# AI for manufacturing

> In a small or mid-sized manufacturer, AI agents and automation apply best to the office work around production: quoting, order entry, scheduling communication, supplier follow-up, quality documentation and maintenance requests. Each needs the relevant data to exist in a system of record rather than in email and spreadsheets, and each keeps a person on pricing, commitments to customers, and anything that affects safety or quality sign-off.

Source: https://risendigital.io/industries/ai-for-manufacturing  
Publisher: Risen Digital  
Updated: 2026-09-21

## At a glance

- **Where it applies:** Quoting, order entry, scheduling communication, supplier follow-up, quality documentation and maintenance requests.
- **What has to exist:** Part, customer, supplier and routing data in a system the agent can read, and a defined home for each kind of document.
- **Where people stay:** Pricing and terms, commitments on dates, quality sign-off, and any decision about safety.
- **How to start:** One flow, usually quote to order, connected read-only first, with the estimator approving every draft.

Most of the time lost in a smaller manufacturer is not on the shop floor. It is in the office work around it: quotes assembled from email and spreadsheets, orders re-keyed into the ERP, suppliers chased by hand, quality records typed up after the fact. This page shows where AI agents, automation and custom software each apply, what data they need and where your people stay.

![A machine operator at work on a factory floor](https://risendigital.io/images/experience/factory-machine-1600.jpg)

## Where do AI agents, automation and custom software apply in a manufacturer?

In the office work around production. The six areas below each have a repetitive, documented pattern where an agent can read, draft or complete within limits. Production itself stays with your people and your machines.

| Area | What the agent or automation does | Where a person stays |
| --- | --- | --- |
| Quoting | Extracts part, quantity, material and revision from the request; drafts routing and material cost from past jobs and the supplier price file | Approves every quote; sets margin and terms |
| Order entry | Turns an accepted quote or a customer purchase order into an ERP order without re-keying; checks it against the quote | Resolves mismatches between the order and the quote |
| Scheduling communication | Answers date and status questions from the schedule; tells customers and the floor when a date moves | Decides priorities when jobs conflict |
| Supplier follow-up | Chases open purchase orders, confirms lead times, flags late material before it stops a job | Chooses suppliers; negotiates price |
| Quality documentation | Assembles certificates, inspection records and traceability from the job record; drafts non-conformance reports | Signs off; decides disposition |
| Maintenance requests | Takes a request from the floor, classifies it, attaches the machine history, raises the work order | Prioritises; decides anything touching safety |

Custom software enters where the systems you have cannot hold the data or the workflow: a quoting tool that knows your routings, a floor-facing screen for requests, a portal where customers see their order status without calling. Automation handles the fixed steps between systems. Agents handle the variable ones, where something has to be read and judged. Most useful flows use all three.

## What does a quote-to-production request look like today versus connected?

Today it crosses email, two or three spreadsheets and the ERP, with the same details typed several times and the drawing revision easily lost. Connected, the request becomes one record that carries from quote to order to the floor.

_Interactive on the web page: One request, two ways. Today a request for quote arrives by email with a drawing, is copied into a spreadsheet by sales, estimated in another spreadsheet by engineering, priced by purchasing over email, assembled into a document, accepted by email, re-keyed into the ERP, discovered by planning at the weekly meeting, and started on the floor against a print that may not match the quoted revision, with status questions answered by three people checking three systems. Connected, an agent extracts the request into a quote record linked to the drawing, drafts routing and material cost from past jobs and the current price file for an estimator to approve, acceptance creates the ERP order without re-keying and with the revision locked, and status is answered from the order record with a person pulled in only when a date moves._

## What data has to exist before AI is useful in a manufacturer?

Parts, customers, suppliers, routings and prices in a system the agent can read, and an agreed home for each kind of document. The agent can help fill gaps, but it cannot invent your routings.

- **Item master** Part numbers, descriptions, materials and current revisions, with old revisions marked as such.
- **Customer and supplier records** One record each, with contacts, terms and the price agreements that apply, not a mix of ERP entries and spreadsheet tabs.
- **Routings and past jobs** What operations a part goes through, on which machines, and what similar jobs actually took. Past jobs are what makes quote drafting possible.
- **A schedule in a system** Even a simple one. If the schedule is a whiteboard, the agent cannot answer a date question.
- **A home for documents** Drawings, certificates and inspection records filed against the part or the job, not in individual inboxes.

