Pull Systems, Kanban and Lean Inventory

Introduction

Lean operations seek to create customer value with fewer resources, shorter lead times, and less waste. Inventory is central to this challenge. Inventory can protect a system from uncertainty, but it can also conceal problems, consume cash, increase handling, and lengthen feedback time. Pull systems provide a disciplined way to replenish materials in response to actual consumption rather than producing solely from forecasts or schedules.

A pull system does not mean that forecasting is unnecessary. Forecasts remain important for capacity planning, purchasing agreements, workforce decisions, and aggregate production planning. The distinctive feature of pull is that authorization to replenish or produce is triggered by downstream consumption, within defined capacity and inventory limits.

This chapter examines Kanban, supermarkets, CONWIP, Just-in-Time (JIT), buffer design, supplier integration, demand variability, and stockout risk. It also considers where pull systems work well, where they require modification, and how digital technologies can support—rather than replace—sound operating discipline.

Learning Objectives

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

1. Distinguish push, pull, Kanban, CONWIP, supermarket, and JIT systems.
2. Design a basic Kanban replenishment loop and calculate card or container requirements.
3. Determine safety-stock and reorder-point requirements under stated assumptions.
4. Analyse the effects of demand variability, lead-time variability, batch size, and service-level targets.
5. Evaluate supplier integration and inventory trade-offs.
6. Diagnose failure modes and recommend an appropriate replenishment policy.
7. Assess the role and limitations of digital Kanban, RFID, sensors, and AI-enabled planning.

1. Push and Pull as Operating Logics

In a push system, production or replenishment is authorized primarily by a plan, forecast, or material requirements schedule. Each process attempts to meet a planned quantity, potentially creating excess work-in-process when the next process is unable to consume it.

In a pull system, a downstream customer, process, or point of use signals the need for replenishment. Upstream activity is therefore constrained by actual consumption and by the number of replenishment authorizations in circulation. Pull is a control logic, not simply a low-inventory target.

A mature operation often uses a hybrid system: push for long-range capacity and supplier planning, and pull for execution and replenishment near the point of use.

2. The Lean Inventory Perspective

Lean does not define inventory as intrinsically wasteful. Inventory has legitimate functions:

• Cycle stock covers normal replenishment quantities.
• Safety stock protects against uncertainty in demand or replenishment.
• Pipeline inventory is in transit or in process because of lead time.
• Anticipation inventory is accumulated for seasonality, planned shutdowns, or known events.
• Decoupling inventory separates processes whose rates or reliability differ.
• Strategic inventory protects against major supply disruption or geopolitical risk.

The managerial question is not “How can all inventory be eliminated?” but “What uncertainty, process condition, or policy is each unit of inventory addressing, and is that protection economically justified?” Excess inventory can hide unreliable equipment, long changeovers, poor quality, inaccurate records, and unstable suppliers. Reducing it exposes these conditions, which must then be solved rather than merely tolerated.

Relevant performance measures include inventory turns, days of supply, fill rate, stockout frequency, order-line service, throughput, lead time, schedule adherence, and total cost. A reduction in inventory is not a lean improvement if it causes disproportionate shortages, expediting, overtime, or lost customers.

3. Just-in-Time and the Architecture of Pull

JIT is a management philosophy and operating system that aims to provide the right item, in the right quantity, at the right time, while eliminating activities that do not add value. It depends on several mutually reinforcing conditions:

• stable and levelled demand where possible;
• short and dependable replenishment lead times;
• small, economical transfer and production batches;
• high equipment availability;
• quality at the source;
• standard work and visual control;
• capable, integrated suppliers; and
• rapid problem escalation.

JIT is often misunderstood as “zero inventory.” In practice, it seeks inventory levels appropriate to process capability and risk. If a process is unstable, reducing its buffer without improving capability may produce repeated shortages rather than lean flow.

A pull architecture commonly contains three elements:

1. Demand point: the location where material is consumed.
2. Replenishment signal: a card, electronic message, container, barcode scan, or other authorization.
3. Supply point: the upstream process or supplier that replenishes the withdrawal.

