Pull Systems, Production Levelling and Lean Supply Networks

Introduction

Lean production is often described as the elimination of waste, but its deeper purpose is to create a system in which value-adding work is triggered by actual downstream need. This is the logic of a pull system. In contrast, a push system releases work according to a forecast, schedule, or local target, often creating excess work-in-process (WIP), long queues, and hidden quality problems.

Pull does not mean that every operation waits passively for a customer order. It means that replenishment, production, and material movement are governed by explicit consumption signals and controlled inventory boundaries. Effective pull therefore requires stable processes, short and dependable replenishment times, visual controls, disciplined data, and coordinated suppliers.

This chapter examines Kanban, supermarkets, CONWIP, replenishment policies, heijunka, pitch, demand segmentation, supplier integration, logistics, bullwhip reduction, and the strategic choice between make-to-stock (MTS) and make-to-order (MTO). The central managerial question is not simply whether an organization uses Kanban, but whether its entire network converts demand into timely, economical, and reliable replenishment.

Learning objectives

By the end of this chapter, the reader should be able to:

1. Explain the principles and operating conditions of pull systems.
2. Design Kanban, supermarket, CONWIP, and hybrid replenishment systems.
3. Calculate container quantities, replenishment signals, takt time, pitch, reorder points, and safety stock under stated assumptions.
4. Apply heijunka and demand segmentation to reduce variability.
5. Evaluate supplier integration, milk runs, cross-docking, and information-sharing practices.
6. Diagnose bullwhip amplification and recommend countermeasures.
7. Select appropriate MTS, MTO, assemble-to-order, or hybrid policies.
8. Critically assess the operational, strategic, ethical, and technological limitations of pull systems.

1. The logic of pull

A pull system has four defining characteristics:

• Downstream authorization: a process produces or moves material because a consuming process has used material.
• Explicit WIP limits: inventory between processes is bounded by a defined number of containers, cards, slots, or jobs.
• Replenishment discipline: the signal is returned upstream and represents a real requirement.
• Abnormality visibility: shortages, quality defects, and process interruptions become visible rather than being concealed by large buffers.

A push system can be useful where long procurement lead times, batch economics, or regulatory release processes require advance planning. Lean management therefore does not imply that all planning disappears. Aggregate planning, sales and operations planning, capacity planning, and material requirements planning remain important. Pull normally operates within a broader planning hierarchy: forecasts and plans set the overall direction, while pull controls execution and replenishment at the appropriate time horizon.

A useful distinction is between demand information and execution authorization. A forecast may indicate that 10,000 units are expected next month. It should not automatically authorize every upstream operation to produce 10,000 units immediately. A pull signal authorizes a defined replenishment quantity after consumption has occurred.

Flow, WIP, and lead time

Little’s Law links average WIP, throughput, and flow time:

Here, L is average WIP in units, λ is the average throughput rate in units per unit of time, and W is average flow time in the same time unit. The relationship assumes a stable system over the period observed: average WIP and throughput are finite, and material entering the system eventually exits. It applies to manufacturing and service processes, although “units” may mean orders, patients, cases, or transactions.

For example, if a line completes 240 units per day and average WIP is 480 units, then average flow time is:

Reducing WIP while maintaining throughput reduces average flow time. However, reducing inventory without addressing interruptions can produce starvation and missed deliveries. Pull is therefore a control architecture, not merely an inventory-reduction exercise.

2. Kanban and supermarket design

A Kanban is a replenishment signal. It may be a physical card, electronic message, empty-bin trigger, barcode scan, RFID event, or digitally released order. The signal normally specifies the part, quantity, source, destination, and replenishment rules. A Kanban loop is not simply a scheduling board: it is a closed authorization cycle linking consumption to replenishment.

A supermarket is a controlled inventory point containing a defined range and quantity of items. A downstream process withdraws what it needs; the withdrawal signal triggers replenishment by the supplying process. Supermarkets decouple processes when continuous flow is impractical because of distance, product variety, differing cycle times, setup requirements, or transportation constraints.

