AI for Manufacturers Watches the Machine. The Order Dies in the Inbox

At four o'clock on a Friday, Tania is at her desk above the floor of her sheet metal and fabrication shop in Dandenong South, reading an email she has read a version of before.

"Thanks for the quote. We've gone with another supplier on this one."

Her machines had a good week. The laser ran two shifts. The new press brake's controller logged every bend. A vendor had spent Tuesday morning showing her how a sensor on the spindle would predict a bearing failure weeks before it happened.

None of it noticed that the drawing for that job arrived eight days ago, sat in the shared inbox for three of them, and then waited another four for her estimator, Col, who was also running the floor that week. The customer did not choose the other shop because it was cheaper. They chose it because it answered first.

A factory measures every minute a machine is idle. It never measures the days a quote is idle.

Nobody decided to be slow. The shop simply watches the floor, never the front office. That is the gap in how AI for manufacturers is being sold: predictive maintenance, vision inspection, cobots, digital twins, all of it aimed at a process that is already running. I build these systems for Australian businesses, so let me say the uncomfortable half out loud. Most small manufacturers are not short of capacity. They are short of the one person who can read a drawing and put a price on it.

What AI for manufacturers gets right

Give the floor tools their due, because the case for them is real.

If an unplanned breakdown on your one laser stops the whole shop, a sensor that warns you early is worth having. A camera that checks every part against the drawing does not get tired at the end of a twelve-hour run. For a high-volume line, these pay their way.

But notice what all of it touches: the machine. The industry builds there because a spindle produces clean numbers and a factory floor makes a great demo video. A request for quote is a PDF, a vague email and a phone call, and nobody films that.

The work you lose before it reaches the floor

Now walk the same week from the front office.

A job shop lives on quotes. Every one needs someone to open the drawing, work out the material, the cut time, the bends, the welding, the finish and the outwork, check what the last similar job actually cost, and write a price the shop can live with. In most small manufacturers that someone is one or two people, and they are also the ones everyone runs to when something goes wrong on the floor.

So quotes queue. The easy ones go out in a day. The awkward ones wait. And a buyer with three suppliers on the email usually starts with whoever replies first.

Run it as arithmetic. Say you receive forty requests a month and win a quarter of them. Say a faster reply wins you three more jobs a month at an average of four thousand dollars. That is twelve thousand dollars of monthly work that was already in your inbox. Those numbers are invented and yours to replace. The shape survives any honest figure.

Your drawings folder already knows how to price most of your quotes. Nobody has time to ask it.

Start with the quote queue, not the spindle

Pick one workflow, not the whole factory.

A system reads every incoming request as it lands: the drawing, the quantity, the material, the finish, the date the customer needs it. It finds the closest jobs you have already made, pulls what they actually cost from your job costing, not what you quoted, and drafts a price with the working shown. Col opens a quote that is eighty per cent done, checks the parts he knows are tricky, adjusts, and sends it the same day.

When a drawing is genuinely new, the system says so and puts it at the top of his pile instead of pretending.

That gives you two numbers each month: time from request to quote, and quotes won. The first one drives the second.

The due date that slips in silence

Then point the same loop at the orders you already won.

Your job board knows when material arrived late, when the powder coater pushed a batch back a week, when a job sat waiting for a machine. It does not tell the customer. So the first time the buyer hears the date has moved is when they ring to ask where their parts are, and that is the call that loses the next order.

A system that watches those same records can flag every job likely to miss its date the moment the risk appears, and draft the update to the customer for someone to send. Bad news delivered a week early reads as a reliable supplier. The same news delivered on the due date reads as a shop that did not know.

Customers forgive a late job. They do not forgive finding out themselves.

Automate the quoting, never the engineering

Here is the line that keeps this safe.

No system decides whether a part can be made as drawn, whether a weld is structural, whether a tolerance is realistic, or whether a job is worth taking at that price. Those calls sit with the people whose name is on the work. A model that confidently misreads a note on a drawing will cost you more than any job it won.

The system reads, finds and drafts. Your estimator checks, prices and signs.

What AI for manufacturers costs

Work down this list in order and stop the moment something works.

• The tools already inside your job management or ERP software. Quote templates, job costing reports, due-date alerts. Usually half configured. Turn them on before you spend anything.
• A quoting add-on or estimating subscription, priced per user each month. Cheap enough to trial on the next month of requests.
• A custom build that reads your incoming requests and past jobs, drafts the quotes and the customer updates, and writes back into the software you already run. A few thousand up to the mid teens of thousands, depending on how many systems it must talk to.

Hold that against the jobs you lost last quarter to a faster reply. What catches owners out is the running cost rather than the build, and I broke those layers apart in what AI actually costs a small business. If you have already trialled something that never left the demo stage, why your AI pilot is stuck is worth ten minutes first.

The Australian layer: shutdowns and other people's drawings

Two things separate a system built for an Australian manufacturer from an overseas template.

The first is the calendar. Many Australian shops close over Christmas and early January, and buyers try to lock in orders before the break. A quoting backlog that builds in November does not just lose jobs. It loses the ones that fill your first quarter.

The second is the drawings. They are your customers' intellectual property, and some arrive under confidentiality terms, or with defence or export restrictions attached. Ask where they will be stored, whether they train someone else's model, and whether they ever leave Australia. Get those answers in writing before anything reads your inbox.

Common questions about AI for manufacturers

What should a small manufacturer automate first?

The quote queue. It runs on requests and job history you already hold, and two numbers measure it inside a quarter: days to quote, and quotes won.

Will AI replace estimators?

No. Reading a drawing for what can go wrong is the skill. Finding the last similar job and doing the arithmetic around it is not.

Is predictive maintenance worth it for a small shop?

Sometimes, if unplanned breakdowns are genuinely what costs you. Check your records first. If your problem is lost quotes, a machine that never stops just waits for work faster.

Go back to Tania. She did not lose that job because her laser was slow or her price was high. She lost it because the drawing sat in an inbox for a week while the one person who could price it was busy keeping the floor running.

If you want a straight read on what your quote queue lost this year, that is what a first call is for. Book a strategy call and bring your last three months of quote requests and which ones you won. I will count the days with you before we talk about building anything.