AI in Manufacturing: 12 Use Cases That Run on Your ERP Data

AI in manufacturing pays off first in planning, purchasing, quality and maintenance, where a model reads records you already keep and suggests the next move. Here are 12 use cases, the ERP data each one needs, and a start-small plan.

A factory floor where AI runs on ERP data: a large production machine with a control screen sits in the centre with an AI core glowing above it. A stock rack full of boxes, work-order sheets on a table, a quality gauge and a pallet of finished goods surround it, an ERP cabinet holds the records, and the plant head stands in front with a tick to approve the AI's suggestions
Published 10 min read
Large US manufacturers using AI or ML at facility or network level (Deloitte, 2025)
29%
Piloting generative AI in the same survey
38%
Use cases in this guide, each mapped to ERP data
12
Use cases in the first three months of our start plan
3

Where things stand

How is AI used in manufacturing today?

In short

Most of the manufacturing industry is still early with AI. Every use case that works runs on ERP data: sales orders, BOMs, work orders, job cards, stock ledgers and inspection readings. If that data is patchy, the AI is guessing.

In Deloitte's 2025 Smart Manufacturing and Operations Survey of 600 executives at large US manufacturers (surveyed August to September 2024, published 1 May 2025), 29% said they use AI or machine learning at the facility or network level. Another 23% were piloting it. For generative AI the numbers were 24% deployed at that scale and 38% piloting.

What smart manufacturing delivered, on average

Deloitte's survey also reports what smart manufacturing programmes as a whole delivered since implementation. Those figures cover the whole smart-factory push, cloud, analytics, IoT and AI together, not AI on its own.

Read them as a ceiling for a well-run programme, not a promise for one tool.

10-20%
Production output
7-20%
Employee productivity
10-15%
Unlocked capacity

Vendors are moving towards the ERP

1

Anthropic: Claude for Manufacturing

A session scheduled for 17 November 2026 shows Claude drafting the quote, the non-conformance report and the morning report from plant records, with the plant's experts approving them. The page is explicit that nothing writes to production control or a system of record on its own.

2

IFS and Anthropic: Resolve

Announced on 13 November 2025, an industrial AI tool for maintenance that reads schematics, sensor data, images and technician notes, then warns of likely failures and suggests repairs.

Notice the pattern in both: AI prepares, a person approves, and the ERP stays the record.

AI use cases in manufacturing

Which AI use cases in manufacturing pay off first?

Our pick

Start with the ones where your data is already clean. For most plants that's the daily morning report, cost variance alerts and non-conformance report drafts. These are the 12 we see most often in ERPNext projects.

#Use caseERP data it needsWho decides
1Demand forecasting2-3 years of sales orders and invoices by item, customer and monthSales and planning head
2Production planning and schedulingProduction Plans, Work Orders, BOM operations, workstation hoursPlanner
3Reorder and purchase suggestionsStock ledger, reorder levels, supplier lead times, BOMsPurchase
4Supplier delivery and price watchPurchase Orders, receipt dates vs schedule dates, price historyPurchase
5Quality trend analysisQuality Inspection readings and rejections by item, supplier and batchQuality head
6Non-conformance report draftsRejected inspections, job cards, batch numbersQuality head
7Predictive maintenanceDowntime Entries, asset maintenance logs, machine sensor dataMaintenance head
8Cost variance alertsBOM cost vs actual material and operating cost on manufacture entriesFinance and plant head
9Quote and estimate draftsBOMs, past quotations, item prices, operation costsSales and estimating
10Daily morning reportWork Orders completed, job cards, downtime, shortagesPlant head
11Shop-floor assistantLive Work Order status, job cards, stock levels, SOP documentsSupervisors
12Document reading for inward paperworkSupplier invoices, delivery challans, test certificatesAccounts and stores

Planning: forecasting, scheduling and reorders

1. Demand forecasting

Reads order history by item and month, spots seasonality and customer patterns, and proposes next quarter's numbers. It needs clean item codes. Say one product sits under three codes because someone created duplicates. Then the forecast splits its demand three ways.

2. Production scheduling

ERPNext's Production Plan already pulls items from Sales Orders or Material Requests, explodes the BOM, and creates Work Orders and Material Requests. AI adds judgement: a sequence that cuts changeovers, a workstation booked past its hours, which order to slip when a rush job lands.

3. Reorder suggestions

ERPNext already nets projected quantity against BOM requirements. An AI agent reads actual consumption and supplier lead times and says, "this item's reorder level is set for last year's volume."

Purchase and supplier risk

4. Supplier delivery and price watch. Compare each Purchase Order's schedule date with the actual Purchase Receipt date, and you have a supplier reliability score you didn't have to build by hand. Add price history and rejection counts, and the AI can tell purchase which supplier is slipping before the line stops.

Typical mistake: back-dated receipts. The supplier score only works if the Purchase Receipt is entered on the day goods arrive. Back-dated entries hide late deliveries.

5. Quality trend analysis

ERPNext's Quality Inspection records readings against minimum and maximum values, or a formula, for incoming, in-process and outgoing goods. After a few months of readings, AI finds the pattern a person misses: one parameter drifting towards its limit on one supplier's batches, or rejections clustering on the night shift.

6. Non-conformance report drafts

A good first job for generative AI in manufacturing. An AI agent collects the failed readings, the job card, the batch and the supplier, and drafts the report. Your quality head edits and signs. That's the "AI drafts, expert decides" split Anthropic describes for its manufacturing work.