A well-designed supermarket requires:

• clearly defined item locations;
• maximum and minimum quantities or card counts;
• FIFO discipline where age or quality matters;
• replenishment routes and frequencies;
• ownership of counting and discrepancy resolution;
• rules for shortages, defects, engineering changes, and obsolete stock;
• an escalation mechanism for abnormal demand.

The supermarket should be placed close enough to the customer process to provide rapid access, but not so close that it becomes an uncontrolled pile of material. It is a buffer with a purpose, not a substitute for process capability.

Kanban quantity calculation

A basic card or container requirement can be estimated as:

N is the required number of containers or Kanban cards, D is demand per unit of time, T is total replenishment time in the same time unit, α is the allowance for demand and process variability expressed as a decimal, and C is the quantity per container. Because a partial container cannot be used, N is rounded up to the next whole number.

Total replenishment time should include authorization delay, picking, handling, transport, queueing, production, inspection, and return of the signal. The formula assumes reasonably stable average demand, a defined container quantity, and that the allowance is deliberately chosen rather than used to conceal unreliable operations. It is not a substitute for statistical safety-stock analysis when demand and lead-time distributions are known.

Worked example. A component is consumed at 480 units per day. The complete replenishment loop takes 0.75 day, the planned variability allowance is 15%, and each container holds 60 units.

The system requires 7 containers. Nominal controlled inventory is therefore 420 units, although actual inventory may be lower as containers are consumed. If the loop is reduced from 0.75 to 0.50 day, the calculated requirement becomes 5 containers after rounding. Lead-time reduction can therefore remove inventory without reducing service protection.

Kanban operating rules

Common rules include:

1. The consuming process withdraws only the authorized quantity.
2. The supplying process produces only against a valid signal.
3. Defective material is not passed forward; the signal is escalated.
4. The number of signals is changed only through a controlled review.
5. Every signal has one location and one status.
6. Emergency signals are visible, time-limited, and investigated.

Kanban is most effective for repetitive demand, relatively stable routings, dependable quality, and short replenishment times. It becomes difficult when demand is highly erratic, products are unique, routing is unstable, or minimum supplier order quantities are much larger than the required replenishment quantity.

3. CONWIP and hybrid pull systems

CONWIP—constant work-in-process—controls the total number of jobs in a production loop rather than assigning a separate card quantity to every part number. When one job leaves the system, a release authorization becomes available for one new job. The total number of circulating authorizations remains approximately constant.

CONWIP is valuable in high-mix environments where individual Kanban loops would require too many cards and where jobs share common resources. It controls aggregate WIP while allowing the next job to be selected according to a dispatching rule, such as due date, product family, or customer priority.

Kanban and CONWIP answer different questions:

Hybrid designs are common. A plant may use CONWIP to control total WIP through fabrication, supermarkets for standard components, and final assembly triggered by customer orders. Such a design should define the boundaries and hand-offs clearly; otherwise, multiple controls can compete and create contradictory priorities.

4. Production levelling: heijunka and pitch

Heijunka means levelling the type and quantity of production over a repeated time interval. It does not eliminate demand variation; it converts irregular customer requirements into a smoother production sequence, subject to capacity and service constraints.

Suppose weekly demand is 400 units of A and 200 units of B, and the available production time is 30 hours. The takt time is:

For 600 units per week:

Takt time is a demand-based rhythm, not the actual processing time. A process can meet takt only if effective cycle time, including losses and required capacity, is compatible with that rhythm. Takt calculations must use the relevant available time: planned breaks, meetings, maintenance, and non-operating periods should be excluded if they are not available for production.

Pitch is the quantity or time interval associated with a repeated withdrawal or production cycle. If the finished-goods container holds 20 units and takt is 3 minutes per unit, the pitch is:

A one-hour pitch provides a practical cadence for material movement, production checks, and schedule management. Smaller containers reduce response time but increase handling and transaction frequency.

A heijunka pattern for 400 A and 200 B could use a six-unit repeating family ratio of A:A:B, adjusted to actual container sizes and setup constraints. The exact sequence must consider changeover time, product compatibility, quality risk, labour skills, and customer promise dates. Levelling is not blind averaging: it is a designed compromise between smoothness, responsiveness, and economic batch size.

