Can AI Quote an Injection Moulding Job? What an Engineer Must Check

AI can lay out an injection moulding quote and do the arithmetic once the inputs are right, but it cannot supply the four figures that decide the price. An engineer must verify tooling cost, cycle time, resin usage and scrap, because a plausible guess on any one of them can turn a profit into a loss.

The useful way to think about it: a quote sheet is a formula, and AI is good at formulas and layout. What goes into the formula comes from the tool, the machine, the resin datasheet and your own history. Take an illustrative job of 10,000 clips: on a 22-second cycle the piece price works out at about $0.153; on a chatbot's guessed 15 seconds it drops to about $0.119, 22 percent low, with every line on the sheet still looking tidy.

Follow me on Instagram@sagnikteaches

Why a confident wrong quote is the real risk

Chatbots do not hesitate. Ask one to price 10,000 moulded parts and you get a tidy answer with a unit price, a tooling figure and a delivery estimate. Nothing in the tone distinguishes a number it calculated from one it made up. The consequences are not hypothetical for businesses in general: a tribunal in February 2024 held an airline responsible for what its website chatbot told a customer, rejecting the argument that the chatbot was a separate entity. The principle carries over. Whatever goes out under your name is your commitment, however it was produced.

Connect on LinkedInSagnik Bhattacharya

Sheet-metal and machined work has the same trap, covered in what goes wrong when ChatGPT quotes machine shop jobs: a believable price built on a cut length and bend count the chatbot estimated rather than measured. In a demo by Paperless Parts, which sells quoting software, a custom GPT came in 18 percent short on the laser cut length of a sheet-metal enclosure and counted 8 of its 12 bends. Moulding has its own versions of those hidden inputs, and they are more expensive because tooling is paid upfront and cycle time repeats on every part.

Subscribe on YouTube@sagnikteaches

The quote sheet: what has to be on it

An injection moulding quote has two parts, the tool and the parts. Keep them separate on the sheet, because they have different owners and different risks.

Part price: five lines

• Resin. Part weight plus the runner share, times the price per kilogram, adjusted for any regrind you can reuse.
• Machine time. Cycle time divided by cavities gives seconds per part; multiply by the machine hourly rate.
• Labour. Operator time per hour of run, divided by parts per hour.
• Scrap. The extra cost of parts lost to start-up, rejects and trimmed material.
• Secondary and packing. Inserts, degating, assembly, labels, cartons.

Batch costs and tooling

• Setup. Tool change, material change, first-off approval.
• Tooling. Quoted by the toolmaker, charged upfront or amortised into the piece price, with ownership and maintenance stated.
• Margin and payment terms. Applied last, by a person.

Ask the AI to build this as a spreadsheet with visible formulas, not as a paragraph. Formulas can be audited; prose cannot.

Build a moulding quote sheet as a table I can paste into a spreadsheet.
Rows: resin, machine time, labour, scrap, secondary and packing, setup per batch, tooling, margin.
For each row give: the formula in words, the inputs it needs, and a blank cell for the input value.
Do NOT fill in any input values. Mark every input as one of: FROM CAD, FROM TOOLMAKER, FROM DATASHEET, FROM OUR HISTORY, ENGINEER JUDGEMENT.

The labels at the end are the important part. They turn a blank sheet into a list of who owns each number.

The worked example: 10,000 cable clips

Say a solar installer wants 10,000 polypropylene cable clips for its own accessory range. All figures here are illustrative, not real market prices. The CAD model gives a part volume of 6.9 cubic centimetres. Polypropylene is roughly 0.9 grams per cubic centimetre, so the part weighs about 6.2 g. The proposed tool is a four-cavity cold-runner tool with a runner share of about 3.2 g per shot.

The sum, line by line. Parts per hour: 4 parts per shot x 3,600 seconds / 22 seconds = about 654. Machine cost per part: $45 / 654 = $0.0688. Resin per shot: 28 g = 0.028 kg x $1.80 = $0.0504, so $0.0126 per part. Labour: $10 / 654 = $0.0153. Add scrap on resin and machine time: ($0.0688 + $0.0126) / 0.97 = $0.0839. Packing $0.01. Setup spread over 10,000 parts: $0.0135.

