How to use AI in manufacturing: from quote to delivery

AI is used in manufacturing for five jobs above all: preparing quotes, answering the question of where an order is, keeping quality inspection records, matching delivery notes, invoices and certificates, and gathering dealer and customer requests in one place. Most of this work happens in the office rather than on the shop floor. Machine settings, prices and quality sign-off stay with the operators, engineers and sales leads who own them today.

Adoption is rising quickly, but small firms are behind. Eurostat's figures for 2025 show that 19.95% of EU enterprises with ten or more employees used at least one AI technology. The share was 17% for small enterprises, 30.36% for medium-sized ones and 55.03% for large ones. Turkey shows the same gap: TurkStat's 2026 figures, published on 2 October 2026, put AI use at 14% of enterprises with ten or more employees, rising to 37.1% among those with 250 or more staff and falling to 12.8% among those with 10 to 49.

The more telling figure is what manufacturers use it for. According to Eurostat, manufacturing firms in the EU that used AI in 2025 applied it mostly to marketing and sales (30.41%) and to organising business administration and management (27.05%). In other words, AI tends to earn its place first in sales and the back office of a factory.

Which manufacturing jobs can AI help with?

1. Quotes: material, labour and subcontracted lines are taken from the customer's drawing and quantity, and a draft quote is priced from your rate sheet.
2. Order status: the stage of every order, from materials to dispatch, is visible on one screen.
3. Quality inspection records: inspection forms are filled in on a tablet, non-conformances are logged with a photo and corrective actions are followed up.
4. Paperwork: delivery note, invoice and certificate details are matched to the order, and anything that does not match is flagged.
5. Dealer and customer requests: questions on stock, lead times and prices from email, WhatsApp and the phone land in one queue.

How can AI shorten quote preparation?

In a small manufacturer, quoting is often one person's job. The sales lead opens the drawing, works out the material and its weight, asks the shop floor how long each operation takes and asks suppliers to price painting, galvanising or heat treatment. Everything ends up in a spreadsheet. This can take days, and while the customer waits they ask other firms for a price too.

The setup starts by putting your own data in one place: the rate sheet, machine hourly rates, material prices and subcontractor prices. When a request comes in, material, dimensions, quantity and operations are read from the email and the drawing and turned into a list of line items. Each line is priced from those tables. A line with no match in the tables is never filled with a guess; it is left blank and flagged.

Reading drawings needs care. AI can read dimensions and notes, but an engineer checks tolerances, material grade and surface treatment. The draft then goes to the sales lead's approval queue, and the sales lead decides discount, lead time and payment terms. The approved quote goes to the customer and is attached to their record. We describe this on our document and report generation page.

If discount decisions always rest on the same criteria, such as customer type, quantity and payment terms, those criteria are written down once. For each new quote a suggestion is prepared against them, with the reasoning shown, and the sales lead still makes the call. See repeatable decision support for the detail.

What does a setup look like in a workshop?

Take an imagined example. A metalworking shop receives a technical drawing and a quantity by email. A draft list of materials and operations is produced from the drawing, and a quote is prepared from the rate sheet. The sales lead checks it, sets the discount and sends it.

When the customer confirms, a work order is opened. The customer is notified when the material arrives, when production starts, when the parts pass inspection and when the shipment leaves. Before calling the customer, the sales lead can already see which stage the order has reached.

How is the order status question answered?

When a customer asks about an order, most firms run the same chain. The customer calls sales, sales calls the production supervisor, and the supervisor walks over to the shop floor. The answer arrives hours later, and another quote waits in the meantime.

In the setup, each order gets a work order with defined stages: awaiting material, in production, in inspection, ready to ship, shipped. The stage comes either from a one-tap update on a phone or tablet on the shop floor or from your ERP. If the ERP has no external connection, we work from exported files.

When the stage changes, the customer is notified by email or WhatsApp. If a delivery date is going to slip, the alert goes to the sales lead first, and the sales lead decides what to tell the customer. We describe this flow on our workflow automation page.

How do quality records move off paper?

In many workshops the inspection form is still on paper. Non-conformance reports sit in a folder, and the follow-up of corrective actions lives in someone's memory. When an audit approaches, records are hunted down one by one.

