Why Your Shipments Slip — and How to See the Bottleneck in Time
Shipments slip because the floor problem — a bottleneck station, piling WIP, a line down since morning — only surfaces as an ERP variance after the buffer protecting the ship date is already spent. The fix is an early floor signal, read directly from the physical floor rather than from machine sensors, while you can still add a shift or re-sequence.
Here is the part the machine-monitoring vendors cannot say: you get that signal without instrumenting a single machine. No PLC tap, no MES rollout, no operator keying data into a terminal. A camera watches the line the way a supervisor would, if a supervisor could stand at every station at once.
Key takeaways
- Slippage is not a surprise. It is the late discovery of a problem that started hours or days earlier on the floor.
- The costliest bottlenecks in a labour-intensive Indian plant are usually a person or a missing trolley — the exact stalls machine telemetry cannot see.
- The earliest signal of all is a growing queue of WIP, which a camera counts directly and which predicts lateness before cycle time even drops.
- Cameras feed the one actual your ERP cannot fake: physical reality, not self-reported logs.
Why do shipments slip without warning?
Picture a mid-size plant in Faridabad or Coimbatore with an export container booked for Friday. On Wednesday afternoon the planner reconciles the production board and finds the assembly cell two days behind. The director now has three bad options: air-freight at a punishing cost, ship short, or call the buyer for an extension.
The information had been in the building since Monday. An operator knew Line 2 was slow. A supervisor knew a machine had been down since the morning tea break. WIP was stacking up before the paint booth. None of it reached the person who could act until the manual board was tallied and the gap was undeniable.
That is the core failure of on-time-in-full (OTIF) as most factories run it. OTIF is scored after the shipment (OTIF, Wikipedia). It tells you how often you were late. It does not tell you, on Monday at 11am, that Friday is now at risk.
Why is slippage a floor problem before it is a schedule problem?
Delivery dates do not slip in the ERP. They slip at a workstation. The classic mechanism is a bottleneck: the one resource whose pace sets the pace of the whole line. When the bottleneck loses time, the order loses time, and speeding up every other station recovers nothing. The idea comes from Eliyahu Goldratt's Theory of Constraints, laid out in his 1984 book The Goal (Theory of Constraints, Wikipedia).
Two things make this dangerous in practice:
- The bottleneck migrates. Fix the paint booth and the constraint jumps to packing. Yesterday's plan assumed a bottleneck that is not the bottleneck today. Static planning sees the migration only a shift later, in the result.
- The loss is silent. A machine running slow, a changeover that runs long, a cell idle while an operator hunts for a trolley — none of these trip an alarm. They quietly eat the buffer that was supposed to protect Friday.
The bottleneck your machine sensors cannot see
Every machine-monitoring pitch assumes the bottleneck is a machine. Bolt a sensor to the spindle, read the PLC, and you will know when that machine stalls. On a labour-intensive mid-size Indian floor, that assumption quietly fails.
More often the bottleneck is not a machine at all. It is a person — an absentee on a shift, or one skilled setter stretched across three cells. It is a material-handling gap — a cell sitting idle because the trolley it needs is somewhere else. It is a manual bench with no controller to read. A spindle sensor is structurally blind to all of these. It sees its own machine running fine and reports green while the station in front of it starves.
A camera does not care whether a station has a controller. It sees an unmanned position, a missing trolley, a WIP stack that is not moving. Video is the only continuous sensor that catches the stalls between the machines — which, on these floors, is where most of the lost time lives.
What early signals show a ship date is at risk?
The point of floor intelligence is to convert each root cause into a signal you can see the same shift, not a variance you read about after the container is gone. The table below maps six common causes of slippage to the physical signal a camera reads and the same-shift action it buys you.
| Slippage cause | What the camera sees on the floor (no sensor needed) | What the director does with a same-shift alert |
|---|---|---|
| Bottleneck station falling behind | Output count at the constraint lagging its takt rate by mid-shift | Re-sequence orders, add a hand, or authorise overtime today |
| WIP piling before one machine | A growing queue of trolleys or pallets at one station | Rebalance the line; pull labour from a starved downstream station |
| Line down since morning, unflagged | No motion and no output at a station for an extended window | Escalate maintenance now, not on the evening board |
| Changeover overrun | Changeover running well past its standard time | Send a setter, or hold the next changeover until the buffer recovers |
| Manpower short on a shift | Stations unmanned versus plan | Redeploy from non-critical areas before the gap compounds |
| Slow running while the station "looks up" | Throughput below rate although the operator is active | Check tooling and material feed before it costs a full shift |
Notice the middle column. None of it comes from a controller. It is physical proxy — queue length, occupancy, motion, unit count — read at stations that have nothing to instrument.
The earliest signal of all: the WIP queue
Cycle time dropping tells you the bottleneck is already losing. There is an earlier tell. Little's Law — the relationship between how much work sits in a queue, how fast it arrives, and how long it takes to clear — means a queue that starts growing predicts lateness before the station's cycle time visibly falls (John Little, 1961). A lengthening line of pallets in front of a machine is the leading indicator.
And a queue is a visual quantity. A camera counts pallets. You do not need a sensor on anything to know the queue at station 7 has doubled since 9am — you need something that can see and count, continuously. That is the one leading indicator the machine-telemetry crowd cannot read, because the pallets waiting between two machines belong to neither.
How does a camera know a line is "90 minutes behind plan"?
