AI can extract the numbers a fabrication quote needs from a drawing: parts, lengths, weights, cut counts and weld lengths. It cannot decide the price. A fabricator must check the drawing revision, weld symbols, material grade, finish and lead time, then apply their own rates and margin before the quote goes out.
The temptation is to treat the extraction as the quote. It is not. A drawing gives you geometry and notes; a quote needs your shop's rates, your supplier's lead times, your risk appetite and a view on how the customer behaves. So split the work: let the tool read the drawing into a parts and weld list, then build the price yourself from rates, stock lengths and calendars it has never seen.
What a quote from a drawing is made of
Every fabrication quote breaks into the same components. If you set them out in the same order each time, you can tell a tool exactly what to extract and what you will add yourself.
| Component | What the drawing gives | What a tool can extract | What you add |
|---|---|---|---|
| Material | Section sizes, plate thickness, grade notes | Parts list with lengths, weights | Stock-length waste, current price per kg |
| Cutting | Part lengths and end angles | Number of cuts, angled cuts | Minutes per cut on your saw or profiler |
| Fit-up and welding | Weld symbols, joint types | Weld length by joint type | Minutes per metre, process, inspection |
| Finishing | Paint or coating spec, area | Surface area estimate | Subcontractor price and lead time |
| Handling and delivery | Overall size, weight | Heaviest sub-assembly | Lifting, packaging, transport |
| Risk and margin | Tolerances, certification notes | A list of anything unusual | Your allowance and your margin |
The pattern is the same as with other trades: extraction on the left, judgement on the right. The best use of AI is to fill the third column accurately and quickly, so you spend your attention on the fourth.
A main example: a machine guard frame, from drawing to price
Illustrative job: a free-standing machine guard frame in 40 x 40 x 3 mm square hollow section, four 150 x 150 x 6 mm base plates, powder coated. The drawing lists 28 parts and 28 welded joints. All rates here are invented for illustration; use your own.
Step 1: weight from geometry
Steel has a density of about 7,850 kg per cubic metre. A 40 x 40 x 3 mm hollow section has a cross-section of 40 x 40 minus 34 x 34, which is 444 square millimetres. That gives about 3.49 kg per metre. Published section tables show a little less, around 3.4 kg, because real tube has rounded corners, so check against your supplier's table, which is the number you will be invoiced on. Total length from the drawing is 14.0 m, so the sections weigh about 49 kg. Four base plates at 0.15 x 0.15 m is 0.09 square metres; 6 mm plate weighs about 47.1 kg per square metre, so about 4.2 kg.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowStep 2: buy by stock length, not by use
You do not buy 14.0 m of tube; you buy 6 m lengths. Three lengths is 18 m, or about 62.8 kg, and the 4 m remaining may or may not be reusable. This is the step beginners and chatbots both skip. Quoting on 49 kg understates the steel bill by more than a fifth. At an illustrative $2.00 a kilogram, three lengths cost about $126. Plate from an offcut or a small sheet share: $20. Material: $146.
Step 3: cutting, welding, finishing
Illustrative shop times: 28 cuts at 3 minutes each including marking and deburring is 84 minutes. Weld length is 28 joints, each welded all round its 160 mm perimeter, about 4.5 m in total. At 25 minutes per metre including fit-up, tacking, welding and clean-up, that is about 112 minutes. Grinding and prep: 50 minutes. Handling and packing: 30 minutes. Total: 276 minutes, or 4.6 hours. At a shop rate of $70 an hour, labour is $322.
Powder coating is outsourced, with an illustrative minimum charge of $60 for this size. Consumables (wire, gas, discs) are $18, and delivery is $35.
Step 4: cost, markup and lead time
Cost: $146 + $322 + $60 + $18 + $35 = $581. A 30 percent markup is $174, giving a price of $755, a 23 percent margin on price. Lead time: steel two working days, cut and weld one day, coating five working days at the subcontractor, plus a day to deliver: about nine working days from deposit. Quote ten.
That is a complete quote, apart from setup time, which the quantity table further down adds back. The tool's part in reaching it is narrower than it looks.
