Optimization and Operations Analytics

Introduction

International businesses routinely allocate scarce production capacity, transport goods across borders, assign employees to projects, replenish inventories, select logistics routes, and manage waiting lines. Optimization converts these decisions into explicit mathematical models. Operations analytics then combines model results with costs, service requirements, risk, regulation, and managerial judgment.

An optimal mathematical solution is not automatically an implementable decision. Data may be uncertain, shipments may require whole containers, employees may have legal working-hour limits, and suppliers may fail. Managers must therefore test assumptions, examine sensitivity, and translate outputs into feasible actions.

Learning Objectives

After studying this chapter, students should be able to:

• formulate linear programming models;
• interpret feasible regions, binding constraints, slack, and alternative optima;
• use dual values and sensitivity ranges in managerial analysis;
• formulate and solve transportation and assignment problems;
• apply basic inventory, network, and queuing models;
• assess assumptions, uncertainty, ethics, and implementation constraints; and
• integrate optimization results into international business decisions.

1. Optimization as a Decision Process

An optimization model contains four main elements:

1. Decision variables—quantities controlled by the decision-maker.
2. Objective function—a measure to maximize or minimize.
3. Constraints—resource, policy, technical, or regulatory restrictions.
4. Parameters—known inputs such as costs, capacities, and demand.

A model is useful only when its variables and constraints correspond to real decisions. A disciplined process is:

• define the managerial problem and planning horizon;
• identify controllable decisions;
• collect and validate data;
• formulate the model;
• solve and verify it;
• conduct sensitivity and scenario analysis;
• assess operational feasibility; and
• implement, monitor, and revise the decision.

2. Linear Programming

Linear programming (LP) optimizes a linear objective subject to linear constraints. A general maximization model is:

subject to

Here, x_j is decision variable j, c_j is its objective contribution per unit, a_ij is the amount of resource i consumed per unit, and b_i is the available amount of resource i. Units must be consistent: if c_j is dollars per unit and x_j is units, then Z is measured in dollars.

2.1 Assumptions

Classical LP assumes:

• Proportionality: each unit has a constant contribution and resource use.
• Additivity: total effects equal the sum of individual effects.
• Divisibility: fractional variable values are permitted.
• Certainty: coefficients are known during the planning period.
• Non-negativity: negative activity levels are normally impossible.

When indivisible decisions are required, integer or mixed-integer programming is more appropriate.

2.2 Feasibility and optimality

A solution is feasible if it satisfies every constraint. The set of all feasible solutions is the feasible region. A binding constraint holds with equality at a solution; a nonbinding constraint has unused capacity called slack.

For an LP with a nonempty bounded feasible region, at least one optimal solution occurs at an extreme point. This does not mean that the optimum is always unique. An entire edge or higher-dimensional face can be optimal when the objective is parallel to a binding boundary.

3. Worked Production Model: Alternative Optimal Solutions

A furniture exporter produces tables and cabinets. Let:

• x = table batches produced per week;
• y = cabinet batches produced per week.

A table batch contributes $80 and uses 4 timber units and 2 labor units. A cabinet batch contributes $60 and uses 3 timber units and 3 labor units. Weekly capacities are 48 timber units and 36 labor units.

subject to

Because

the objective cannot exceed 20(48) = 960. Every feasible point using all 48 timber units therefore earns $960.

On the timber boundary,

Substituting this expression into the labor constraint gives:

Non-negativity of y requires x \le 12. Consequently, every plan satisfying

is optimal. There are infinitely many optimal solutions along the feasible timber boundary from (6,8) to (12,0). Examples include (6,8), (9,4), and (12,0), each producing $960.

This flexibility has managerial value. The exporter can choose among the optimal plans using considerations absent from the LP, such as contractual demand, product diversification, worker skills, storage space, or currency exposure.

4. Duality, Shadow Prices, and Sensitivity Analysis

4.1 The dual model

For a primal maximization model in the form Ax \le b with x \ge 0, the dual is:

subject to

Each dual variable represents the marginal value of a corresponding resource, subject to the validity range of the current solution structure.

For the furniture model, let u₁ be the timber shadow price and u₂ the labor shadow price. Since both products are positive at (6,8), their dual constraints hold with equality:

The second equation gives u₁+u₂=20. Dividing the first by 2 gives 2u₁+u₂=40. Subtraction yields:

Thus, within the applicable RHS range, one additional timber unit increases maximum contribution by $20, while an additional labor unit has no marginal value. Labor has a zero shadow price because timber remains the economically limiting resource and alternative optimal production mixes exist.

4.2 Reduced cost

For a maximization problem, the reduced cost of a nonbasic variable x_j can be written as:

Under this convention, \bar{c}_j \le 0 at an optimum. It measures how much the unit contribution must improve before a currently zero variable can become attractive, assuming the current basis remains relevant.

