When Kanban Becomes a Second Planning System

A pull system connects downstream consumption to an upstream replenishment action. When material is consumed, a signal communicates what is required, how much is authorised and where it must go. The supplying process or material handler responds to that signal rather than producing or moving material because a separate schedule says it might be needed.

The card is only one possible signalling device. An empty container, marked space, electronic message or other clear trigger can perform the same function. What matters is the connection between consumption, the signal, an authorised quantity, ownership of the response, replenishment and the way abnormalities are handled.

Lean Enterprise Institute describes pull production as downstream activity signalling its needs to upstream activity. In a supermarket pull system, withdrawal sends information upstream and authorises replacement of what was consumed. That definition is useful because it makes the control mechanism explicit: actual use creates the authority to replenish.

If I was looking at this on the shop floor, I would not start by counting cards or checking whether every rack has a label. I would watch how the next production decision is actually made. Does an empty container trigger a defined response, or does someone wait for the planner’s list? Does the supplying process replenish a withdrawal, or does a supervisor change the sequence through a message or conversation?

Two planning systems can run at once. The pull loop creates one set of instructions while the spreadsheet, production schedule or supervisor creates another. When they agree, the weakness remains hidden. When they conflict, people have to decide which instruction is real. The system they trust enough to act on is the system controlling production.

None of this means the people involved are resisting improvement. It means the formal pull loop is not reliably answering the operational question: what should be replenished next? A workaround has answered it instead.

The first thing I’d want to know is why people stopped trusting the pull loop. If a material handler misses collections, a card may sit unnoticed while the consuming process runs out. If quantities were calculated for an old demand pattern, the signal may consistently arrive too late. If no one owns an empty supermarket location, the shortage can remain visible without producing a response.

Production mix also matters. A loop designed around stable, repeated consumption can become unreliable when product variety, batch size, supplier timing or changeover behaviour changes. Scheduling may then override the signal to protect a customer order. One exception becomes a routine, and soon the spreadsheet carries the practical knowledge that the pull system was supposed to contain.

The interesting bit is what happens when something goes wrong. If the normal loop has no defined abnormal response, people create one. Supervisors expedite, planners issue new lists, material handlers take verbal priorities and operators build protective stock. Those actions may be keeping production alive. Removing them without repairing the failed connection would make the operation worse, not leaner.

An operator keeps two extra boxes beside the machine ‘just in case’. The easy response is to remove them because they exceed the defined line-side quantity. I wouldn’t start by removing the buffer. I’d want to know what weakness in the system the operator is compensating for.

Perhaps the milk run sometimes misses the collection point. Perhaps supplier delivery varies by several hours, consumption rises sharply on one product variant, or the Kanban quantity was based on an unrealistic replenishment lead time. The operator may have learned that following the formal quantity creates a predictable shortage, while two extra boxes keep the line running.

The additional stock still has a cost. It hides shortages, consumes space, weakens inventory accuracy and can conceal quality or revision problems. But it is also evidence. Record when the buffer is used and what would have happened without it. Then correct the replenishment weakness so the local protection becomes unnecessary and can reduce naturally.

Instead of auditing the Kanban board, stand beside the consuming process and follow one container. Watch the last part being taken. What physical or digital event happens next? If the answer is ‘someone normally notices’, the trigger is already weaker than it appears on the procedure.

Follow the empty container or signal to the next person. Note when it becomes visible, what quantity it authorises and whether that person has both the responsibility and capacity to respond. Continue until replenished material returns to the correct location. Use actual times. Do not replace observation with the planned milk-run interval or the lead time written in the standard.

Then ask what happens if the expected response cannot occur. If stock is unavailable, does the empty location or card expose the shortage to the right owner? Is there a defined escalation, or does the material handler put the card aside and start making calls? Does someone create an urgent order outside the loop? This is where the second planning system usually becomes visible.

Lean Enterprise Institute’s description of withdrawal and production Kanban shows how the signals work together: withdrawal authorises movement to the downstream process, while the corresponding production signal authorises replacement upstream. It also notes that cards must return in a timely and reliable way for the control information to remain accurate. That reliability is something to verify on the floor, not assume from the design.

A useful audit does not need dozens of checks. These five questions test whether the essential links exist and whether they control real decisions. Ask them at the point of use, then confirm the answers by observing a complete replenishment cycle.

A pull system should not depend on perfect conditions. Late suppliers, quality holds, demand spikes, missing containers, missed milk runs, machine downtime and production-sequence changes will occur. The system needs to make those conditions visible early enough for a controlled response.

The same applies after downtime. A supplying process may try to recover by producing a large batch, while downstream cards authorise a different sequence. If the recovery rule is unclear, the schedule or the loudest shortage takes over. The cards remain visible, but they no longer govern the decision.

The practical test is simple: does the system expose the abnormality, or do people have to quietly compensate for it? A healthy loop does not prevent every disruption. It reveals the disruption, identifies who owns the response and preserves one clear method for authorising production and movement.

When supervisors repeatedly check shortages, chase material, reorder priorities and tell operators what to run, expediting has become part of production control. It can look like an individual working style, but it is often a rational response to unreliable information and unclear ownership.

A supervisor is close enough to see which order is about to stop, which supplier is late and which process can change sequence. Their judgement bridges gaps that the formal loop does not handle. The problem is that this knowledge remains in conversations and urgent decisions. Other people cannot see the same priority, and the system depends on the supervisor being present.

Do not begin by telling the supervisor to stop expediting. Capture the conditions that trigger intervention. Which signals were late, missing or wrong? Which exception had no owner? Which schedule instruction conflicted with actual consumption? Repair those links and confirm that the normal loop can make the decision before removing the workaround.

Start with one part family or one replenishment route where the problem is visible. Clarify the consumption event that creates the signal and define the exact quantity it authorises. Confirm the collection frequency, replenishment lead time and stock required to cover real variation. Use observed performance rather than an assumed route time.

Give each step an owner. Someone must collect the signal, someone must supply or produce the authorised quantity, and someone must respond when that cannot happen. Make shortages and overdue signals visible where action is taken. Define how quality holds, schedule changes and supplier failures enter the same control logic without creating an independent priority list.

Then remove competing instructions carefully. A spreadsheet may contain necessary sequencing logic that the pull design never addressed. A supervisor’s shortage list may be the only current abnormality system. Transfer those functions into the repaired loop before retiring the workaround.

Only after the basic process is stable should digital signalling or automation be considered. Technology can shorten transmission time and improve visibility, but it cannot resolve ambiguous triggers, wrong quantities or absent ownership. Automating two competing planning systems only makes the conflict faster.

A working pull system is usually less dramatic than the system it replaces. Consumption creates a clear signal. The signal authorises a known quantity. The responsible person responds within a dependable period, and an exception becomes visible before the consuming process runs out.

Operators no longer need unofficial stock because the normal quantity and response protect production. Supervisors spend less time checking, chasing and rewriting priorities. Planners manage the system boundary, demand and genuine exceptions rather than issuing a second stream of shop-floor instructions.

Most importantly, the pull signal actually controls material movement. Cards, containers and supermarket markings support that control, but their appearance is not the result. The result is one trusted connection from consumption to replenishment, with a visible response when the connection fails.