Whoever is nearest answers the phone, and that's rarely their job
Call a typical plant and the person who answers is rarely a receptionist, because most plants this size don't have one. It's a planner between two meetings, a machinist who happened to walk past the office phone, or the one person in the admin team who also does the books, the payroll, and half of purchasing. Whoever picks up stops what they were doing, deals with the call as best they can, and goes back to the job that was actually theirs — except now it's ten or fifteen minutes later and the thing they were concentrating on has to be picked back up from wherever they left it.
None of the calls themselves are hard. Confirming a delivery date, taking down an RFQ, telling a haulier which dock to use — a competent person handles any one of them in under a minute. The cost isn't the call. It's the interruption, repeated several times a day, always landing on someone whose actual job is estimating, planning, or running a line, not answering phones. A voice AI system built around this kind of call absorbs the interruption instead of routing it to whoever is unlucky enough to be near the desk.
The calls that keep pulling someone off what they were doing
The shape of these calls is consistent across plants that make very different things.
- Customers chasing order status and delivery dates. "Where's my order" and "will it still ship Thursday" are the single most common reason a plant's phone rings, and they are also the calls least suited to interrupting a planner who has a schedule to hold together.
- RFQ enquiries that arrive by phone, not email. Plenty of buyers still call before they send anything in writing — to ask whether a plant can even make the part, or to get a rough sense of lead time before committing a request for quote to paper. Losing that call to voicemail often means losing the RFQ to whichever competitor answered.
- Supplier and haulier calls about deliveries. A supplier confirming a raw-material drop, a haulier asking which dock to reverse into, a driver running late and needing a new window — these calls need an answer now, not a callback once someone's free.
- Main reception for a site with a very small admin team. A site running real production volume can still have one or two people covering the entire front office — purchasing, HR queries, visitors, and the phone, all at once. The phone is usually the thing that loses.
- Out-of-hours calls when the plant runs shifts but the office doesn't. Production can run nights and weekends while the office runs nine-to-five, which means every call that comes in outside those hours — a night-shift supplier issue, a customer in a different time zone — goes to voicemail or nowhere at all.
What the system does on a normal day
The system sits on the number your plant already publishes and answers every call, not only the ones the admin team can't get to in time.
For an order status call, it identifies the order — by order number, customer name, or PO reference — checks the live status in your ERP, and gives the caller a real answer: what's shipped, what's still on the floor, and the current delivery date, not a guess based on when the order was placed. For an RFQ call, it captures the part, the quantity, the material, and the requested lead time, checks it against what your team already quotes for, and either logs it as a structured request for an estimator to price or tells the caller honestly that someone will call back with a number, rather than inventing one on the spot. For a supplier or haulier call, it confirms the delivery window, the dock, and the paperwork required, and updates the record so the person expecting that delivery isn't finding out about a change secondhand.
Anything the system can't resolve on its own — a complaint, an unusual request, a caller who simply wants a person — gets handed off immediately, with a summary of the call already prepared for whoever picks it up. The system doesn't try to be the last word on anything that needs judgment; it exists to close the gap between a call coming in and a person being free to take it.
Three calls, start to finish
An order status call ten minutes before the delivery window closes
A customer calls at 9:40 asking whether an order due today is still on track, because their own dock closes at 10 and they need to know whether to hold a crew back. The system looks up the order against the PO number the caller gives it, confirms it shipped the previous afternoon, and reads back the carrier and the expected arrival window. The caller thanks the system and hangs up having got a real answer in under a minute, and the planner who would normally have fielded that call never knew it happened.
An RFQ that comes in by phone before the paperwork does
A buyer calls to ask whether the plant can quote a part before he sends anything in writing — he's calling three shops in the same afternoon and wants to know quickly whether it's worth his time to send the drawing. The system asks for the part type, material, approximate quantity, and target lead time, checks that against the kind of work the plant already does, and confirms honestly that it's within scope and that an estimator will follow up once the drawing comes through. The call is logged with everything the buyer said, so when the RFQ does land by email, the estimator already knows it's coming and roughly what it involves — a workflow that runs alongside automating the RFQs and order paperwork that arrive in writing rather than duplicating it.
A haulier calling after the office has gone home
It's 6:50 p.m. and the office closed at 5, but the plant is still running a second shift. A haulier calls to say he's running forty minutes behind on a delivery expected before close and wants to know if someone will still be there to receive it. The system checks the delivery record, confirms the plant runs a night shift and that receiving stays staffed, and gives the haulier the correct dock and a contact on the floor to ask for when he arrives. No call gets lost to a closed office, and nobody on shift is surprised by a truck showing up with no warning.
Connecting to your ERP and phone system
A voice system for a plant's front line only earns its place if it plugs into what you already run, not a new phone platform your team has to learn.
Your existing number. Most builds sit on the number you already publish, through call forwarding, a SIP trunk into your current provider, or a direct integration with the phone system you use today. Nobody has to learn a new number, and nothing about how customers, suppliers, or hauliers currently reach you needs to change.
ERP and order data. Order status calls only work if the system is reading the same live data your team sees, not a stale export from last week. We connect to your ERP through an API where one exists, and through a scheduled feed or a shared database where it doesn't — the same mix of modern and older systems covered in AI in manufacturing operations generally.
Routing rules. What the system answers directly, what it escalates, and who it escalates to are all decisions your team makes, not defaults we impose. A supplier call about a late delivery might route differently than a customer complaint, and those rules get set during scoping, not guessed at afterward.
Where the call touches a document. Some calls end with someone needing to check a spec, a drawing, or a certificate the caller mentioned — that retrieval sits with a knowledge system built for manufacturing rather than the phone system itself, and the two connect cleanly where a plant runs both.
A modest number of calls, not a call centre
It's worth being honest about the shape of this problem, because it isn't the one most voice AI marketing describes. A plant isn't fielding thousands of calls a day the way a consumer support line does. It's fielding a modest, fairly predictable number — order status, an RFQ here and there, a handful of supplier and haulier calls, the occasional evening call nobody was there to take. What makes that expensive isn't volume. It's that every one of those calls currently pulls a skilled person — an estimator, a planner, a quality lead — away from the work they were actually doing, several times a day, every day.
That's also why this tends to be a smaller, cheaper build than a high-volume consumer line: fewer call types, a narrower scope, and a shorter path to something live and useful. The voice AI cost guide breaks down what typically drives the number for a build like this, and the after-hours coverage use case goes deeper into the specific gap between a plant that runs shifts and an office that doesn't.