reworded terms are a contract risk. A prompt that works:

Write a quote email using the approved figures below. Use my standard
terms paragraph exactly as supplied, without editing it. Keep the email
under 150 words above the table. State the revision, material, finish and
the assumptions listed. Do not add promises about delivery or quality.

Approved figures: [paste the estimator's table]
Assumptions: revision C; 6061-T6; type II black anodise by our usual vendor;
no material certificate unless requested; quote valid 30 days.
Standard terms: [paste your terms]

A good draft gives the customer a clear price table, names the assumptions and ends with one question or call to action. A bad one sprinkles in "we pride ourselves on", promises "fast turnaround" and rewrites a payment term. Reject it, tighten the prompt, and keep the standard terms outside the model's discretion.

The assumptions list deserves attention. If you quoted on revision C and the customer meant revision D, the written assumption is the reason you can reprice without an argument.

The reply that stops a bad quote before it starts

For an RFQ that fails the template's gate, the best use of the assistant is a clear, short request for what is missing. This is also where the strongest time saving sits, because the old pattern was a long email thread of one question at a time.

Hello [first name],

Thanks for the enquiry on BR-2210. To quote it accurately we need:
1. Confirmation of revision (the file names show C; your email says "latest").
2. Whether material certificates or an inspection report are required.
3. Delivery terms, and whether five weeks is a hard date.
4. Quantities 50, 200 and 500 as asked, or any others.

We will return the quote within [your usual turnaround] of hearing back.

[your name]

Notice that it contains no price and no guess. A customer who answers four numbered questions in one reply is a customer you can quote in one pass.

Four other RFQ shapes and how the template copes

The repeat order. A customer asks for "the same as last time". The assistant's job is to find the previous quote and job record, show the old revision, price and actual hours, and flag any change in bar price since. The estimator decides whether last time's setup hours were realistic. Actual hours are better data than any estimate, so keep them where the assistant can read them.

The prototype of one. Quantity one makes setup the whole cost, and the quote usually includes programming time. Add a "programming hours" line to the card for single and low-quantity parts, because the assistant will not include one unless the card says so.

The multi-line spreadsheet. A buyer sends forty lines. Let the assistant split them into one template per line and sort them into ready, held and decline. The estimator quotes the ready ones, the held ones go back for answers, and nobody retypes forty lines.

The assembly with bought-in parts. Here the draft must list each purchased component with a supplier price and lead time. Use the assistant to extract the parts list from the bill of materials, then check it with the method in checking a bill of materials with AI. A missing fastener on an assembly quote is a small cost you only discover after the order.

When a draft nearly went out wrong

An illustrative scenario that shows how the failure arrives. A shop's assistant draft for a stainless fitting looked flawless: a clean extraction, a sensible price table, a polished cover note. The extraction table said "316 stainless". The customer's email said "stainless" and the drawing note said 304. The assistant had filled in 316 because it is the grade most often mentioned in the same sentence as fittings. Nobody typed it, nobody saw it, and the table looked the same as every other.

The difference in bar price was real, but the bigger problem was a customer who would have received a quote naming a material they had not specified. What caught it was not cleverness but the rule written into the prompt, "never infer material", and the estimator's tick-box that says "material confirmed against the print title block".

The lesson generalises. Extraction errors do not look like errors. They look like tidy, confident, complete rows. So the control is a human check on a short list of fields where a wrong value changes the price (material, revision, quantity, finish) and a prompt rule that forces NOT STATED when the email is silent.

Keeping customer drawings out of the wrong place

Drawings and models are the customer's property, often covered by a non-disclosure agreement, and some carry export-control or secrecy markings that forbid sharing them with outside services. Before pasting any drawing file into an AI tool, check three things: what the NDA says about third-party processing, whether the print carries a control marking, and what your plan's data settings do with uploads (some consumer plans may use content to improve models unless you switch it off; business plans typically do not, but confirm on the vendor's own page).

The safe pattern for most shops is to give the assistant the email text and file names only, and to have the estimator type the part facts they need into the template. That loses little, because the assistant was never going to price from the print anyway. For the wider question of what you can safely paste, see how to classify business data before using AI tools.

