The manufacturing order is late and the reason was visible three weeks ago

The gap between the floor and the office

AI automation for manufacturing is about moving what the floor already knows to the person who can act on it. Order status visible without a phone call, material shortages flagged before the line stops, downtime and quality captured with reason codes that make the record useful later, and dispatch paperwork assembled once. Capture has to work on a shared phone, not a laptop.

A factory usually knows what is happening. The line supervisor knows the machine has been stopping. The store keeper knows the accessories have not arrived. The quality checker knows a particular operation is producing rework. Each of those facts exists inside somebody's day and none of them reliably reaches the person who could still change the outcome.

So the merchandiser tells a buyer the order is on schedule, because that was true when the plan was made, and the shortfall becomes visible when there are eleven days left and nothing can be done except pay for air freight or ask for an extension. The information was available in week two. It was simply not written anywhere that could be read from an office.

This is what automation is genuinely good at in a manufacturing setting, and it is not what factories are usually sold. The pitch tends to be prediction and optimisation. The actual return is much duller: capturing what people already know, at the moment they know it, in a form that adds up.

Capture has to survive the shop floor

Every failed factory system fails the same way. Somebody designs a screen, it requires a laptop in an office, the supervisor is on the floor for eleven hours, and the data gets entered at the end of the week from memory or not at all. The system then reports a version of the factory that never existed, and management makes decisions from it, which is worse than having no system.

The design constraint is absolute. Capture happens where the work happens, on a phone, in a few seconds, using a photograph, a voice note or a code scanned at the machine. A defect is recorded by selecting from a short list, not by typing a description. A stoppage is recorded by choosing a reason and a duration. If it takes longer than the walk to the next machine, it will not happen.

Language matters here too. Supervisors and operators do not work in English, and a system that demands it will be filled in by one literate person for everybody else, which reintroduces the memory problem it was bought to remove.

Materials, and the shortage discovered at the line

The most expensive scheduling failure in this market is not a machine breaking. It is accepting an order and discovering, once production has started, that a component or a material is short. The line stops, the plan for every other order shifts, and the delivery date that was promised in good faith becomes impossible.

The check that prevents it is arithmetic, not intelligence. When an order is accepted, compare what it requires against what is physically in the store and what is on open purchase orders with realistic lead times, and raise the shortfall then. Doing that consistently, for every order, at the moment of acceptance, is precisely the sort of repetitive checking that people skip when the sales side is pushing to confirm.

The prerequisite is a bill of materials that is actually correct and a store record that matches the store. Factories where the register and the shelf disagree will get confident shortage reports that are wrong in both directions. Counting the store honestly is the first project in that situation, and it is not a technology project.

Quality, downtime and the reason codes that make them useful

Quality records in most factories exist as paper that satisfies an inspection and is then filed. They are rarely analysed, because analysing handwriting across a hundred sheets is nobody's job. The same information captured at the point of inspection, against fixed defect codes, with the line, the shift, the operation and a photograph attached, becomes something that answers questions: which operation, which shift, which machine, and whether last month's corrective action changed anything.

Machine downtime has the identical shape and the identical failure. Recorded as breakdown, which explains nothing. Recorded against a short list of causes with a duration, it produces the two or three recurring causes that account for most lost hours, and those are usually addressable without capital expenditure.

Buyer facing quality documentation benefits from the same capture. Inspection records, corrective actions and traceability that can be produced on request rather than assembled the night before an audit, which is when errors get introduced.

There is an honest caveat. Measuring downtime and doing nothing about it is worse than not measuring, because it costs supervisor time and teaches everybody that recording data is theatre. Agree who reviews the top causes and how often before the first record is captured.

A downtime register that nobody reviews is a tax on the supervisor's day, collected weekly and spent on nothing.

Dispatch, export paperwork and the buyer who wants an update

The same information gets typed several times on the way out of a factory. Packing list, delivery challan, gate pass, commercial invoice and, for export orders, the further set of documents where a discrepancy between two of them holds a shipment at a stage where holding is expensive. Every retyping is an opportunity for a mismatch in a quantity, a carton count or a description.

Assembling the document set from one order record, and checking the documents against each other for consistency before a person signs them, removes a category of delay that has nothing to do with how well the factory produces. The filing and the submissions themselves are done by the people and agents who are accountable for them. The system prepares and it checks.

Buyer communication closes the loop. A weekly production status against the delivery date, generated from the order record and approved by the merchandiser before it goes, changes the relationship from one where the buyer chases to one where the buyer is informed. It also forces the internal conversation about a slipping order to happen in week two rather than week nine, which is the actual point.

The order to build in is usually this: order status visibility, material shortage checking at acceptance, downtime and quality capture, dispatch document assembly, then buyer reporting on top.

• One order record that every department reads, so nobody walks to the floor to answer a buyer
• Shortage checking at order acceptance rather than at the line
• Fixed reason codes for defects and stoppages, captured on a phone
• Document sets assembled from one source and cross checked before signature
• A weekly buyer update generated from the record and approved by a person

Factories with overseas buyers, and what those buyers attach to the order

The gap between the floor and the office is the same in every country. What changes when the buyer is abroad is how much of that gap the buyer expects to see.

