In a manufacturing business the ERP holds the orders, stock and costs, but much of what feeds it arrives by email. Customer purchase orders come as PDFs, quote requests arrive with drawings attached, supplier confirmations turn up as replies, and licence cards are sent in as photos. Customer service, purchasing, stores and accounts read each one and key it in. Automation takes the predictable steps, the lookups, matching, routing and chasing, and runs them as fixed, tested software that behaves the same way every time. AI is used only where something has to be read or judged, such as a free-form order or a supplier’s reply, and its output is checked before anything leaves the system.
These are patterns we see in manufacturing businesses, not client case studies. Your process gets its own map in the assessment.
Customer purchase orders keyed into the ERP
A customer’s purchase order might arrive as a PDF from their procurement system, a spreadsheet, or a line in an email that says “same as last time, but double the brackets”. Customer service reads it, finds the account, works out which of your part numbers the customer’s description means, checks the price against the contract, and keys each line into the ERP. AI reads the order into lines and maps the customer’s descriptions and part numbers to your SKUs. From there the work follows set rules, so every line is checked against the customer, product and price records before the order is written and the confirmation goes out. Lines that don’t match cleanly wait for customer service before anything posts.
Quoting from drawings and RFQs
Requests for quote come in with drawings, a material spec and several quantity breaks, sometimes followed by a phone call that changes the quantities. The estimator reads the drawing, works out material, machine time, finishing and freight, then builds the quote in a spreadsheet they maintain themselves. The estimator stays in charge and approves every quote before it goes, because a wrong price on a production run can lose money for as long as the job runs. The groundwork is where the automation sits. AI reads the request and its attachments and drafts the quote lines, and your price book, routing costs and margin rules are applied the same way on every quote, with the template filled and the follow-up scheduled.
Buying from suppliers
Purchasing starts most days with the shortages report. Someone raises purchase orders for steel, fasteners or packaging, emails them out, then spends the week chasing confirmations and ship dates so production planning knows what it can schedule. Replies come back as email text, marked-up PDFs or a phone call. Here the purchase orders are generated from your reorder rules and approval limits, and a chase schedule follows up any that haven’t been confirmed. AI reads supplier replies for confirmed dates, part substitutions and quantity changes, and those are written back to the ERP. A date that slips past the build date, or a substituted material, goes to the buyer to decide.
Imported raw materials and components come with a commercial invoice, packing list, bill of lading and sometimes certificates of origin or treatment, each in the supplier’s own layout. Someone checks them against the purchase order and sends them to the customs broker, and a missing certificate holds the container at the wharf. Here AI reads the supplier documents into fields, and set rules check them against the PO for quantities, prices and part numbers, confirms the required certificates are present, and sends the set to your broker with the shipment reference. Differences, or a certificate that hasn’t arrived, go to the purchasing officer before the vessel does.
Supplier rebate and volume incentive agreements sit in contracts and email threads, each with its own tiers, periods and claim deadlines. Tracking them means pulling purchase history, working out which tier you reached, and remembering to lodge the claim. A rebate earned and never claimed stays with the supplier. AI reads each agreement’s terms into rules once, and accounts checks every rule against the contract before it’s used. From then on, purchases are tracked against each agreement, the entitlement is calculated and claims are lodged on schedule. Accounts can see what’s owed, what’s been claimed and what’s been paid.
Paying suppliers and closing the month
Supplier bills in manufacturing rarely match the order neatly. The steel price moved, freight was added, a part-shipment was invoiced in full, or the store hasn’t entered the goods receipt yet. Capture tools already read standard invoices well, and we use them where they fit. The work left is the exceptions. Matching each invoice against the purchase order and the goods receipt is rule work, and so are applying your price and quantity tolerances, routing approvals to the right cost-centre manager, chasing the ones sitting unapproved and reconciling supplier statements. AI steps in only for invoices the capture tool can’t read. Anything outside tolerance goes to accounts payable with the difference shown.
Many manufacturers run more than one entity, perhaps an operating company, a property trust and a sales arm in another state. Month-end means bank reconciliations, intercompany charges, stock and WIP accruals, and variance checks against standard cost, mostly in spreadsheets the financial controller built years ago. Enterprise close software is sized for much larger companies. Nearly all of this runs as fixed software: pulling balances, matching intercompany entries, posting recurring accruals and flagging variances over your thresholds. AI drafts the variance commentary, such as why material usage ran over or labour recovery fell short, and the financial controller edits it before signing off the close.