4.3 Allowable objective-coefficient ranges

An objective-coefficient range identifies values for which a selected solution remains optimal when one coefficient changes and all other data remain fixed.

At (6,8), let the table and cabinet coefficients be c_x and c_y. Optimality requires nonnegative dual multipliers satisfying:

Holding c_y=60 gives u₁+u₂=20. Solving for the multipliers produces:

Both are nonnegative when:

Therefore, with cabinet contribution fixed at $60, the point (6,8) remains optimal when table contribution lies between $40 and $80. At $80, the entire timber segment is optimal; for 40<c_x<80, (6,8) is the unique optimum.

Holding c_x=80 instead gives:

Hence the allowable one-at-a-time range is:

These ranges are not simultaneous-change guarantees. If both coefficients change, the full optimality conditions must be checked.

4.4 Allowable RHS ranges

RHS analysis asks how much resource availability can change while the current basis remains feasible. Let timber capacity be b₁ and labor capacity be b₂. Under the basis in which both production variables are positive and both constraints bind:

Solving gives:

Holding b₂=36, feasibility requires:

Holding b₁=48, feasibility requires:

For example, increasing timber capacity from 48 to 52 while labor remains 36 gives:

The new objective value is $1,040, an increase of $80. This equals the timber shadow price multiplied by the four-unit increase: 20 \times 4 = 80. Shadow-price extrapolation outside the allowable range is not justified because the active solution structure can change.

5. Transportation Models

A transportation model minimizes the cost of shipping a homogeneous product from origins to destinations. Let:

• x_ij = quantity shipped from origin i to destination j;
• c_ij = cost per unit shipped;
• s_i = supply at origin i;
• d_j = demand at destination j.

For a balanced problem, total supply equals total demand:

The model is:

subject to

An unbalanced model can be balanced with a dummy origin or destination, but the dummy cost must represent the actual meaning of shortage, unused capacity, or disposal rather than being assigned mechanically.

5.1 Worked transportation solution

Two distribution centers, A and B, serve three markets. Unit shipping costs are:

Suppose an initial feasible allocation is x_A1=20, x_A2=10, x_A3=10, x_B2=30, and x_B3=30. Its cost is:

Use the closed loop B3(+), A3(−), A2(+), B2(−). The largest feasible adjustment is 10 units. The revised allocation is:

Every supply and demand requirement is satisfied. The cost is:

The loop saves 10(10-7-5+4)=20 dollars. Using transportation potentials for the occupied cells gives nonnegative opportunity costs for the unused A3 and B1 routes, confirming optimality. The optimal transportation cost is therefore $540.

International applications may also include tariffs, insurance, carbon charges, port handling, sanctions, lead-time penalties, and route-capacity restrictions.

6. Assignment Models

The assignment model allocates agents to tasks on a one-to-one basis. Let x_ij=1 if agent i is assigned to task j, and 0 otherwise.

subject to

Consider the following completion times in hours:

The optimal assignment is P→2, Q→1, and R→3, with total time 2+6+1=9 hours. The Hungarian algorithm solves larger square assignment problems efficiently. Dummy rows or columns can balance unequal numbers of agents and tasks.

Real assignments may require skill, language, visa, independence, or workload constraints. These extensions generally require binary programming rather than the basic assignment algorithm.

7. Inventory Analytics

Inventory balances service continuity against ordering, holding, shortage, obsolescence, and capital costs.

7.1 Economic order quantity

The basic EOQ model assumes constant known demand, constant lead time, instantaneous replenishment, no shortages, and constant ordering and holding costs. Annual relevant cost is:

where D is annual demand in units, S is ordering cost per order, H is annual holding cost per unit, and Q is order quantity.

Differentiating and setting the derivative to zero gives:

so the economic order quantity is:

If D=12{,}000 units per year, S=75 dollars per order, and H=3 dollars per unit per year:

The annual ordering and cycle-stock holding costs are each approximately $1,162, giving a combined relevant cost of about $2,324.

7.2 Reorder point

With deterministic demand and lead time:

where d is demand per time period and L is lead time in matching periods. If daily demand is 40 units and lead time is 6 working days, ROP=240 units.

Under uncertainty:

where SS is safety stock selected for a target service level. International supply chains often require larger safety stocks because customs delays, strikes, exchange controls, and maritime disruption make lead time variable.

8. Network Models

Networks represent locations or activities as nodes and relationships as arcs.

8.1 Shortest path

A shortest-path model minimizes total distance, time, cost, or risk from a source to a destination. Arc weights must be defined consistently; combining cost and risk requires a defensible common scale.

Figure: Directed logistics network; numbers indicate arc costs.

In the illustrated network, the least-cost path from A to E is A→C→B→D→E, with cost 2+1+5+2=10. Dijkstra's algorithm applies when all arc weights are nonnegative.