The signal should contain sufficient information to identify the item, quantity, source, destination, and handling requirements. It should not authorize production beyond the agreed number of active signals.

4. Kanban Systems

Kanban means a visible signal or signboard. In a conventional two-card system, a withdrawal kanban authorizes movement of a specified quantity from a supermarket to a consuming process, while a production kanban authorizes the supplying process to make a replacement quantity. A one-card system may combine these functions where movement and production are closely linked.

A supermarket is a controlled, deliberately located inventory point from which a downstream process withdraws material. The upstream process replenishes what has been withdrawn. Unlike a conventional warehouse, a supermarket in a lean system has defined item locations, standard quantities, replenishment rules, and visual limits.

A basic operating cycle is:

1. The customer process withdraws a container or quantity from the supermarket.
2. The withdrawal signal is returned to the supplying process.
3. The supplier produces or retrieves the specified item.
4. The replenished container is delivered to the supermarket.
5. The customer consumes the item and repeats the cycle.

The fundamental rule is that a kanban is an authorization, not merely an information label. If cards are duplicated, lost, or released without consumption, the system becomes a hidden push system.

4.1 Kanban quantity calculation

A widely used starting point for the number of kanban containers is:

where:

• N = number of kanban cards or containers;
• D = average demand rate, in units per time;
• L = replenishment lead time, in the same time unit;
• S = safety factor expressed as a decimal, such as 0.10 for 10%;
• C = standard quantity per container, in units; and
• ⌈x⌉ = round upward to the next whole number.

The formula assumes relatively stable average demand, a defined replenishment lead time, consistent container quantity, and that the safety factor adequately represents uncertainty. It is a planning approximation, not a substitute for validating actual service performance. If demand and lead time are highly variable, a probabilistic safety-stock analysis is preferable.

Worked example: Kanban containers

A machining cell consumes 240 components per eight-hour shift. Replenishment takes 0.5 shift. Management applies a 15% allowance for variability. Each container holds 30 components.

Thus, five containers or five equivalent production authorizations are required under the stated assumptions. The allowance should be reviewed after observing actual shortages, excess capacity, and lead-time performance. Increasing the number of cards is not the preferred first response to every shortage; the root cause may be quality loss, downtime, inaccurate inventory records, or oversized production batches.

4.2 Designing the kanban loop

A sound design specifies:

• demand rate and demand pattern;
• container quantity and ergonomic constraints;
• replenishment lead time, including transport and queue time;
• supermarket location and maximum quantity;
• signal ownership and scanning responsibility;
• delivery frequency and route discipline;
• quality-release rules;
• escalation rules for shortages; and
• rules for changing the number of signals.

Container size should balance handling efficiency, product protection, working capital, and replenishment frequency. Smaller containers reduce average inventory and expose problems sooner, but may increase handling and transport cost. The appropriate decision is a total-system trade-off.

5. CONWIP and Other Pull Variants

CONWIP—constant work in process—controls the total number of jobs in a line or production loop rather than assigning a separate kanban for every item-location combination. When a finished job exits, a card or electronic authorization becomes available for a new job to enter. CONWIP is particularly useful in high-mix environments where product-specific kanban loops would require too many cards.

Kanban controls replenishment of defined items between specified locations. CONWIP controls aggregate work-in-process. A hybrid system may use CONWIP at a fabrication line and kanban supermarkets for common components.

6. Inventory and Replenishment Calculations

6.1 Reorder point

For continuous-review replenishment, the reorder point is generally:

where μ_L is expected demand during lead time and SS is safety stock. If average demand per unit time is d and lead time is L, then:

These equations assume that demand and lead time are measured consistently and that the reorder point is based on inventory position—on-hand inventory plus on-order inventory minus allocated or backordered quantity—not merely physical on-hand stock.

6.2 Safety stock with variable demand and lead time

If demand per period has standard deviation σ_d, lead time has standard deviation σ_L, average demand per period is d̄, and demand and lead time are treated as independent, the standard deviation of demand during lead time can be approximated by:

where L̄ is average lead time. A safety stock based on a cycle-service target is:

where z is the standard-normal value associated with the selected probability of not stocking out during a replenishment cycle. For example, a 95% cycle-service target corresponds approximately to z = 1.645 under a normal approximation.

