Skip to content

Pull System: Definition, Kanban & How It Really Works

By Christian Fieg · Last updated: April 2026

What is pull control?

Pull control is a production principle in which every process step only starts work when a downstream step signals real consumption. Nothing is produced "just in case" based on a forecast. Nothing moves forward until the next station is ready to take it. The customer order at the end of the value stream is the only legitimate trigger for activity upstream, propagated station by station through physical signals, Kanban cards or digital equivalents.

The principle sounds simple and is brutal in its implications. A pull system exposes every weakness in the process: unreliable equipment, variable cycle times, poor quality, long changeovers. That is not a bug, it is the point. Toyota built the Toyota Production System on pull precisely because push systems hide those problems behind inventory buffers.

In the Lean literature, pull is one of the five Lean principles defined by Womack and Jones (Value, Value Stream, Flow, Pull, Perfection). In ISO 22400, the related KPIs are on-time delivery, inventory turns and work-in-process. On a factory floor, it is a Kanban card, a coloured bin, or a digital signal that says "produce one more, now".

Push vs. pull: the defining difference

The clearest way to understand pull is to compare it side-by-side with the push logic that still dominates most ERP-driven plants. The two operating models produce very different plant behaviour even when the equipment is identical.

Dimension Push system Pull system
Trigger Forecast, MRP run, schedule Downstream consumption signal
Inventory strategy Safety stock to cover forecast error Capped supermarket at each step
Response to a problem Keep producing, inventory hides it System stops, problem becomes visible
Typical lead time Long, driven by queue time Short, driven by cycle time
Behavioural effect Keep machines busy Produce only what is needed
Tolerance for disruption High, absorbed by WIP Low, forces root-cause fixes

The practical consequence is that pull works best in environments where demand is reasonably stable, product variants are controllable and quality is reliable. In genuinely chaotic demand or low-reliability processes, a naive pull rollout will either starve the line or collapse into a push system wearing a Kanban costume.

The building blocks of a real pull system

"We do pull" often means very different things in different plants. A rigorous pull system contains a specific set of mechanisms, each doing a distinct job. Missing any of them turns pull from a discipline into a slogan.

Element What it does Typical tool
Takt Converts demand rate into the beat of the line Takt-time calculation, visual Andon
Supermarket Capped buffer between processes that absorbs short-term variation Defined racks, min/max levels, SAP/MES inventory
Kanban signal Authorises the upstream step to produce or move material Physical card, e-Kanban, MES pull signal
Pacemaker process Single scheduling point that sets the rhythm of the entire value stream Usually last process before shipping
Heijunka Levels the product mix so pull signals arrive in a predictable pattern Heijunka box, sequencer
FIFO lane Sequenced buffer where supermarkets don't fit (low-volume variants) Marked lane with a capacity limit
Quick changeover Makes small-lot production economical SMED

Pull, JIT and Kanban: how the terms relate

The three terms are often used as synonyms. They are not.

Term What it is Scope
Pull control Principle: produce only on downstream demand The why
Just-in-Time (JIT) Delivery strategy: right part, right quantity, right moment The what
Kanban Mechanism: cards or signals that carry the pull instruction The how

You can have JIT without strict pull (scheduled sequencing can also deliver parts on time), and you can use Kanban boards for non-pull applications (software teams do it routinely). But if you want a true Toyota-style pull system, you need all three, reinforcing each other.

When pull works — and when it doesn't

Pull is not a universal operating model. After 25+ years of running and watching production systems across four continents, the pattern is clear: pull delivers spectacular results in the conditions it was designed for, and fails quietly in others.

Condition Pull works well Pull struggles
Demand pattern Repeating, levellable, medium-to-high volume One-off, highly seasonal, extreme variance
Product variety Finite variant family (tens to hundreds) Pure project business, thousands of rare SKUs
Process reliability Stable quality, predictable cycle time Frequent unplanned stoppages, high scrap
Changeover time Short enough to run small lots Multi-hour setups that force big lots
Supply chain Reliable, short lead times from suppliers Volatile sourcing, allocation markets

In practice, most mid-sized manufacturers are a mixed case: some value streams are perfect pull candidates, others belong on a classical MRP/push schedule. Recognising the difference is the job; applying pull as a blanket policy almost always backfires.