8.2 Maximum flow and project networks

A maximum-flow model maximizes movement through a network subject to arc capacities and flow conservation at intermediate nodes. Applications include port throughput, telecommunications, and pipeline capacity.

Critical path analysis represents project activities and precedence relationships. The critical path is the longest-duration path through the project network and determines minimum completion time under fixed activity durations. Activities with zero total slack require particular managerial attention.

9. Basic Queuing Analytics

Queuing models examine congestion when customers, documents, vehicles, or transactions arrive for service. The M/M/1 model assumes:

• Poisson arrivals at mean rate \lambda;
• independent exponential service times at rate \mu;
• one server;
• first-come, first-served processing;
• a sufficiently large calling population and waiting capacity; and
• steady-state stability, requiring \lambda<\mu.

The utilization rate is:

Steady-state performance measures are:

Here, L and L_q are expected numbers in the system and queue; W and W_q are expected times in the system and queue. Time units follow the units used for the rates.

If customs documents arrive at 8 per hour and one analyst processes 10 per hour:

High utilization can create disproportionate delays. The formulas should not be used when arrivals are highly scheduled, service times are not approximately exponential, priorities dominate processing, or the operation has multiple servers without an appropriate model.

10. Illustrative Integrated Case: Meridian Furnishings

Background

Meridian Furnishings is an illustrative exporter serving three regional markets. Management must coordinate production, pallet distribution, inventory replenishment, route selection, and export-document processing.

Problem

Timber and labor constrain weekly production. An existing 100-pallet order must be allocated from two distribution centers. Imported fittings require economical replenishment, while document congestion threatens delivery reliability.

Analysis and solution

The production model yields maximum weekly contribution of $960. It has infinitely many optimal mixes from (6,8) to (12,0). Management chooses (9,4) because it preserves both product lines while achieving the same optimal contribution.

The transportation model allocates 20 pallets from A to Market 1, 20 from A to Market 2, 20 from B to Market 2, and 40 from B to Market 3. Total cost is $540. In particular, B→3 must carry 40 pallets for destination demand and origin supply to balance.

For imported fittings, the EOQ calculation recommends approximately 775 units per order under the stated deterministic assumptions. Meridian nevertheless tests higher safety stock because border lead times vary.

Network analysis identifies A→C→B→D→E as the least-cost route with total arc cost 10. The M/M/1 document model reports 80% utilization and an expected system time of 30 minutes. Management therefore considers cross-training a second analyst during peak periods rather than treating average processing capacity as sufficient.

Outcome

The models produce a coordinated baseline: $960 production contribution, $540 transport cost, a 775-unit EOQ, a least-cost network route of 10, and explicit service-delay measures. Managers supplement these outputs with product-demand commitments, disruption scenarios, emissions targets, and regulatory checks.

Learning points

• Alternative optima create managerial flexibility rather than model failure.
• Transportation allocations must satisfy every supply and demand equation.
• Shadow prices are local marginal values, not permanent market prices.
• Deterministic inventory and queue models require assumption testing.
• Integrated decisions should be stress-tested before implementation.

11. Implementation, Governance, and Contemporary Practice

Optimization can produce harmful decisions when objectives omit employee welfare, resilience, environmental impact, or legal obligations. International models may need constraints for sanctions, customs rules, local-content requirements, cabotage, carbon limits, labor standards, and supplier due diligence.

Data governance is equally important. Cost and capacity data should have identified owners, timestamps, currencies, and exchange-rate assumptions. Personally identifiable employee or customer data should be minimized and protected. Managers should document overrides of model recommendations and monitor unequal workload or discriminatory assignment outcomes.

Modern solvers can handle large mixed-integer, stochastic, and robust models. Digital twins can update capacities and travel times, while machine learning can forecast coefficients such as demand or lead time. However, predicted parameters remain uncertain. Combining an inaccurate forecast with an exact optimizer can produce a precisely calculated but poor decision.

Best practice includes independent arithmetic checks, reconciliation with operational totals, scenario analysis, model version control, audit trails, and post-implementation comparison of predicted and actual outcomes.

12. Challenges and Limitations

Common limitations include:

• uncertain demand, costs, exchange rates, and lead times;
• nonlinear quantity discounts or congestion effects;
• indivisible containers, shifts, and facilities;
• multiple conflicting objectives;
• omitted resilience and sustainability considerations;
• poor-quality or stale data; and
• organizational resistance to analytically generated decisions.

Sensitivity analysis addresses parameter variation near the baseline but is not a substitute for stochastic, robust, simulation, or scenario-based analysis when uncertainty is substantial.

Chapter Summary

Optimization translates business decisions into variables, objectives, and constraints. Linear programming supports resource allocation and may produce a unique optimum or infinitely many alternative optima. In the furniture model, every feasible point on the timber boundary from (12,0) to (6,8) earns $960. The corresponding shadow prices are $20 per timber unit and $0 per labor unit within their valid ranges.