Why levelling matters

Without levelling, a supplier may receive a weekly order in large bursts. The supplier then creates overtime, temporary labour, expedited transport, and excess inventory, even though the customer’s average demand is unchanged. Heijunka reduces this artificial variability. It also supports standard work, stable staffing, faster detection of abnormalities, and more predictable logistics.

However, levelling can be inappropriate when demand is genuinely seasonal, capacity is constrained, products are perishable, or urgent service has a high social value. In such cases, the organization should make the trade-off explicit rather than pretending that a fixed sequence can absorb unlimited variability.

5. Replenishment policies and demand segmentation

Pull design should reflect demand characteristics. A single policy for all items is usually wasteful. Useful segmentation dimensions include demand volume, variability, value, criticality, lead time, supply risk, and product life-cycle stage.

For a periodic-review order-up-to policy, the target stock level can be expressed as:

S is the order-up-to level in units, μ is average demand per unit time, L is replenishment lead time, R is the review interval, and SS is safety stock. The formula assumes that demand is represented by a stable average over the protection period L + R. The order quantity at a review is commonly:

Inventory position equals on-hand inventory plus scheduled receipts minus backorders. The policy must be adjusted for minimum order quantities, shelf life, capacity, and supplier constraints.

When demand during the protection period is approximately normal and lead time is fixed, a common safety-stock approximation is:

Here, z is the standard-normal service factor, σ is the standard deviation of demand per time unit, and L is lead time in the same time unit. This expression assumes independent demand periods, a stable distribution, fixed lead time, and a service measure based on the probability of avoiding a stockout during the protection period. If both demand and lead time vary, the expression is insufficient; a combined-variability model or empirical simulation is preferable.

For example, daily demand has standard deviation 12 units, lead time is 4 days, and the chosen service factor is z = 1.65. Then:

The result is approximately 40 units, not a guarantee that every demand cycle will be served. A cycle-service target is different from a fill-rate target, and managers must specify which service measure they are optimizing.

6. Make-to-stock, make-to-order, and postponement

The MTS–MTO choice is a positioning decision concerning where inventory is held relative to the customer order decoupling point.

• Make-to-stock: finished goods are produced before a specific customer order. This supports rapid delivery when demand is predictable but creates forecast and obsolescence risk.
• Make-to-order: production begins after a customer order. This reduces finished-goods risk and supports customization but increases customer lead time.
• Assemble-to-order: common modules are stocked and final configuration occurs after the order. This often balances variety, response time, and inventory risk.
• Engineer-to-order: design and production are initiated for a specific requirement, usually with long lead time and high customer involvement.

The decision should consider demand predictability, required response time, customization, component commonality, product value, obsolescence, capacity flexibility, and the cost of lost sales. The customer order decoupling point should be placed where postponement creates the greatest economic value.

A common error is to classify the entire product line as MTS or MTO. A product may be MTS at the component level, assemble-to-order at the module level, and MTO at final configuration. This layered approach often allows lean replenishment of common items while protecting customers from long waits for final customization.

7. Lean supply networks

A lean supply network extends pull principles beyond the factory. Its essential elements are synchronized demand signals, transparent capacity and inventory information, reliable logistics, collaborative problem solving, and shared improvement objectives.

Supplier integration

Supplier integration can include:

• common item definitions and engineering-change protocols;
• electronic schedules and consumption visibility;
• supplier capacity reviews and risk assessments;
• shared quality standards and root-cause analysis;
• supplier development and process audits;
• returnable packaging and agreed container quantities;
• long-term relationships with fair performance expectations.

Integration is not the same as transferring risk to suppliers. Requiring suppliers to absorb indefinite schedule changes, price pressure, and inventory ownership can create fragile networks and unethical commercial relationships. A robust partnership aligns total cost, quality, delivery, resilience, and improvement benefits.

Logistics patterns

A milk run uses a planned route to collect or deliver material among multiple suppliers or facilities. Compared with independent full truckloads, it can improve vehicle utilization and delivery frequency, although route feasibility depends on geography, time windows, loading constraints, and demand variability.