The total cost per part is $0.0839 + $0.0153 + $0.01 + $0.0135 = $0.1227. With a 25 percent markup the piece price is about $0.153, so the 10,000 parts bring in roughly $1,534 plus the tooling invoice. The run takes about 15.3 hours of press time. A quote this small looks trivial, which is exactly why it gets sent on a quick check. The next section shows how little it takes for it to go wrong.

What a wrong cycle time does to that price

Suppose a chatbot, with no knowledge of your wall thickness or mould cooling, assumes a 15-second cycle for the same part. Cooling is usually the longest and least visible part of a cycle, and it grows rapidly as walls get thicker: the standard cooling-time formula has the wall thickness squared in it, so a clip with a 2.5 mm boss can cool far slower than its thin walls suggest.

Rerun the sum at 15 seconds: parts per hour 960, machine cost $45 / 960 = $0.0469, labour $0.0104, scrap-adjusted machine and resin ($0.0469 + $0.0126) / 0.97 = $0.0613. Total: $0.0613 + $0.0104 + $0.01 + $0.0135 = $0.0952, a piece price of about $0.119 against $0.153. The quote is 22 percent low. Over 10,000 parts that is about $350 of revenue that never arrives, and the press actually runs about 5 hours longer than the quote allowed (15.3 hours against 10.4).

A 22 percent error on a $1,500 job is annoying. On a 500,000-part programme the same percentage is five figures, and the margin is usually lower than 25 percent, so a miss like this can erase it. Piece price in moulding moves almost directly with cycle time, which is why the cycle time cell is the one an engineer must own.

Quick sums like this are a skill in their own right; why AI is bad at maths explains what to re-check whenever a chatbot calculates a total itself.

The four figures, who checks them and how

Each row has a different failure and a different cure. Do not collapse them into "check the numbers". The engineer's note against each cell should say where the figure came from, not just that it looks right.

Where AI does earn its place in a moulding quote

None of this makes AI useless here. It handles the repetitive, text-heavy and arithmetic-after-inputs parts well:

• Extracting the requirements from the RFQ. Part number, annual volume, resin, colour, tolerance notes, packaging and delivery pattern. A prompt that lists unclear points saves a round of emails.
• Building and checking the sheet. Once inputs are in, ask it to recompute the total from your formulas and show every step. Compare with your spreadsheet. Disagreement means someone has a mistake.
• Generating what-if tables. Piece price at 5,000, 10,000, 50,000 and 100,000 parts, or with cycle times of 20, 22 and 25 seconds, so the customer discussion rests on sensitivities, not a single number.
• Writing the covering text. Terms, assumptions and exclusions, in your voice.
• Reviewing the finished quote. Asking it which RFQ requirements the quote does not address.

For the RFQ-to-draft workflow in machined work, the machine shop quote template shows a close cousin of this routine.

Here is a sample what-if prompt and an illustrative output:

Using these formulas and inputs, show piece price at 5,000, 10,000, 50,000 and 100,000 parts, and at cycle times of 20, 22 and 25 seconds. Show the working for the 10,000 / 22 s case only. Flag any cell where setup is more than 10% of the piece price.

Qty 5,000, 22 s: $0.170 (setup is 20% of price - flagged)
Qty 10,000, 22 s: $0.153
Qty 50,000, 22 s: $0.140
Qty 100,000, 22 s: $0.138
Qty 10,000, 20 s: $0.144
Qty 10,000, 25 s: $0.168

Treat output like this as arithmetic to verify, not as the answer. Recompute two cells by hand. If two match, the sheet's logic is probably sound; if either misses, find out why before using the others. This sample has a slip worth catching: at 10,000 parts the setup charge is about $0.017 of a $0.153 price once marked up, roughly 11 percent, so all three 10,000-part rows should have been flagged as well. The prices are right and the rule it was given was not applied, which is exactly the kind of error a quick hand check finds.

A resin swap the quote never noticed

Imagine the same clip enquiry arrives again later, this time from a security installer who wants a stiffer version for a camera mount. The RFQ says "glass-filled nylon". The earlier sheet is reused, and the AI, working from the template, keeps the polypropylene density and the 22-second cycle.