In the setup, inspection forms are filled in on a tablet or phone. If a measurement is out of tolerance, a non-conformance is opened automatically and a photo is attached. The owner, the corrective action and the closing date stay on record, and overdue items are flagged to the quality manager. Voice notes left by operators are transcribed and added to the record. At the end of each month, non-conformances are grouped by product, machine and supplier, so a recurring fault shows up in the table.

During an audit, a record can be found by batch number, customer or date. ISO published a new edition of its quality management standard, ISO 9001:2026, this year. ISO itself does not issue certificates; certification is carried out by independent certification bodies. Accepting or rejecting a product remains the quality manager's decision.

Computer vision systems that look for surface defects with cameras also exist. They need a separate setup for each line and many sample images. For a small manufacturer, the records usually come first, and computer vision can follow if it is needed.

What can AI do for delivery notes, invoices and certificates?

The delivery note, the invoice and any material certificate for a shipment all belong to the same order, yet they often live in different programs and details are copied between them by hand. In Turkey, many firms issue delivery notes and invoices electronically through systems run by the Revenue Administration (GIB); which documents a firm must issue electronically is a question for its accountant.

In the setup, order, delivery note and invoice records are matched on the order number. A record whose quantity, unit price or customer name does not match lands on a list and goes to accounts. Material certificates that suppliers send by email are read and linked to the purchase order and batch number, so when a customer asks for the certificate of a shipment, it opens from the record instead of being searched for.

This setup does not issue invoices or submit anything to the tax authority. Invoices and delivery notes continue to be issued through your accounting software or e-invoicing provider. We cover the detail on our integrations and invoice automation pages.

How are dealer and customer requests handled?

Dealers and regular customers ask similar things through the day: stock levels, lead times, the current price list, a copy of a delivery note. The questions come by email, WhatsApp and phone, and answering usually means someone in sales opening yet another program.

These questions are gathered in one queue. Where the answer is in a record, a reply is drafted: stock from the ERP, order stage from the work order, a delivery note copy from the document archive. Price negotiation, new product requests and complaints always pass to the sales lead, and the other side is told clearly that they are writing to an assistant. We describe the setup on our request handling page.

Where do drawings, prices and customer data stay?

Technical drawings and price lists are trade secrets, and the worry shows in the statistics. Among EU enterprises that had considered AI but did not use it in 2025, 48.83% cited concerns about data protection and privacy, according to Eurostat.

In the setup, only the question and the relevant document are sent to the model; the archive as a whole is never sent. Documents are not used to train models and access rights stay as they are in the source. The names, phone numbers and email addresses of customer and dealer contacts are personal data under the GDPR and, for Turkish operations, under the KVKK. Where the data is processed is agreed in writing before the build. We explain this approach on our AI integration page.

Where should AI not be used in manufacturing?

• Price and discount: a draft quote is prepared; the sales lead approves the price and discount.
• Accepting or rejecting a product: a non-conformance is logged; the quality manager decides.
• Machine settings and safety: machine set-up, tool changes and stopping work stay with operators and the safety officer.
• Numbers without a source: a price, lead time or stock figure does not go to a customer unless it is tied to a record.

Where should a manufacturer start?

Among EU enterprises that had considered AI but did not use it, 70.89% named a lack of relevant expertise, the most frequent reason in the Eurostat survey. Starting does not need a large project or an in-house AI team. The first question is which task takes the most time or loses the most customers. For one workshop it is quoting, for another the order status question, for another pulling records together before an audit. Start with that one task: build the workflow, measure it for a month, then move to the next.

ALTAI Digital starts with a discovery of about a week, covering your quote files, ERP and accounting software, quality forms and repeated tasks. We then prepare a demo, put your requirements in writing through a detailed form and build the real thing. For a single workflow, first live use usually opens within 4-8 weeks, and your timeline is confirmed in writing after discovery. Measurement and reporting continue after go-live.

Every part joins one panel, so quotes, orders, quality records and paperwork can be followed on one screen. We describe the sector setup on our manufacturing solutions page and bespoke programs on our custom software development page.