The formula excludes an interaction term because demand and lead time are assumed independent. If they are correlated, an additional covariance term is required:

The approximation is most defensible when demand is reasonably represented by a stable distribution and replenishment lead time is not extremely skewed. Intermittent demand, promotion-driven demand, supply disruption, and highly non-normal data require empirical or simulation-based methods.

Worked example: reorder point

A service-parts operation has average daily demand of 80 units and average replenishment lead time of 4 days. Daily demand standard deviation is 12 units, lead-time standard deviation is 1 day, and demand and lead time are assumed independent. The target cycle service level is 95%, so z = 1.645.

Expected lead-time demand is:

Demand during lead-time standard deviation is:

Safety stock is:

Rounding to whole units, safety stock is approximately 138 units and the reorder point is:

This result is sensitive to lead-time variability. The example also illustrates why a policy that considers demand variation but ignores supplier or transport variation can materially understate protection requirements.

6.3 Service measures and cost trade-offs

Cycle service level is the probability that no stockout occurs during a replenishment cycle. Fill rate is the proportion of total demand supplied immediately from available stock. These measures are related but not identical: one stockout can create a low cycle-service result while affecting only a small fraction of unit demand, or it can create a large fill-rate loss if the shortage is severe.

The economic order quantity model is sometimes used to examine ordering trade-offs:

where D is annual demand in units per year, S is ordering or setup cost per order, and H is annual holding cost per unit. The model assumes relatively constant demand, constant lead time, no quantity discounts, instantaneous replenishment, and no stockouts. These assumptions often conflict with lean objectives, especially when setup reduction makes small batches economical. Therefore, Q^* should be treated as a benchmark rather than an automatic batch-size recommendation.

7. Demand Variability, Level Loading and Stockout Risk

Pull systems perform best when demand is levelled. Heijunka, or production smoothing, reduces abrupt changes in product mix and volume. It does not eliminate demand uncertainty, but it converts large, irregular releases into a more stable sequence where possible.

Variability has several sources:

• random customer demand;
• seasonality and promotions;
• product mix changes;
• supplier delays;
• machine downtime;
• quality failures;
• transport disruption; and
• inaccurate transaction data.

The appropriate response depends on the source. Increasing safety stock may be justified for irreducible market uncertainty, but it is a poor substitute for preventive maintenance, supplier development, faster changeovers, or accurate inventory transactions. Managers should separate common-cause variation, inherent in the process, from special-cause variation, arising from identifiable abnormal events.

Stockout risk should be segmented by item criticality. A low-cost standard fastener and a safety-critical spare should not necessarily have the same service target. A useful policy combines value, demand frequency, operational criticality, obsolescence risk, and recovery time. High service targets can create disproportionately large safety stocks when variability is high; management should evaluate the cost of shortage against the cost and risk of carrying inventory.

8. Supplier Integration and External Pull

Extending pull beyond the factory requires more than sending a supplier a daily consumption file. Effective supplier integration commonly includes:

• agreed item definitions, packaging, and quantities;
• stable delivery windows and route schedules;
• shared visibility of consumption and inventory position;
• clear ownership of quality and nonconforming material;
• capability and capacity assessments;
• contingency plans for disruption;
• electronic data interchange or API integration where beneficial; and
• joint review of lead time, variability, and improvement actions.

Supplier-managed inventory can reduce administrative work, but governance must define ownership, replenishment authority, data access, payment terms, liability for obsolescence, and escalation procedures. Supplier integration must not become unilateral transfer of inventory risk to a weaker partner. Ethical procurement requires fair commercial arrangements, realistic lead times, responsible labor expectations, and transparency regarding forecast changes.

Milk-run logistics, frequent scheduled deliveries, returnable containers, and supplier parks can support external pull. However, frequent deliveries may increase transport emissions unless routes are consolidated and vehicle utilization is considered. Lean decisions should therefore evaluate total environmental impact rather than only factory inventory.