How an MES makes pull actually run

Paper Kanban was the original Toyota innovation, and for stable, collocated plants it still works. But as soon as you have multi-site production, digital ERP integration, or even moderate product complexity, paper cards become the bottleneck. This is where a cloud MES earns its place in a Lean stack.

Pull requirement Paper/manual approach With SYMESTIC MES
Consumption signal Empty bin, card on board Automatic e-Kanban from process data or scanner event
Supermarket levels Visual check once per shift Real-time min/max, alarm on breach
WIP visibility Manual counts, lost cards Live WIP per process, per shift, per plant
ERP sync Backflush at end of shift, always lagging Bidirectional ABAP IDoc to SAP at cycle level
Pull across sites Impossible at scale Cross-plant signals via the same cloud tenant

At Meleghy Automotive, press-line consumption triggers pull signals across six plants through SAP-integrated MES data. Carcoustics runs a similar model on 500+ injection-moulding and punching machines with IXON gateways and MQTT-based signalling into Azure. In both cases, the pull logic is classical Lean; the execution layer is cloud-native and digital.

FAQ

Is pull the same as Just-in-Time?
No. JIT describes the delivery outcome — right part, right quantity, right moment. Pull is the mechanism most commonly used to achieve JIT, where downstream consumption authorises upstream production. You can deliver JIT through tight sequencing without a formal pull loop, and you can run a pull loop that is not strictly JIT (for example, supermarket-based pull with safety stock). In Toyota's original design, the two are deliberately combined, which is why they are often confused.

Does pull control require Kanban cards?
No. Kanban is the best-known mechanism but not the only one. Other valid pull signals include CONWIP (constant work-in-process) limits, polca (paired-cell overlapping loops of cards with authorisation), reorder-point replenishment, two-bin systems and digital e-Kanban. The cards themselves are irrelevant; what matters is that no upstream work starts without an explicit downstream signal, and that work-in-process is capped.

Can we run pull if our demand is unpredictable?
Only with heavy adaptation. Classical pull assumes that demand can be levelled (Heijunka) into a repeating pattern. If 80 % of your demand is repeating and 20 % is one-off, run pull on the 80 % and keep the 20 % on a classical schedule. If the ratio is inverted, pull will force you to either carry large safety stocks (defeating the purpose) or starve the line. In that case, invest first in SMED and flexibility, not in Kanban boards.

How does pull control interact with ERP?
ERP still owns master data, demand planning, customer orders, BOM and costing. Pull operates at the shop-floor execution layer, below the ERP scheduling horizon. The clean architecture, per ISA-95, is: ERP sets the forecast and master schedule, the MES converts real consumption into pull signals, and the ERP receives backflush data from the MES. Where companies fail is when they try to run pull through ERP MRP runs directly — MRP is a push planning engine by design, and bolting pull onto it creates nervous signals and instability.

What is the first sign that a pull system is failing?
Rising WIP inside the capped loops. In a healthy pull system, WIP stays at or below the designed limits, with visible empty spaces in the supermarkets. When WIP silently climbs — extra cards appear "temporarily", bins overflow, informal buffers emerge near the pacemaker — pull has collapsed into push with extra paperwork. The second warning sign is cards being overridden by supervisors under shipment pressure. Both are symptoms of an underlying reliability or changeover problem, not of a broken pull concept.


Related: Kanban · Just-in-Time · Takt Time · Heijunka · Lean Production · SMED · Cycle Time

About the author
Christian Fieg
Christian Fieg
Head of Sales at SYMESTIC. Six Sigma Black Belt with 25+ years in automotive manufacturing. Ran JIT/JIS and pull-based interior lines for Johnson Controls across plants in China, Mexico, the US and Europe. Author of "OEE: One Number, Many Lies". · LinkedIn
Start working with SYMESTIC today to boost your productivity, efficiency, and quality!
Contact us
Symestic Ninja
Deutsch
English