Three things changed at once. Glass-filled nylon is much denser, typically around 1.3 to 1.4 grams per cubic centimetre (one 30 percent glass-filled PA66 grade lists 1.37; check the datasheet for yours), so the part is about half as heavy again. The resin usually costs considerably more per kilogram, so get a current supplier price. And nylon must be dried before moulding (that same datasheet gives 80 degrees Celsius for two to four hours) and runs with a much hotter mould, so the cycle and the setup both change. The quote came out far too low, with every line looking tidy.

The source-label column would have caught it: RESIN was marked "FROM DATASHEET" in the template, and nobody refreshed the datasheet. A rule fixes this: when the resin grade changes, every cell labelled DATASHEET, ENGINEER JUDGEMENT or FROM SUPPLIER is blanked and refilled.

Other inputs an engineer should look at before the sheet is trusted

The four headline figures are not the whole list. A few smaller inputs routinely hide a surprise, and AI will not raise them unless you ask. Here is what each looks like in practice, with a sample question the engineer asks:

• Which press. A part with a large projected area needs more clamp force than a small one, and a four-cavity tool needs a press whose platens, tie-bar spacing and shot size suit it. If the quote assumes the 80-tonne press at $45 an hour but the tool belongs on a 150-tonne machine at a higher hourly rate, every machine-time line is wrong. Question: does this tool physically fit the press we priced?
• Runner type. A cold runner wastes resin every shot unless it is reground and allowed by the customer. A hot runner saves resin and shortens cycles but raises tooling cost and can bring colour-change delays. Question: is the runner in the quote the runner in the tool design?
• Regrind rules. Some customers forbid reground material, particularly for cosmetic or safety parts. Question: does the contract allow what the quote assumes?
• Inspection and tolerances. Tight dimensions can mean statistical process control, a first-article report or 100 percent checking. These cost time that does not appear in cycle time. Question: what inspection does the drawing demand, and is it priced?
• Colour and additives. Masterbatch, colour matching and special additives cost money and can lengthen purging between jobs. Question: is colour a standard or a custom match?
• Packing and delivery. Clips shipped loose in a carton cost almost nothing. Parts that must be bagged in fives and labelled add a labour line. Question: what exactly goes in the box?

Before and after: a vague RFQ reply and a useful one

When the RFQ leaves these open, the best use of AI is drafting the questions to the customer. A weak first reply from an unsupervised chatbot reads like this:

Thank you for your enquiry. Based on the information provided, we can supply 10,000 parts at $0.15 each with tooling at $18,000. Delivery in 10 weeks.

The same enquiry handled with the checklist produces a reply that protects the quote:

Thank you for your enquiry. To confirm pricing for 10,000 clips we need: the resin grade and colour reference, whether reground material is acceptable, the annual volume behind this order, the inspection level the drawing requires, and how the parts should be packed. Tooling will be quoted separately once we confirm four cavities and the steel specification. We will return a priced quote within two working days of your answers. [your name]

The second takes a few minutes longer and prevents the swap described above. It also tells the customer you know what moves the price.

The ten-minute engineer sign-off

This sheet sits on the front of every quote. It is the product of the engineer's checks, not of the AI. A filled-in example (illustrative):

QUOTE: 10,000 PP cable clips, 4-cavity tool    DATE: ________
Prepared by: ________ (AI-assisted layout; inputs entered by hand)

Tooling
[x] Written toolmaker quote attached: $18,000, ownership: customer
[x] Cavities, steel, expected tool life stated on quote

Cycle time (22 s)
[x] Source: comparable tool, same resin, wall 1.6 mm, thickest boss 2.5 mm
[x] Cooling time checked against thickest section

Resin
[x] Grade: PP copolymer, datasheet density 0.90 g/cm3 (attached)
[x] Runner share 3.2 g from tool design, regrind policy: not reused on this job
[x] Price confirmed with supplier this week

Scrap
[x] 3% from our last six jobs on similar parts

Arithmetic
[x] Two cells recomputed by hand: match

Assumptions printed on quote: [x]   Exclusions printed on quote: [x]
Engineer: ________  Approver (margin): ________

The two "recomputed by hand" ticks and the "assumptions printed on quote" tick are often the difference between a quote you can defend and one you cannot. If a customer questions the price later, the sheet shows how each figure was obtained.