An overseas buyer usually asks for records the factory already half keeps: traceability from a finished lot back to the material and the supplier, quality results held per batch rather than per shipment, and evidence from social or environmental audits. Certification to a recognised quality management standard is commonly a condition of supply rather than a line on a company profile. If those records live in a register on the floor, an audit becomes a fire drill and every buyer question takes a person off the line.

This is the strongest argument for capture at the point of work in an export factory. The records are being kept anyway. The only open question is whether they can be produced in an hour or in a week.

Time zone decides the update habit. A European buyer's morning is a Karachi afternoon and a Gulf buyer sits within an hour of Karachi, so a same day answer is realistic. A buyer on the United States east coast opens at about the time a Karachi day ends, so the update has to be published on a schedule the buyer can read without asking. Delivery calendars split the same way: Eid production peaks at home and in the Gulf, back to school and Christmas deliveries for Europe and North America.

When the constraint is the floor, not the office

A job shop running a handful of orders at a time, where the owner walks the floor daily and knows the state of every machine, does not need this. The reporting overhead would cost more attention than it returns.

If there is no accurate bill of materials and the store record does not match the shelves, every downstream calculation inherits that error and states it with more confidence than a person would. Counting honestly comes first.

If the real constraint is power or gas availability, better scheduling data will tell you in more detail what you already know. It has some value in negotiating delivery dates honestly with buyers, and it is not the productivity gain being imagined.

And if management is not going to review what gets captured, do not capture it. A supervisor who records stoppages for six months and never sees anything change will stop recording accurately, and the system will then produce numbers that are worse than no numbers because people believe them.

How soon the floor sees anything, and the two cases where this does not work

The first week is the audit and in a factory most of it goes on watching capture rather than reading a process map, because a system a supervisor will not use during a shift reports a version of the factory that never existed. One piece is live inside a fortnight, normally downtime or defect capture on a shared phone against a short fixed list, since it needs nothing integrated and starts producing the two or three recurring causes almost at once. Material shortage checks at order acceptance come later, because they depend on a bill of materials that is correct and a store record that matches the store.

There are two situations where this does not work, and both belong before a quotation rather than after one. Where the register and the shelf disagree, shortage reports come back confidently wrong in both directions, and counting the store honestly is the first project rather than a technology purchase. And where nobody has agreed who reviews the top downtime causes and how often, measuring them is a tax collected from the supervisor every week and spent on nothing. Automation cannot supply the decision at the end of the data, and a supplier who will not raise that is selling you a register.

Nothing about a quality release, a corrective action or whether an order can be promised is decided by the system. It assembles the evidence, checks a document set against itself for a mismatch in a quantity or a carton count, and then hands it to a person to sign, because the signature is the thing the buyer is relying on. Anything that changes a delivery date already given to a buyer waits for human approval, since that conversation is commercial rather than clerical. Ibrahim owns build and workflows at Wobble and tests capture by watching one supervisor record one event during an ordinary shift without help, which is slower than a demonstration and considerably more honest.

A factory with a methodical planner can build the shortage check in-house on a spreadsheet, and for a single line operation that is frequently the right answer, because the arithmetic is the value and not the software. What is harder to keep on your own team is document set assembly across export paperwork, where a discrepancy is paid for in days at a port. Whatever does get built sits in accounts under your own logins, so you own the system and the reason code history stays with the factory rather than with a supplier. Wobble works from Karachi and bills month to month, across 25 engagements in six countries.

Common questions

What can a Pakistani manufacturer realistically automate first?

Order status visibility, so anyone answering a buyer can see the stage an order is at without walking to the floor, and material shortage checking at the moment an order is accepted rather than when the line stops. Both are arithmetic against records that mostly already exist, and both address the failure where information the floor had never reached the person who could act on it.

Why do factory data systems usually fail?

Because capture is designed for an office and the work happens on the floor. If recording an event requires a laptop or takes longer than the walk to the next machine, it gets done at the end of the week from memory or not at all, and the system then reports a factory that never existed. Capture on a phone, in seconds, in the language people actually work in, is the whole design constraint.

How does automation help with quality records?

By capturing inspection results against fixed defect codes at the point of inspection, with line, shift, operation and a photograph attached, so the records aggregate instead of being filed. That answers which operation and which shift are producing rework and whether a corrective action changed anything, and it means buyer documentation can be produced on request rather than assembled the night before an audit.

Can AI track machine downtime?

It can make recording it fast enough that it actually happens: a short list of causes and a duration, selected on a phone at the machine. That surfaces the two or three recurring causes behind most lost hours, which are often addressable without new equipment. Agree who reviews those causes and how often before starting, because measuring and ignoring is worse than not measuring.

Can export documentation be automated?

The assembly and the consistency checking can be. One order record produces the document set, and the documents are checked against each other for mismatched quantities, carton counts and descriptions before a person signs. The filing and the submissions stay with the people and agents accountable for them. What is removed is the retyping that creates the discrepancies in the first place.

What does a factory need before starting?

A bill of materials that is correct, a store record that matches the shelves and one order identifier that every department uses. Without those, the calculations inherit the error and state it more confidently than a person would. Counting the store honestly is often the real first project, and it is not something a supplier can do for you.