9. Digital Kanban and AI-Enabled Replenishment

Electronic Kanban replaces physical cards with scans, mobile devices, RFID, sensors, or system messages. Digital systems can improve traceability, reduce lost signals, and provide real-time status. They cannot correct poor process definitions, inaccurate bills of material, incorrect locations, or unstable lead times.

AI and advanced analytics may assist by:

• detecting abnormal consumption;
• forecasting demand at suitable aggregation levels;
• identifying lead-time changes;
• recommending dynamic safety-stock adjustments;
• predicting equipment failure that could affect replenishment; and
• simulating alternative buffer policies.

Governance is essential. Algorithms should expose assumptions, data quality, confidence ranges, and override rules. Automatic increases in inventory can mask process deterioration, while automatic reductions can create shortages during structural demand changes. Human accountability remains necessary for critical items, supplier disruptions, and safety-related decisions.

10. Implementation Framework

A practical implementation sequence is:

1. Define the customer and consumption point. Establish what demand the system is intended to serve.
2. Map the material and information flow. Identify queues, decoupling points, supermarkets, transport loops, and delays.
3. Stabilize the process. Improve quality, equipment reliability, standard work, changeovers, and transaction accuracy.
4. Classify items. Segment by demand, criticality, value, variability, and replenishment characteristics.
5. Select the control method. Use item-specific kanban, CONWIP, reorder point, scheduled replenishment, or a hybrid policy.
6. Set quantities and signals. Calculate an initial requirement, round to practical containers, and document assumptions.
7. Pilot and measure. Track availability, stockouts, lead time, WIP, inventory, expedites, and signal accuracy.
8. Adjust scientifically. Change one policy variable at a time where possible and distinguish symptoms from causes.
9. Standardize and audit. Define ownership, visual limits, transaction rules, and escalation.
10. Improve continuously. Reduce lead time and variability so that buffers can be reduced without damaging service.

Applied Case: Illustrative Medical-Device Assembly Cell

Background

A medical-device manufacturer assembles a family of diagnostic units. A final assembly cell consumes a common connector from a supermarket. The connector supplier delivers twice daily, and the plant uses returnable containers of 40 connectors. Average consumption is 360 connectors per shift. The replenishment loop from empty-container scan to delivery averages 0.75 shift. The initial variability allowance is 20%.

Problem

The cell has experienced intermittent shortages despite apparently adequate average inventory. Investigation finds that delivery lead time varies from 0.5 to 1.25 shifts, quality inspection occasionally holds a container, and operators sometimes remove connectors without scanning the empty container. The production manager proposes adding three more kanban cards.

Analysis

The basic calculation is:

However, the calculation alone does not diagnose the shortage. The current loop contains nine containers, but missing scans reduce the effective number of active signals. Quality holds also extend replenishment lead time. Adding cards may protect service temporarily, but it increases inventory and can conceal the causes.

Approach and solution

The team introduces a standardized scan at point of use, a quarantine location for held material, and an escalation rule requiring supplier notification within 15 minutes of a quality hold. It separates production authorization from quality-release authorization and establishes a fixed milk-run schedule. A short CONWIP limit is applied to the assembly cell so that work-in-process does not expand when connectors are unavailable. Supplier and plant personnel jointly review the lead-time distribution weekly.

A controlled pilot retains nine cards for two weeks, then tests eight cards after scan compliance reaches 100% and quality-hold response improves. The decision is based on achieved service, not on an arbitrary inventory-reduction target.

Outcome

The pilot reduces unplanned expedites and improves inventory-record accuracy. The plant does not claim that eight cards are universally correct; the number is valid only for the measured demand, container size, lead-time performance, and service requirement. The case demonstrates that pull design is socio-technical: containers and formulas matter, but standard work, quality, supplier behavior, and data integrity determine whether the system functions.

Learning points

1. A calculated buffer is not effective if signals are not captured.
2. Adding inventory can treat a symptom rather than a cause.
3. Quality holds and transaction failures are part of replenishment lead time.
4. Pull systems require explicit ownership and escalation.
5. Buffer reduction should follow process improvement and verified service performance.