Tooling needs its own rule

Tooling deserves a separate warning because it is paid before the first part and it is the hardest figure to put right afterwards. Never let a chatbot's tooling number reach a customer. Get a written quote from the toolmaker and have the engineer confirm that its assumptions match the part: the number of cavities, whether hot or cold runner, any side actions, texture, steel grade and the shot life it is designed for. Then AI can help you tabulate several toolmakers' quotes side by side, which comparing supplier quotes with AI covers, because you supply the quotes and it only reorganises them.

Also write down who owns the tool, who pays for maintenance and where it is stored. Those clauses matter more than the unit price of the tool when a customer moves production later.

Putting it to work without slowing the quote down

A workable routine for a shop quoting several jobs a week:

1. AI reads the RFQ and produces the requirement list and the blank sheet with source labels (5 minutes).
2. The engineer fills in the inputs they own and attaches the evidence (15 to 30 minutes, depending on novelty).
3. AI recomputes the total, runs the what-if table and drafts the covering text (5 minutes).
4. The engineer and approver complete the sign-off sheet (10 minutes).

Most of the saving comes from the first and third steps, which used to be copying and re-typing. The human steps do not shrink, and should not. If purchasing resin and components is the next bottleneck after a win, automating purchase orders and supplier emails is the natural follow-on, and the checks for any vendor who tells you their AI can quote for you are in evaluating an AI implementation proposal. If you would like help designing the sheet and the sign-off for your own shop, that is the sort of work an AI implementation consultation covers; you can book one here.

Questions from moulders about AI quoting

Can AI estimate injection mould tooling cost?

It can structure the estimate and list what drives cost, such as cavity count, steel, slides and texture. The price itself must come from a toolmaker's quote or your own history, because a chatbot has no current figures for your tool and a wrong tooling number sinks the whole job.

What is the biggest quoting risk in moulding?

Cycle time. Piece price moves almost one for one with it, and cooling time depends on wall thickness and resin. A chatbot's guess can be 20 percent out without looking odd. Take cycle time from a flow analysis, a similar tool or a trial.

Should the quote include scrap?

Yes. Allow for start-up scrap, rejects and the resin lost in runners if they are not reground. Use your own recent scrap rates by part type, not a generic percentage from a chatbot, and say what the quote assumes.

Can I paste a part drawing into ChatGPT?

Check your customer's confidentiality terms and your plan's data settings first. Even then, the chatbot reads the drawing rather than measuring it, so volume, wall thickness and feature counts should come from the CAD model.

Who should sign off an AI-built moulding quote?

A process or tooling engineer who knows your machines and resins, plus the person who owns the margin. Their initials and the date belong on the quote sheet, along with the source of each key figure.

Further reads

• AI Freight Quoting: Turn Customer Emails Into Draft Quotes — Another trade where customer emails become draft quotes.
• AI Supplier Management: Track Prices, Lead Times, and Risk — Track resin prices, lead times and supplier risk over time.
• How to Match Supplier Invoices to Purchase Orders Automatically — Match resin and tooling invoices to orders automatically.
• How to Review an AI Tool After 90 Days: Keep, Fix or Cancel — Decide after three months whether an AI quoting aid is working.
• AI Hallucinations Explained for Business Owners: Causes and Fixes — Why confident numbers from a chatbot can still be invented.
• Can AI Draft a Lockout Tagout Procedure? What Must Be Verified — Another shop-floor document where AI drafts and an engineer verifies.
• Scrap Rate Tracking: Group Reject Reasons From Shop Notes With AI — The scrap rate formula, reason codes grouped from free-text shop notes with AI, and a weekly Pareto list operators help you check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: RJG's breakdown of injection moulding cooling time (thickness squared in the cooling formula); Paperless Parts demo page comparing a custom GPT's sheet-metal quote with a human estimator's, https://www.paperlessparts.com/demos/llms-vs-manufacturing-aware-ai/ (read 5 November 2026); a Celanese Zytel PA66-GF30 datasheet listing (density 1.37 g/cm3, drying at 80 C for 2-4 hours); polypropylene density from public reference data; public summaries of the February 2024 tribunal decision holding an airline responsible for its website chatbot's wrong answer. Every price, rate and cycle time in the worked example is illustrative.