Cross-docking transfers incoming material to outbound movement with little or no storage. It is effective when inbound and outbound schedules are synchronized and product identity, quality, and timing are reliable. It is not simply “zero inventory”; poorly synchronized cross-docking can convert storage into queues at docks.

Supplier-owned or vendor-managed inventory can reduce transactions and improve availability, but governance is essential. Ownership, valuation, expiry, access rights, data accuracy, and responsibility for obsolete stock must be defined contractually.

Reducing the bullwhip effect

The bullwhip effect is the amplification of order variability as information moves upstream. Causes include forecast updating, order batching, temporary price promotions, rationing and shortage gaming, long lead times, and poor visibility of true consumption.

Countermeasures include:

1. sharing point-of-sale or actual consumption data;
2. reducing order and transport batch sizes;
3. using stable pricing rather than frequent promotions;
4. shortening and stabilizing lead times;
5. allocating scarce supply by transparent rules;
6. coordinating replenishment through collaborative planning;
7. measuring variability at each network echelon.

One analytical indicator is the variance amplification ratio:

A is dimensionless, Var(orders) is the variance of orders received by an upstream stage, and Var(demand) is the variance of downstream consumption over comparable periods. A greater than 1 indicates amplification. The ratio is descriptive, not causal; the observation period, aggregation level, promotions, and lead-time effects must be examined before conclusions are drawn.

8. Applied case: Northstar Medical Devices (illustrative)

Background

Northstar Medical Devices assembles a high-volume disposable diagnostic cartridge. Its final assembly line consumes a common plastic housing, a sensor module, and a sterile package. The factory uses a weekly forecast, large supplier deliveries, and several local production schedules. Finished-goods availability is acceptable, but WIP is high and engineering changes frequently create obsolete components.

Problem

The housing supplier receives orders in large weekly batches even though Northstar consumes approximately 1,200 housings per day. Supplier deliveries are frequently expedited near the end of the week. Northstar holds 18,000 housings, equivalent to 15 average consumption days, while the supplier experiences alternating overtime and idle periods. The assembly line also changes product variants in large campaigns, creating shortages of one variant and excesses of another.

Analysis

The team maps the replenishment loop and finds that the nominal supplier-to-line loop is 1.5 days, but actual variability is caused by two-day order batching, irregular transport, and inspection queues. A supermarket is designed at the point of use with 12 containers of 100 housings each. Average demand is 1,200 units per day, the loop target is 0.75 day, and the planned allowance is 20%:

The design rounds up to 11 containers, or 1,100 units. The twelfth container is retained temporarily as a controlled transition buffer and removed after delivery reliability is demonstrated. The supplier receives electronic withdrawal signals through a daily milk run rather than a weekly forecast-only order.

For production levelling, customer demand is translated into a repeating sequence of cartridge variants using a two-hour pitch. The sequence is reviewed weekly, but execution is controlled by consumption and available capacity. Common housings remain replenished through the supermarket; variant-specific sensor modules are assembled closer to the customer order point.

Outcome

After a stabilization period, average housing inventory falls, emergency freight declines, and supplier workload becomes smoother. The plant does not eliminate all safety stock: sterile packaging remains protected by a risk-based policy because sterilization capacity is constrained and regulatory release takes time. The case demonstrates that pull is selective. Different items require different decoupling points and controls.

Learning points

• Inventory reduction followed lead-time and variability reduction rather than occurring as an isolated target.
• The supplier received consumption signals, but aggregate capacity planning remained necessary.
• Heijunka addressed artificial order bursts; it did not deny genuine customer variation.
• Medical-device traceability and release controls limited the use of unconstrained supermarket replenishment.
• The transition buffer was governed by explicit exit criteria rather than becoming permanent hidden stock.

9. Implementation, measurement, and governance

Implementation should begin with a product family and a clearly bounded value stream. The organization should establish demand, process time, replenishment time, quality performance, container quantity, and escalation rules before calculating card counts. A pilot should test not only average inventory but also shortages, response time, schedule adherence, quality escapes, and workload stability.

Useful measures include:

Metrics must not encourage local optimization. A supplier’s high on-time percentage may be achieved by costly expediting, while a factory’s low inventory may be achieved by shifting shortages to customers. Governance should therefore review total-system performance, including safety, environmental impact, labour conditions, resilience, and customer outcomes.