You do not need all of it perfect before starting. What you need is one flow whose data is good enough, and a plan for the rest. Quoting often goes first because past jobs and the price file are usually the best-kept records in the business, and because the estimator is a natural reviewer for every draft.

> **The read-only fortnight** Before any agent drafts anything, connect it read-only and ask it questions: what did we quote this customer last year, which jobs used this material, what is open with this supplier. The wrong answers tell you exactly which data to fix first.

## Where do people stay in an AI-assisted manufacturer?

On every price and every promise, on quality sign-off, on safety, and on the judgement calls when jobs compete for the same machine. The agent prepares those decisions; it does not make them.

The instinct in a smaller manufacturer is right: the business runs on relationships and on people who know what a part really takes. AI does not change that. It changes how much of their day those people spend re-typing, chasing and answering the same status question. The estimator still sets the margin, but from a drafted routing instead of a blank sheet. The planner still decides which job jumps, but the customer hears about it the same hour. The quality manager still signs, but the pack is already assembled.

- **Pricing and terms** The agent drafts the cost and shows the history; the person sets price, lead time and conditions.
- **Commitments to customers** Dates and promises go out under a person's name, even when the agent drafted the message.
- **Quality and safety** Certificates, non-conformance disposition and anything touching a safe working practice are signed by a qualified person.
- **Supplier relationships** The agent chases and reports; the buyer decides who gets the next order.
- **Exceptions** Anything the agent flags as unsure goes to a person with the evidence attached, and the reason is logged so the rules improve.

## How does a small manufacturer start with AI without disrupting production?

Start in the office, not on the floor. Pick one flow, connect the systems read-only, let the agent draft while the people who own the flow approve, and widen on evidence. Production is never in the loop until the office work is reliable.

1. **Choose the flow with the best data and the loudest pain** For most, that is quoting or order entry. Write down what done looks like and who approves.
2. **Connect read-only and ask questions** Two weeks of the agent answering questions from the ERP and the document store, with every wrong answer traced to the data behind it.
3. **Run in draft mode with the estimator or the order desk approving** Every quote, order or message is a draft until a person releases it. Track how often drafts go out unchanged.
4. **Grant narrow write access** Let the agent post the case types it gets right almost every time: a repeat order under a value limit, a status reply, a supplier chase. Everything else stays in the queue.
5. **Add the next flow** Supplier follow-up, then quality documentation, then maintenance requests, each reusing the integrations from the last.

We work with manufacturers this way because the shop floor cannot absorb a failed experiment. The office can, and the improvements there reach the floor as fewer wrong prints, fewer surprise orders and fewer interruptions for a status update.

## Frequently asked questions

**Q: Is AI only for large manufacturers with big IT teams?**  
A: No. The office work that AI agents and automation apply to, quoting, order entry, chasing suppliers, assembling quality packs, is the same in a small plant as in a large one, and the smaller plant usually has fewer systems to connect. What it needs is one system of record for each kind of data and a person who owns the flow.

**Q: Do we need a new ERP first?**  
A: Usually not. Most ERPs used by smaller manufacturers can be read and written through an API or an export, and the agent works alongside them. A new ERP is a separate decision; if one is planned, sequence the AI work so the integrations are built once against the new system.

**Q: Can an AI agent do our quoting?**  
A: It can draft a quote: extract the request, propose a routing from past jobs, price material from the current file, and present it for approval. The estimator sets the margin and releases the quote. Over time the drafts need fewer changes, but a person approving every quote is the design, not a temporary stage.

**Q: What about the shop floor and machine data?**  
A: Machine monitoring and predictive maintenance are their own projects with their own sensors and data. This page is about the office work around production, which is where most smaller manufacturers lose time first and where the data already exists. Once the maintenance request flow is in place, machine data has somewhere to go.

**Q: How do we keep quality records right if an agent assembles them?**  
A: The agent assembles the pack from records that already exist on the job: material certificates, inspection results, traceability. It does not create measurements. A qualified person reviews and signs, and the pack shows where each item came from so the review is fast and the audit trail is complete.

## Related pages

- [AI agent use cases](https://risendigital.io/guides/ai-agent-use-cases)
- [Reducing manual data entry errors](https://risendigital.io/guides/reducing-manual-data-entry-errors)
- [What is a business operating system?](https://risendigital.io/guides/what-is-a-business-operating-system)
- [AI for logistics and distribution](https://risendigital.io/industries/ai-for-logistics-and-distribution)