Challenges and Limitations

Pull systems can fail when demand is highly erratic, products are engineered to order, lead times are extremely long, suppliers are unreliable, or capacity is insufficient. In such settings, a pure consumption-triggered system may respond too slowly. Forecast-based planning, strategic inventory, capacity reservations, or risk pooling may be necessary.

Other limitations include signal proliferation, excessive transaction discipline, gaming of service measures, local optimization, and the mistaken belief that high resource utilization is always desirable. A bottleneck may require a protective buffer to preserve throughput. Similarly, reducing inventory at one location may merely move it upstream or downstream.

Managers should watch for these warning signs:

• frequent emergency overrides;
• large unexplained differences between system and physical stock;
• kanban cards added without root-cause analysis;
• stockouts concentrated in one supplier or process;
• excessive finished goods despite low WIP; and
• service targets that ignore item criticality or customer consequences.

Managerial and Professional Implications

Operations managers must balance flow, responsiveness, cost, resilience, and sustainability. The appropriate inventory policy is a strategic choice shaped by customer promise, process capability, supply-network risk, and cash constraints. A low inventory figure is not automatically evidence of excellence, just as a high service level is not automatically economically optimal.

Professionals should be able to explain the assumptions behind a buffer, challenge unreliable data, distinguish a shortage caused by policy from one caused by execution, and communicate trade-offs across purchasing, production, logistics, finance, quality, and sales. Pull systems work when the organization manages the entire replenishment loop rather than optimizing isolated departments.

Chapter Summary

Pull systems authorize production or replenishment from actual downstream consumption, while push systems rely primarily on plans and forecasts. Kanban uses visible or electronic signals to control item-specific replenishment, often through supermarkets. CONWIP controls the total amount of work in a production loop and is useful in high-mix environments. JIT combines pull with stable flow, small batches, reliable equipment, quality at the source, and capable suppliers.

Kanban quantities can be estimated from demand, replenishment lead time, variability allowance, and container size, but the result must be validated against actual service and process conditions. Reorder points combine expected lead-time demand with safety stock. When both demand and lead time vary, safety stock should reflect both sources of uncertainty and the assumptions of the statistical model. Service level, fill rate, inventory cost, shortage cost, and criticality must be considered together.

Successful pull systems depend on stable processes, accurate data, disciplined signals, supplier integration, level loading, and rapid problem solving. Digital systems and AI can improve visibility and analysis, but they cannot replace process capability or managerial judgment. Inventory should be treated as a deliberate risk-control resource, not as an isolated target for reduction.

Key Terms

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

Push system: A system in which production or replenishment is authorized mainly by forecasts or schedules.

Kanban: A visual or electronic signal authorizing movement or production of a defined quantity.

Supermarket: A controlled inventory point from which downstream users withdraw material and which upstream processes replenish.

Withdrawal kanban: A signal authorizing movement of material from a supplying point.

Production kanban: A signal authorizing production of a specified quantity.

CONWIP: Constant work in process; a pull method that limits total jobs in a defined loop.

Just-in-Time: An operating philosophy focused on supplying required items in required quantities at required times with minimal waste.

Cycle stock: Inventory used to meet normal replenishment-cycle demand.

Safety stock: Inventory held to protect against uncertainty.

Pipeline inventory: Inventory currently in transit or moving through a process.

Reorder point: Inventory position at which replenishment is triggered.

Inventory position: On-hand plus on-order inventory minus allocated or backordered quantity.

Cycle service level: Probability of avoiding a stockout during a replenishment cycle.

Fill rate: Proportion of demand supplied immediately from available inventory.

Heijunka: Production smoothing or levelling of volume and product mix.

Lead-time variability: Variation in the time required to replenish an item.

Supplier-managed inventory: An arrangement in which the supplier monitors and replenishes agreed customer inventory.

Inventory turns: A ratio indicating how many times inventory is consumed or replenished over a period.

Quality at the source: Designing work so that defects are detected and prevented at the point where they occur.

Review Questions

Short Answer Questions

1. What is the fundamental difference between a push system and a pull system?
2. What is the function of a supermarket in a Kanban system?
3. Distinguish cycle stock from safety stock.
4. Why is inventory position more useful than on-hand inventory alone for reorder decisions?
5. State two conditions necessary for successful JIT operation.