Technology and AI

Electronic Kanban, manufacturing-execution systems, barcode scanning, RFID, and supplier portals can improve signal accuracy and traceability. Analytics can identify recurring shortages, estimate lead-time distributions, detect abnormal demand, and simulate card quantities. Machine-learning forecasts may support aggregate planning, but they do not replace pull discipline. A highly accurate forecast can still create excess inventory if every stage independently converts it into production.

AI-enabled recommendations require data governance. Managers should examine data ownership, cybersecurity, explainability, bias in supplier-risk scores, human override, and the consequences of automated release decisions. In regulated environments, electronic records and changes to replenishment logic may require validation, access control, audit trails, and documented approval.

10. Challenges and limitations

Pull systems have important limitations:

• They can fail when demand exceeds designed capacity for prolonged periods.
• Reducing buffers too rapidly can expose customers and employees to instability.
• Long, global, or disruption-prone supply chains may require strategic decoupling stock.
• Minimum order quantities and transportation economics can conflict with small-lot replenishment.
• Product proliferation can make part-specific Kanban administratively complex.
• A supermarket can hide poor quality if defective stock is replenished without root-cause action.
• Heijunka may reduce responsiveness if treated as an inflexible schedule.
• CONWIP controls total WIP but does not automatically select the best job sequence.
• Supplier integration can create dependency and concentration risk.

The appropriate response is not to abandon pull, but to define where pull applies, where planning is necessary, and where resilience buffers are economically and socially justified. Lean supply networks should be designed for both efficiency and recovery from disruption.

Chapter Summary

Pull systems authorize production and movement through downstream consumption rather than relying exclusively on forecasts or local targets. Kanban uses part-specific signals and is particularly suitable for repetitive, stable flows supported by supermarkets. CONWIP controls total WIP and is useful in high-mix environments. Their design depends on demand, replenishment time, container quantity, variability, quality, and capacity.

Heijunka smooths product mix and quantity, while takt time expresses the customer-demand rhythm and pitch translates that rhythm into a practical replenishment interval. Demand segmentation enables differentiated policies: stable common items may use Kanban, intermittent or critical items may require statistical or managerial review, and customized products may be controlled through MTO or assemble-to-order systems.

Lean supply networks extend these principles through supplier integration, milk runs, cross-docking, shared consumption data, and coordinated replenishment. Bullwhip reduction requires smaller batches, shorter lead times, stable commercial policies, and visibility of actual demand. MTS, MTO, and postponement decisions should locate inventory at the point that best balances response time, customization, obsolescence, capacity, and risk. Successful implementation combines quantitative design with governance, employee involvement, ethical supplier relationships, regulatory compliance, and continuous improvement.

Key Terms

Pull system: A control system in which replenishment is authorized by downstream consumption.

Push system: A system that releases work mainly from forecasts, plans, or upstream targets.

Kanban: A signal authorizing replenishment or production of a defined quantity.

Supermarket: A controlled inventory location used to decouple processes and trigger replenishment.

CONWIP: Constant work-in-process control using a fixed number of circulating production authorizations.

Heijunka: Production levelling by smoothing product mix and quantity over time.

Takt time: Available production time divided by required customer demand.

Pitch: The quantity or time interval associated with a repeated production or withdrawal cycle.

WIP: Work-in-process inventory between the start and completion of a process.

Little’s Law: The relationship L = λW linking WIP, throughput, and flow time.

Replenishment lead time: Time from a consumption signal until replenished material is available.

Safety stock: Inventory held to protect against specified demand or supply uncertainty.

Demand segmentation: Classification of items or customers to assign appropriate control policies.

Make-to-stock: Production completed before a specific customer order.

Make-to-order: Production initiated after receipt of a customer order.

Assemble-to-order: Stocking common modules and completing configuration after order receipt.

Milk run: A planned logistics route serving multiple collection or delivery points.

Bullwhip effect: Amplification of demand variability as information moves upstream.

Customer order decoupling point: The location separating forecast-driven activity from order-driven activity.