The extraction prompt, with its realistic output
Give the tool the drawing's text content, a parts table or a clean PDF, whichever your plan accepts, and ask for a structured list. Always request page references so you can check.
From this fabrication drawing, extract:
1. A parts table: part number, section or plate size, length, quantity.
2. Total length per section size.
3. A weld list: joint number, joint type, symbol as drawn (fillet, butt,
one side or both sides, all-around), weld size, estimated length in mm.
4. Finish specification as written.
5. Every note about tolerance, material grade, certification or inspection.
6. Anything ambiguous, with the drawing location.
Do not estimate prices. Do not fill gaps: write "not stated".An illustrative result: 28 parts across three part numbers; 14.0 m of 40 x 40 x 3 SHS; four 150 x 150 x 6 plates; 28 fillet welds listed with sizes; finish "powder coat, RAL as per customer, 60 micron minimum"; note "dimensions in mm, general tolerance +/- 2 mm"; ambiguous item "joint 17: symbol unclear, possibly all-around".
Your job: open the drawing at joint 17. If the symbol is all-around, the weld length for that joint is the full perimeter of 160 mm, not one side of 40 mm. The difference is 120 mm on one joint, but a tool that reads every symbol as all-around, or as single-sided, will be wrong by a large amount across a whole frame. Weld length is where most hours hide, so check every symbol on a drawing with more than a handful of joints.
Five ways a tool-assisted quote goes wrong
1. Wrong revision
A customer sends revision C but the PDF you uploaded was revision B, still in your folder. A tool will extract revision B with perfect confidence. Put revision and date in your prompt and compare them with the title block of what you uploaded.
2. Weld symbols misread
Arrow-side, other-side, all-around and field-weld symbols are small and easy to misread, in a scan especially. Check them against the drawing, one by one, when the job has more than ten joints.
3. Weight from nominal rather than actual sections
A tool may use a generic figure per metre. Replace it with your supplier's table. A 5 percent error on 62.8 kg is only 3 kg here, but on a three-tonne job it is 150 kg, or about $300 at the same illustrative price.
4. Missing standard or inspection requirements
If the drawing says "welds to be inspected to the customer's specification" and the specification is a separate document, the quote must cover it: weld procedure, welder qualification, non-destructive testing. A tool may extract the note without understanding the cost. Highlight any such note and price it yourself, or ask the customer for the document before quoting.
5. Quantity breaks assumed
Setup, programming and first-off inspection are one-off costs. A tool quoting ten pieces by multiplying one piece by ten undercharges the first and overcharges the rest. Ask for setup and run time as separate lines, and decide your quantity breaks yourself.
A mistake that cost a day's work
In an illustrative case, a shop asked a chatbot to total the cut lengths from a parts table and then price the saw time. The tool correctly summed the lengths but treated 12 mitred corners as plain square cuts. Mitres need a different setup and about twice the time each, so the saw labour was 36 minutes short. On its own that is trivial, but the same assumption carried through the fit-up estimate, since mitred joints need more careful tacking. The job lost most of its margin. A single line in the prompt, "list angled cuts separately and count them", would have shown the issue, and a ten-second look at the part table would have caught it.
Finishing and lead time: the lines AI cannot see
Many fabrication quotes are won or lost on lead time, and lead time is made of other people's calendars. Steel arrives when the supplier says; coating happens when the finisher has space; a heavy job waits for a crane. None of that is in the drawing.
Build a small table in your own words and keep it updated: supplier delivery in working days for common sections, finisher turnaround for powder coat, galvanising and wet paint, and your own queue length this week. A chatbot can then assemble a lead time from the table: "Steel 2 days, fabrication 1 day given queue of 3 days, coating 5 days, delivery 1 day, total 12 working days." It is a sum on inputs you maintain. If you leave the table out, it will invent a plausible number, which is the worst thing it can do to a delivery promise.
Ask the tool to state the lead time as a range and to list every assumption. Then add buffer from experience. A promised date missed is worth more in lost goodwill than a longer honest estimate.