Long Answer Questions

1. Explain the architecture of a two-card Kanban system and discuss how signal duplication can undermine pull control.
2. Compare Kanban and CONWIP in terms of control variable, applicability, advantages, and limitations.
3. Discuss how demand variability and lead-time variability affect safety-stock decisions.
4. Explain why inventory reduction should be combined with process stabilization and supplier development.
5. Evaluate the opportunities and governance risks associated with electronic Kanban and AI-enabled replenishment.

Analytical / Application Questions

1. A component has average demand of 150 units per day, replenishment lead time of 2 days, a 10% variability allowance, and a container size of 25 units. Calculate the initial number of Kanban containers using the chapter formula and state the assumptions behind your answer.
2. A plant experiences frequent stockouts even though its calculated Kanban quantity appears sufficient. Develop a diagnostic plan that separates demand variation, lead-time variation, quality losses, equipment downtime, and transaction errors.
3. Design a hybrid replenishment system for a high-mix assembly operation that uses CONWIP for the main production loop and Kanban supermarkets for common components. Justify the control points, measures, and escalation rules.

Multiple Choice Questions with Answers

Question 1

Which statement best describes a pull system?

A. Production is authorized primarily by a long-range forecast.

B. Replenishment is triggered by downstream consumption.

C. Every process maximizes utilization regardless of demand.

D. Inventory is eliminated from every location.

Correct Answer: B. Replenishment is triggered by downstream consumption.

Question 2

A supermarket in a lean replenishment system is primarily used to:

A. Store uncontrolled excess inventory.

B. Replace quality inspection permanently.

C. Provide controlled stock that is replenished after withdrawal.

D. Schedule annual production capacity.

Correct Answer: C. Provide controlled stock that is replenished after withdrawal.

Question 3

Using N = ⌈D × L × (1 + S) / C⌉, what is N when D = 200 units/day, L = 0.5 day, S = 0.20, and C = 20 units?

A. 5 containers

B. 6 containers

C. 7 containers

D. 8 containers

Correct Answer: B. 6 containers

Question 4

What does CONWIP primarily control?

A. The total number of jobs in a defined production loop

B. The safety stock of every individual component

C. The annual purchasing budget

D. The number of suppliers in a network

Correct Answer: A. The total number of jobs in a defined production loop

Question 5

Under the chapter’s reorder-point model, the reorder point equals:

A. Annual demand divided by holding cost

B. Expected lead-time demand plus safety stock

C. Container quantity minus cycle stock

D. Ordering cost plus setup cost

Correct Answer: B. Expected lead-time demand plus safety stock

Question 6

If demand and lead time are both variable, safety stock should generally consider:

A. Only average demand

B. Only container size

C. Both demand variability and lead-time variability

D. Only annual ordering cost

Correct Answer: C. Both demand variability and lead-time variability

Question 7

Which condition most directly supports successful JIT operation?

A. Large batches and long queues

B. Unreliable equipment protected by large buffers only

C. Stable processes, short lead times, and quality at the source

D. Production based exclusively on optimistic forecasts

Correct Answer: C. Stable processes, short lead times, and quality at the source

Question 8

Why can adding Kanban cards fail to solve a shortage problem?

A. Cards always reduce capacity.

B. The shortage may be caused by quality, downtime, lead time, or transaction failures.

C. Kanban cards cannot authorize replenishment.

D. More cards always reduce available inventory.

Correct Answer: B. The shortage may be caused by quality, downtime, lead time, or transaction failures.

Question 9

A 95% cycle-service target under the normal approximation uses approximately which value of z?

A. 0.675

B. 1.000

C. 1.645

D. 2.576

Correct Answer: C. 1.645

Question 10

Which statement about digital Kanban and AI is most accurate?

A. Technology removes the need for accurate inventory data.

B. AI should automatically increase inventory whenever variation occurs.

C. Digital tools improve visibility and analysis but do not replace process discipline.

D. Electronic signals convert every push system into a pull system automatically.

Correct Answer: C. Digital tools improve visibility and analysis but do not replace process discipline.