Review Questions

Short Answer Questions

1. What distinguishes a pull authorization from a demand forecast?
2. State two conditions under which a supermarket is useful.
3. Define takt time and explain why it is not the same as cycle time.
4. What is the principal difference between Kanban and CONWIP?
5. Identify two causes of the bullwhip effect.

Long Answer Questions

1. Explain how a Kanban system should be designed, implemented, and governed in a repetitive manufacturing process.
2. Compare Kanban, CONWIP, and hybrid pull systems, including their advantages and limitations.
3. Discuss how heijunka, takt time, pitch, and setup reduction interact to support production levelling.
4. Evaluate the strategic and operational factors involved in choosing MTS, MTO, and assemble-to-order policies.
5. Explain how supplier integration and logistics design can reduce total-system variability without merely shifting inventory or risk to suppliers.

Analytical / Application Questions

1. A component is consumed at 900 units per day. Replenishment takes 0.8 day, the variability allowance is 10%, and each container holds 75 units. Calculate the required number of Kanban containers and state two assumptions behind the calculation.
2. A product family requires 720 units per day during 7.5 available production hours. Calculate takt time in minutes per unit. If the container quantity is 24 units, calculate pitch in minutes.
3. A firm has stable demand for common modules but highly variable demand for customized final products. Design a hybrid MTS, assemble-to-order, and MTO policy, identifying the likely decoupling point and the controls needed at each stage.

Multiple Choice Questions with Answers

Question 1

Which statement best describes a pull system?

A. Every upstream process produces to its maximum local capacity.

B. Production or movement is authorized by downstream consumption.

C. All production is released from a long-range forecast.

D. Inventory is eliminated from every point in the network.

Correct Answer: B. Production or movement is authorized by downstream consumption.

Question 2

A component has daily demand of 480 units, a replenishment loop of 0.75 day, a 15% allowance, and containers of 60 units. How many containers are required using N = (D × T × (1 + α)) ÷ C, rounded up?

A. 5

B. 6

C. 7

D. 8

Correct Answer: C. 7

Question 3

What is the primary control variable in a CONWIP system?

A. The total number of jobs or WIP authorizations in the controlled loop

B. The safety stock of every individual part number

C. The supplier’s annual forecast accuracy

D. The number of finished goods sold each month

Correct Answer: A. The total number of jobs or WIP authorizations in the controlled loop

Question 4

A line has 30 available hours per week and demand of 600 units per week. What is the takt time?

A. 2 minutes per unit

B. 3 minutes per unit

C. 5 minutes per unit

D. 30 minutes per unit

Correct Answer: B. 3 minutes per unit

Question 5

If container quantity is 20 units and takt time is 3 minutes per unit, what is the pitch?

A. 17 minutes

B. 23 minutes

C. 60 minutes

D. 600 minutes

Correct Answer: C. 60 minutes

Question 6

Which item is generally the strongest candidate for a repetitive Kanban supermarket?

A. A unique engineered product with no demand history

B. A stable, high-volume item with reliable replenishment

C. An obsolete item awaiting disposal

D. A product requiring a new regulatory design approval

Correct Answer: B. A stable, high-volume item with reliable replenishment

Question 7

In the order-up-to expression S = μ(L + R) + SS, what does R represent?

A. The reorder quantity

B. The review interval

C. The risk coefficient

D. The return rate

Correct Answer: B. The review interval

Question 8

Which practice most directly reduces the bullwhip effect?

A. Increasing order batch sizes

B. Sharing actual consumption data and shortening lead times

C. Using frequent temporary price promotions

D. Allowing each echelon to forecast independently

Correct Answer: B. Sharing actual consumption data and shortening lead times

Question 9

Which policy best fits common modules that are configured after a customer order?

A. Assemble-to-order

B. Engineer-to-disposal

C. Pure make-to-order for every component

D. Uncontrolled push production

Correct Answer: A. Assemble-to-order

Question 10

What is a central purpose of heijunka?

A. To maximize local utilization by producing the largest possible batches

B. To smooth product mix and quantity over time

C. To remove all safety stock regardless of risk

D. To replace capacity planning with customer forecasts

Correct Answer: B. To smooth product mix and quantity over time