A second job, scaled: a run of 12 brackets
Illustrative: twelve identical brackets from 8 mm plate, each 220 x 140 mm with four 12 mm holes and a bevelled edge. Parts are laser-cut by a subcontractor and welded in-house to a small upright. A chatbot can take the part dimensions and compute plate area: 0.22 x 0.14 = 0.0308 square metres per bracket, so 0.37 square metres for twelve before nesting. At 62.8 kg per square metre for 8 mm plate, that is about 23 kg. But the subcontractor charges by the sheet and machine time, not by the part area. Ask them for a price per piece for quantity twelve and quantity fifty; the chatbot cannot know either. The tool's contribution is the weight, the hole count and a neat request-for-quote email to the subcontractor. Yours is choosing the break point and your margin on bought-in work.
Setup versus run time: pricing one, ten and fifty
Customers often ask for prices at several quantities. Do the sum in two parts: setup time, paid once, and run time, paid per piece. Illustrative figures for the guard frame: setup (programming, jigs, first-off check) is 90 minutes, or $105 at $70 an hour; run time per frame is 4.6 hours of labour, falling to 3.9 hours each once a jig is built and the cuts are batched.
| Quantity | Setup | Labour | Materials and bought-in | Cost per frame |
|---|---|---|---|---|
| 1 | $105 | $322 | $259 | $686 |
| 10 | $10.50 each | $273 | $259 | $542.50 |
| 50 | $2.10 each | $273 | $250 (volume rate on steel) | $525.10 |
The materials and bought-in column here is the $146 of steel and plate plus $60 coating, $18 consumables and $35 delivery, all illustrative, with the volume rate on steel assumed to save $9 a frame at fifty. A chatbot is useful for laying this table out and recalculating it when you change the setup time. It cannot know whether your supplier really gives a volume rate, or whether your jig will work first time. Ask it to show each sum, and use your own figures for the three assumptions that matter: setup time, the learning-curve drop and the supplier discount.
Notice what the table says about pricing. A single frame at $686 cost is about 26 percent dearer than the same frame in a run of ten, and a customer who asks for one now and fifty later is asking you to carry the setup cost on the first. Some shops quote the single piece at the ten-off price when repeat work is likely, and say so in writing; others charge the setup as a separate line so the customer sees it. Either is a decision for you, and the table makes it visible instead of leaving it in your head.
The table also exposes a gap in the main example. The $755 price was built on $581 of cost with no setup line; once 90 minutes of setup is counted, the single frame costs $686, and the margin falls from 23 percent to about 9 percent ($69 on $755). On one-off work, either price the setup as its own line or decide on purpose to absorb it. A tool will not raise the question unless the prompt asks for setup and run time separately.
Asking the customer the right questions before you quote
A drawing often leaves gaps. Quoting round them is how shops lose money. A chatbot is a good way to turn your list of unknowns into a polite query email, and the list is often the most valuable output of the extraction step.
Write a short, polite email to a customer with these questions about a
fabrication drawing. Keep each question to one line and number them.
Questions: confirm the revision; confirm the finish colour and thickness;
confirm whether weld inspection is required and to which document;
confirm whether certificates are needed for the steel; confirm delivery
address type (loading dock or kerbside); confirm annual quantity if repeat.
Sign off as [your name]. Under 140 words.The customer's answers change the price. "Kerbside delivery" may add a vehicle with a lifting arm; "certificates required" adds a cost per batch from the steel supplier; "annual quantity 200" changes your quantity breaks. Ask before you price, not after. A reply that arrives with four answers saves a revised quote and a bruised relationship. If the customer will not answer, state your assumptions in the quote itself, in writing, so the price is clearly conditional.
Turning the numbers into a quote the customer reads
Once the figures are settled, a chatbot can draft the customer-facing quote: scope, material grade, finish, quantity, price, lead time, what is excluded, validity. Give it the facts and tell it not to add promises. Add an expiry date and a price-rise clause using the wording in how long a quote is valid and what to say about price rises, because steel prices can change between quote and order. For tools aimed at shops like yours, the tools for welding shops tutorial goes into software options, and checking fabrication drawings against the specification covers the drawing side in more depth.