Maintenance

7. Predictive maintenance gets the most attention and needs the most data. Inside the ERP you have Downtime Entries and asset maintenance logs. Real prediction usually also needs machine signals from sensors or a PLC, which most mid-size plants don't yet collect.

IFS's press release on Resolve gives a sense of the starting point at one plant. If you can't say what your own figure is, start by recording every downtime with a reason in the ERP. That record is the training data later.

38%
Repairs that were emergency, not proactive, at William Grant & Sons before Resolve (IFS)

Data predictive maintenance needs

  • Downtime Entries with reasons
  • Asset maintenance logs
  • Vibration
  • Temperature
  • Motor current

Cost and commercial

8. Cost variance alerts

Compares the standard BOM cost with what a Work Order actually consumed, including process loss, and flags the outliers each morning. A person still asks why, but they ask on day one instead of at month-end close.

9. Quote and estimate drafts

Pulls the BOM, current item prices, operation costs and the last few quotes for the same customer, then builds a first draft. Estimating checks margins and lead times before it goes out.

Shop floor and paperwork

10. The morning report

The easiest win on this list. The AI reads yesterday's completed Work Orders, open job cards, downtime and material shortages, and writes one page for the plant head before the 9 a.m. meeting. Nobody spends an hour building it in a spreadsheet.

11. Shop-floor assistant

Answers plain questions: "Where is Work Order 1042?", "How much of this resin is left in the main store?", "What's the SOP for this changeover?" It reads; it doesn't post entries.

12. Document reading

Turns supplier invoices, delivery challans and test certificates into draft ERP entries. Stores and accounts review and submit. This removes a lot of retyping, and it's where FinByz's own DocVision product sits.

Data first

What data must be in the ERP before AI helps?

Short answer

The transactions that describe what actually happened on the floor, entered on the day they happen. In our implementation work, the gaps that block AI projects are rarely about the model. They're about records.

These are the usual ones:

  • Job cards not closed. Without time logs per operation, scheduling and cost variance have nothing to learn from.
  • Material issued in bulk. If raw material moves to the floor in one weekly transfer instead of against Work Orders, consumption per order is invented.
  • Duplicate items and loose BOMs. Forecasting and reorder suggestions inherit every duplicate.
  • No downtime reasons. "Breakdown" as the only reason tells a model nothing.
  • Inspection readings on paper. Quality analysis needs the numbers in Quality Inspection, not a scanned sheet.

Fix these first. It isn't glamorous, but it's also what makes your ordinary ERP reports trustworthy, so the work pays off even before any AI is switched on. Our guide to the benefits of ERP for manufacturing covers the base setup.

Integration

How do you integrate AI into existing manufacturing processes?

In short

You add AI beside the ERP, not inside every screen: an AI agent with read access to ERP data through the API, a short list of jobs it may do, and a person who approves anything that changes a record. For the architecture, see our piece on deploying enterprise AI agents on ERP.

A few rules we follow:

Read first, write later

Start with agents that only read and report. Let them create draft documents (a Material Request, a Purchase Order) only once people trust the drafts.

Use the ERP's own permissions

The agent logs in as a user with a role. If the purchase clerk can't see salary data, neither can the purchase agent.

Keep a log

Every suggestion and every approval is recorded, so you can check later why a decision was made.

One number per use case

Hours spent on the morning report, stock-outs per month, emergency repairs as a share of all repairs. Pick it before you start.

Start small

How should a small or mid-size manufacturer get started?

Here's the plan we suggest to plants with 50 to 500 people. It assumes ERPNext, but the order works on any ERP.

  1. 1Weeks 1-2: data checkPull 3 months of Work Orders, job cards, stock entries and inspections. Count how many are complete and entered on time. That tells you which use cases are possible now.
  2. 2Weeks 2-4: one reporting use caseThe morning report or cost variance alerts. Read-only, low risk, visible every day.
  3. 3Month 2: one drafting use caseNCR drafts, quote drafts or document reading. A person approves every draft.
  4. 4Month 3: one planning use caseReorder suggestions or schedule suggestions, compared side by side with what the planner did.
  5. 5ReviewKeep what moved its number. Drop what didn't. Then fix the next data gap and add the next use case.

Three use cases in three months, each measured, beats a factory-wide AI plan that never leaves the slide deck.

Limits

When is AI in manufacturing the wrong tool?

Be honest about where it doesn't fit:

No transaction history

If you went live on ERP last month, there's nothing to learn from yet. Get six months of clean records first.

Safety-critical decisions

AI can flag; a qualified person decides on anything that touches safety, release of product or regulatory sign-off.

Machine control

An ERP-side AI agent shouldn't send commands to machines. Anthropic's own manufacturing material says nothing writes to production control on its own, and that's the right default.

A broken process

AI on top of a process people work around just speeds up the workaround.

Next step

Where FinByz fits

FinByz has implemented ERPNext since 2016, across 200+ projects in 15+ countries, with a 50+ person in-house team. For manufacturers we set up ERPNext manufacturing properly first (BOMs, routings, job cards, quality and stock), then add AI agents on ERPNext for the use cases above through our AI automation services.

If you want to know which of the 12 your data can support today, talk to us. We'll start with the data check.

  • Data check of Work Orders, job cards, stock and inspections
  • ERPNext manufacturing set up properly first
  • AI agents with ERP permissions and approvals
  • One measured use case at a time
Talk to our Experts

Ready to Scale with Smart Tech? Let's Talk.

Transform your business processes with AI-powered automation and expert implementation.

Trusted by 100+ businesses
Questions buyers ask

Frequently Asked Questions

Find answers to common questions about our services