Proving it works: three numbers to track for a month

1. Time from RFQ received to quote sent. Record it for twenty quotes before the change and twenty after. The gain should show mostly on held RFQs, where the missing-information reply goes out the same day.
2. Revision and material errors caught before sending. Count the times the estimator's check changed a field. A few catches means the gate works. Zero for a month means either the extraction is excellent or nobody is checking, so look at a sample.
3. Quote win rate by line. If faster quotes win more work, you will see it here. If win rate falls, check whether the break tables or the margin rule changed what you quote.

As a rough sum, if an estimator handles 15 RFQs a week and the clerical part falls from 18 minutes to 6, that is 3 hours a week returned to estimating. At the $95 an hour in the example, it is about $285 of capacity a week, though the real gain is quotes going out the same day. These are example figures, so put in your own.

Rolling it out in the shop without upsetting the estimator

Start with one estimator and one kind of RFQ, usually repeat-type milled parts, which have the least variety. Run the assistant in parallel for two weeks: the estimator quotes as usual, and compares the draft after. Note where it differs. You will find the rate card gaps (a missing programming line, a handling percentage nobody wrote down) before a customer does.

Then switch to the draft-first order. Keep the estimator's initials on every quote, as before. When you are confident, add the second RFQ type. If you later want dedicated quoting software, what AI quoting software costs a machine shop shows the price range, and the written rate card you built carries across.

If you would rather map your own quoting steps with someone who has set this up before, that is what an AI implementation consultation covers, at the contact page.

Questions shops ask before trusting an AI draft quote

Can AI read a drawing and price the part by itself?

Not reliably enough to send a number. AI can read a title block, a material note and a tolerance callout, but it misses features, misreads dimensions and cannot judge how hard a part is to hold. Use it to draft the paperwork around an estimator's decisions, and keep the drawing review human.

Which AI tool should I use for RFQ emails?

Any of the mainstream assistants (ChatGPT, Claude, Gemini or Copilot) can extract fields from an email into a table. The tool matters less than the template, the rate card you give it and a rule that nothing leaves the building without an estimator's initials. Check your plan's data settings first.

Is it safe to paste customer drawings into an AI tool?

Only if the customer's NDA allows it and the drawing carries no export-control or secrecy marking. When in doubt, paste the email text and a typed summary of the part rather than the drawing file, or ask the customer. Check the data settings of the plan you use.

How long should a draft quote take once this is set up?

For a simple repeat-type part, the extraction and draft take a few minutes, and the estimator's check takes the time it always took to judge the part. The saving is the retyping, the rate-card lookups and the chasing of missing details, not the thinking.

What if the RFQ has no drawing at all?

Do not price it. Send the missing-information reply instead, asking for a drawing or model, revision, material, quantity and finish. A quote built on a description alone commits you to a part nobody has seen.

Further reads

• AI for Small Manufacturing Companies: Build a Weekly Production Schedule — Turn won quotes into a weekly shop schedule.
• Capture Tribal Knowledge With AI Before Your Best Machinist Retires — Write down the estimator knowledge that lives in one head.
• Capacity Planning Template: Can You Take That Order? Check With AI — Check you can actually make the order before you price it.
• How to Use AI in a Machine Shop: Five Jobs Beyond Quoting — Other shop jobs where AI saves time.
• AI Systems for Manufacturing: MRP Add-On or Chatbot on Your Own Data? — Decide between an MRP add-on and a chatbot on your data.
• AI Tools for Welding Shops: Quotes, Drawings and Job Scheduling — A welding shop owner's guide to AI on the office side: a drawing-based quote check, cut lists, a schedule stress test and where weld sign-off stays human.
• Manufacturing Cost Estimating: How AI Helps Price a New Product — One new steel product costed end to end with AI help: materials, labour minutes, overhead, scrap, tooling, margin and what moves the price.
• WIP Tracking for a Job Shop: A Spreadsheet With AI Status Updates — A two-tab spreadsheet for jobs and operations, the formulas that flag late work, and prompts that draft customer updates from the daily export.
• Wholesale RFQ Response Emails: AI Drafts From Your Price List — How a wholesaler can have AI draft quote replies from its own price list, with price breaks in a table, an approval step and a ten-RFQ test.
• Can AI Quote an Injection Moulding Job? What an Engineer Must Check — Use AI to lay out an injection moulding quote, then see which four figures an engineer must confirm before it is sent.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: worked figures are illustrative shop numbers, not vendor data; AI tool data settings were checked against each vendor's own help pages.