The checks that stay with you
- Drawing revision and date against the customer's latest.
- Every weld symbol on jobs with more than a handful of joints.
- Material grade and certification requirements, and whether you can meet them.
- Stock-length purchasing, not net length.
- Angled cuts, bevels and holes counted separately.
- Inspection and test requirements priced in full.
- Lead time built from your supplier and finisher calendars.
- Price, margin and the decision to quote at all. Some jobs are not worth winning.
An illustrative checking routine takes ten minutes on a simple drawing and thirty on a complex one: revision, weld symbols, stock lengths, special requirements, then a read of the finished quote as the customer would. If you cannot explain every line to the customer, you are not ready to send it.
Testing a tool on a drawing you have already priced
Take a finished job with a final cost record. Strip your own notes, give the drawing to the tool, and compare its extracted weight, cut count and weld length with what you measured. Say it gets weight within 3 percent, cut count exactly, and weld length 18 percent low because it missed two all-around symbols. That tells you the tool is good for weights and counts and needs a symbol check. Repeat on a different kind of drawing, such as a welded assembly with machined parts, before you depend on it. Do not test on a live tender, where a mistake costs real money and the answer arrives too late.
If your drawings come from several customers, expect big variation in quality. Clean CAD exports read well; photographs of hand-marked prints do not. A fair rule is to use the tool on the first kind and do the second by hand. For other ways AI can support a small workshop, the AI hub lists related tutorials, and building a rate sheet for welding jobs shows how to set the hourly and machine rates a quote like this depends on.
Keep customer drawings confidential. Check the terms of the customer's order or non-disclosure agreement before uploading anything, remove titles and part numbers where you can, and use a plan whose data terms you have read. If the terms are unclear, ask the customer.
Fabricators' questions about AI and drawing-based quotes
Can AI read a fabrication drawing accurately?
It can read clean, text-based or well-scanned drawings reasonably, but it misreads weld symbols, tolerances in notes and hand-marked changes. Treat what it extracts as a first-pass parts and weld list, then check every number that drives cost against the drawing yourself.
What should I check before sending an AI-assisted quote?
Drawing revision, material grade, quantities, total weight, number of cuts, weld length and type, finish specification, inspection requirements, outsourced lead times and exclusions. Add your own price, margin and lead time. Nothing the tool produced should go out unread.
Does AI know my shop's rates?
No. Your hourly rate, machine rates, scrap and rework history, and subcontractor prices are yours. Give them to the tool as inputs and ask it to apply them, showing the sums. A rate it suggests on its own is a guess, not a benchmark.
Is it safe to upload customer drawings to a chatbot?
Check the customer's confidentiality terms first. Many drawings are restricted, and some contracts forbid uploading them anywhere. Use a plan with data controls, remove titles and part numbers where you can, and ask the customer if the terms are unclear.
Further reads
- AI in Steel Detailing: Gains for a Small Office, and Tasks to Avoid — Where AI helps a small steel detailing office, and where it does not.
- Best AI for Welding: Can It Keep Weld Procedure Records in Order? — Keeping weld procedures and material certificates in order.
- AI for Welders: What a Welding Business Can Hand Over Today — Which welding-shop jobs AI can take on today.
- AI Software for Construction Drawings: Catch Changes Between Revisions — Catching changes between drawing revisions.
- Markup vs Margin in Construction: A Pricing Sheet Built With AI — Keeping markup and margin straight on every quote.
- AI-Generated Welder Website Marketing: Case Study Pages for Custom Work — Case-study pages that show off custom work.
- Countertop Fabricator Software: Templating, Slabs and Where AI Helps — A vendor question list for templating, slab layout, quotes and cut files, with a scorecard, a yield sum and where ChatGPT helps with quote wording.
- Welding Marketing: Website, Google Profile and Posts Drafted With AI — A welding marketing kit: five job-type pages, a Google Business Profile for a mobile or workshop welder, and weekly posts from real before-and-after photos.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: steel density (about 7,850 kg per cubic metre) from standard engineering references; worked example uses illustrative rates, prices and times for the reader to replace. No vendor pricing is quoted.