Fair question, and it is where most vision pitches wave their hands. A camera does not magically know your schedule. It counts. Establish a station's standard rate — its takt, the units it should complete per hour when healthy — then count actual completed units against that rate through the shift. The running shortfall is a deficit in units, which converts straight to minutes behind. Count fifty against a target of eighty by 11am and the line is thirty units, roughly ninety minutes, behind. Simple, and it needs no data from the machine — only a clear view and a baseline.
The anti-dashboard: one sentence, not ten gauges
The real-time-visibility industry sells the director more screens. A director has time for zero of them. Every dashboard is cognitive load he will not pay for.
Mama is the opposite. The output is one plain-language line a day, the kind of read a sharp shift-in-charge gives you when he catches your eye across the floor:
"Line 2 is 90 minutes behind plan as of 11am. The constraint is the CNC cell, WIP is stacking before it, two stations idle waiting. If it holds, Thursday's dispatch is at risk."
That is the second brain: continuous floor intelligence that surfaces the one thing worth your attention today, early enough to act. It turns the question from "why did we miss it?" — asked to your customer on Friday — into "what do we move now?" — asked to your supervisor on Monday. The same continuous-reading capability sits behind production line monitoring cameras in India, pointed at the delivery date instead of a single metric.
The one actual your ERP cannot fake
Every number in your ERP, MES, or whiteboard is self-reported. An operator or a supervisor logged it. On a manual floor that logging is late and often wrong: jobs marked done that are still on the bench, "ghost WIP" that exists on the board but not in the cell. You plan against a picture that people typed in, under pressure, at the end of a shift.
Video is the one source that is not self-reported. It does not ask the floor what happened; it watches what happened. Its job is to feed the actuals — is the line keeping pace, where is WIP stacking, is a station dark — back against the plan, fast enough to matter.
Be honest about the boundary. A camera sees physical reality: motion, queues, occupancy, pace. It does not see your order quantities, due dates, priorities, or material availability. That lives in your ERP, and it should. So an early brief is a reason to walk over and look, not a verdict — a camera cannot always tell a planned stop from an unplanned one without context you supply. And it needs a ramp-up: the system has to learn each line's normal rate over a shift or two before "behind plan" means anything.
Why this bites harder for Indian exporters
Two India-specific pressures raise the stakes.
Export windows and penalty clauses. Export orders often carry liquidated-damages (LD) clauses — a pre-agreed sum the buyer can claim for late or short delivery. Under Section 74 of the Indian Contract Act, 1872, courts award reasonable compensation not exceeding that stipulated amount, provided the sum is a genuine pre-estimate of loss rather than a penalty (India Code — Indian Contract Act, 1872, Section 74). A slipped container can mean a contractual deduction on top of the rebooking cost, plus a damaged buyer relationship in a market where re-orders are the whole game.
Manual production boards. A large share of mid-size Indian floors still track output on a whiteboard or a shift register tallied at day's end. That is a once-a-shift snapshot of a problem compounding hour by hour — the structural reason slippage is discovered late here specifically.
Neither is solved by working people harder. Both are eased by seeing sooner.
Where to start — reuse the cameras you already own
You do not need a full deployment, and you may not need new hardware at all. Most mid-size Indian plants already have security CCTV — an asset paid for years ago that produces zero operational value, watched only after a theft. Point that same feed at your one worst line, the one that most often puts a ship date at risk. There is nothing to instrument, no MES project, no capex committee.
Baseline that line's normal rate over a shift or two, define the standard rate per station, then watch for a sustained deficit against it and for WIP stacking upstream. In one live deployment we saw a station go dark at 11:08 and stay down for one hour fifty minutes — the exact silent stop a shift register would not surface until evening. One saved dispatch this quarter and the read has paid for itself.
If you want the underlying efficiency metric these signals feed, start with OEE explained for Indian factories — availability and performance losses are the ones that quietly slip your dates. For the rupee value of the lost time, see the cost of factory downtime and how to measure it.
This is the wedge Mama is built around. You record a short phone walkthrough of the floor. It returns a camera-placement plan for the stations where delivery risk accrues — or reads the feeds you already have — and turns them into a plain-language brief that flags "this date is at risk" while you can still save it. No machines to wire, no dashboard to watch.
FAQ
What causes shipment schedule slippage in a factory? A floor problem that surfaced too late — nine times out of ten. A bottleneck station falling behind, WIP piling before one machine, a line down since morning that no one flagged, or slow running that trips no alarm. By the time it shows as a variance on the board or ERP, the buffer protecting the ship date is spent.
Can cameras improve on-time delivery (OTIF)? Yes, but indirectly — by giving an early signal, not by measuring OTIF itself. Cameras read the floor continuously and can flag "Line 2 is behind plan at 11am" while you can still add labour or re-sequence. They do not plan orders; they complement planning by feeding fast, honest actuals against it.
Do camera analytics replace my ERP or planning system? No. Cameras see physical reality — motion, queues, pace, occupancy — not order quantities, due dates, or material status. Those live in your ERP. Floor intelligence catches schedule risk between reconciliations; it does not replace the plan. It is the one actual the plan cannot fake, because nobody typed it in.
Do I need to install sensors on my machines? No, and that is the point. A camera reads output rate, queues, occupancy, and downtime from a clear view of the line — including manual benches and material-handling stalls that have no controller to tap. Many plants can start on the security CCTV they already own.
Why do Indian exporters lose more when a shipment slips? Export contracts often carry liquidated-damages clauses. Under Section 74 of the Indian Contract Act, a late or short container can mean a contractual deduction — reasonable compensation up to the agreed cap — plus a damaged buyer relationship, on top of air-freight or rebooking costs. Many floors also still tally output on manual boards once a shift, so problems are found late.