Buying another manufacturer usually means buying its way of working too, with a different ERP, different part numbering, its own price lists and its own approval habits. For months after settlement, orders, invoices and stock are handled twice, and month-end needs a manual bridge between both sets of books. Tested rules handle the overlap. Customer, supplier and part records are mapped between the systems, transactions are synchronised during the transition, and a daily reconciliation checks both sides and puts differences in an exception queue. Your team decides which processes to keep, and the migration runs against checks instead of memory.
Customers, tenders and reporting
Winning work from a large customer or a government project often means a tender response or a prequalification portal. They ask for capability statements, quality system details, ISO certificates, insurances, safety records, product data sheets, and a compliance matrix against their specification. The same material gets rewritten each time by whoever has the least time. Automation assembles the pack, keeping the content library and certificate register current, building the compliance matrix and putting the documents into the customer’s format. AI drafts responses from your approved content only. Your team edits the drafts, decides which work is worth bidding for, and owns every submission.
A customer rings about a failed pump, a cracked casting or a batch that failed their incoming inspection. Someone raises a return authorisation, arranges the freight and logs the fault. If a bought-in component failed, they also lodge a warranty claim with that manufacturer, on its form, under its evidence rules. Claims get missed when the paperwork lags behind the credit. AI reads the fault description from the customer’s email and the service report. The rest follows your rules, with the return authorisation issued, the goods tracked back in and the manufacturer’s claim form filled from the return record. Quality decides the root cause and whether a credit is due.
Large customers now send suppliers questionnaires about emissions, modern slavery, safety and quality, often through a supplier portal and often asking the same questions in different words. Their own climate reporting is why so many of these ask for supply chain emissions data. The answers usually sit with the owner or the quality manager, and each questionnaire takes days from someone with other work to do. Automation keeps an approved answer library and an evidence register under version control, so answers stay consistent between customers. AI drafts each response from that library. The owner or quality manager approves every submission, and any new question goes to whoever owns that answer.
Mandatory climate reporting reaches a wider group of larger companies from financial years starting 1 July 2027. If your business meets the thresholds for that group, someone has to gather activity data every year from electricity and gas bills for each site, fuel cards, refrigerant top-ups, freight and waste invoices. In a manufacturer those sit across several sites and suppliers. Collection runs automatically, taking activity data from bills, meter data and your accounting system as it arrives, keeping the source record against each entry, and flagging gaps while there is still time to fill them. Your finance team and your auditor keep ownership of the report and the method.
Stores, plant and floor safety
Cycle counts and stocktakes regularly show the ERP and the rack disagreeing. Finding out why means someone tracing issues to work orders, unrecorded scrap, backflushing that consumed the wrong quantity, and receipts entered against the wrong location. It is slow, and it usually happens at the end of the month. The tracing needs no AI. Counts are compared with system quantities as they are entered, each variance is checked against recent transactions for the known causes, and a likely explanation is attached. The store supervisor reviews the variances, approves the adjustments, and can see which causes keep repeating.
On a factory floor, the record of who can operate what lives in several places. High-risk work licences for forklifts and cranes, machine-specific inductions, first-aid certificates and contractor site inductions each have their own file. When a licence lapses, the first person to notice is often an auditor, or an investigator after an incident. Automation keeps a single register per worker. AI reads uploaded licence cards and certificates for the class and expiry date, and the supervisor confirms any it can’t read clearly. Fixed rules send reminders before expiry, notify the supervisor, and flag the worker in the roster so they aren’t allocated to that machine until the ticket is current.
Near misses get written on a form in the lunchroom, incidents get emailed to the WHS coordinator, and hazard reports sit in a supervisor’s notebook. Corrective actions are tracked in a spreadsheet, when they’re tracked, and psychosocial hazards such as fatigue from shift patterns rarely get recorded at all. Automation gives every report the same path. A form on a phone or kiosk captures the report, fixed rules route it to the right supervisor, set follow-up dates and escalate overdue actions, and the register updates itself. AI summarises long free-text reports for the weekly safety meeting. The investigation, and the decision about what changes on the floor, stay with your people.
Presses, compressors, forklifts, overhead cranes and the delivery ute each have their own service interval, inspection requirement or registration date. Most factories track them in the maintenance supervisor’s head, on a whiteboard, or in a spreadsheet that goes stale. Fixed software holds the asset register, reads hour meters where it can reach them, raises service work orders when an interval comes due, and reminds the office about registrations and crane inspection dates. Completed work is logged against the asset, so the history is there when an inspector or insurer asks. The maintenance supervisor still schedules the work around production and decides when a machine comes offline.
If one of these costs you real money every month, describe it in four answers.
Thinking about what a process actually costs you?
The assessment is small, fixed-price, and tells you the real number, whether or not you ever build.
Worth a 30-minute conversation.