Sensitivity analysis evaluates objective coefficients, RHS values, reduced costs, and shadow prices. Transportation models allocate flows between origins and destinations; the worked solution ships 40 units from B to Market 3 and has an optimal cost of $540. Assignment models allocate agents to tasks, EOQ balances ordering and holding costs, network models support routing and flow decisions, and queuing models quantify congestion. Effective managerial use requires assumption testing, governance, scenario analysis, and operational validation.

Key Terms

Alternative optimum: One of multiple feasible solutions attaining the same best objective value.

Assignment model: A binary model allocating agents to tasks, usually one-to-one.

Binding constraint: A constraint satisfied as an equality at a given solution.

Critical path: The longest-duration path determining minimum project completion time.

Decision variable: A controllable quantity selected by the decision-maker.

Dual model: A related optimization model that values primal resources or requirements.

Economic order quantity: The order size minimizing basic annual ordering and holding costs.

Feasible region: The set of all solutions satisfying every constraint.

Linear programming: Optimization of a linear objective subject to linear constraints.

Maximum flow: The largest feasible flow through a capacitated network.

Objective function: The performance measure to maximize or minimize.

Reduced cost: The marginal objective improvement needed for a nonbasic variable to become attractive under a stated convention.

Reorder point: The inventory position triggering replenishment.

Sensitivity analysis: Examination of how parameter changes affect a solution or its optimality.

Shadow price: The local marginal change in optimal objective value from a one-unit RHS increase.

Slack: Unused capacity in a less-than-or-equal constraint.

Transportation model: A model allocating shipments from origins to destinations at minimum cost.

Utilization: The proportion of service capacity used, commonly \rho=\lambda/\mu.

Review Questions

Short Answer Questions

1. What four elements form an optimization model?
2. Distinguish a binding constraint from a nonbinding constraint.
3. Why is a shadow price valid only over a limited range?
4. What assumptions underlie the basic EOQ model?
5. What stability condition is required for an M/M/1 queue?

Long Answer Questions

1. Explain the assumptions of linear programming and discuss situations in which each may fail.
2. Derive the EOQ formula from the annual relevant-cost function.
3. Explain primal-dual relationships and the managerial interpretation of shadow prices.
4. Compare transportation, assignment, shortest-path, and maximum-flow models.
5. Discuss ethical, regulatory, and data-governance issues in international optimization.

Analytical / Application Questions

1. For the furniture model, verify that (9,4) is feasible and optimal, then explain why the optimum is not unique.
2. Recalculate the transportation cost if the B→3 unit cost rises from $7 to $9, keeping the stated allocation fixed. Explain why reoptimization would then be advisable.
3. A service desk receives 12 cases per hour and processes 15 per hour. Assuming M/M/1 conditions, calculate utilization, expected queue length, and expected waiting time in the system.

Multiple Choice Questions with Answers

Question 1

What is the maximum contribution in the furniture production model?

A. $720

B. $840

C. $960

D. $1,120

Correct Answer: C. $960

Question 2

Which statement correctly describes the optimal solutions of the furniture model?

A. Only (12,0) is optimal.

B. Only (6,8) is optimal.

C. Every feasible point is optimal.

D. Every feasible point on the timber boundary from (12,0) to (6,8) is optimal.

Correct Answer: D. Every feasible point on the timber boundary from (12,0) to (6,8) is optimal.

Question 3

What are the timber and labor shadow prices, respectively, in the furniture model?

A. $20 and $0

B. $0 and $20

C. $28 and $11.20

D. $80 and $60

Correct Answer: A. $20 and $0

Question 4

What is the optimal cost of the worked transportation allocation?

A. $470

B. $520

C. $540

D. $560

Correct Answer: C. $540

Question 5

In the optimal transportation allocation, how many units are shipped from B to Market 3?

A. 20

B. 30

C. 40

D. 50

Correct Answer: C. 40

Question 6

Using D=12{,}000, S=75, and H=3, what is the EOQ to the nearest unit?

A. 600 units

B. 775 units

C. 900 units

D. 1,162 units

Correct Answer: B. 775 units

Question 7

What is the minimum total completion time in the worked assignment problem?

A. 8 hours

B. 9 hours

C. 10 hours

D. 12 hours

Correct Answer: B. 9 hours

Question 8

For an M/M/1 queue with \lambda=8 per hour and \mu=10 per hour, what is the expected time in the system?

A. 0.1 hour

B. 0.4 hour

C. 0.5 hour

D. 0.8 hour

Correct Answer: C. 0.5 hour

Question 9

Holding cabinet contribution at $60, over what range does (6,8) remain optimal as table contribution c_x changes?

A. 0 \le c_x \le 40

B. 40 \le c_x \le 80

C. 60 \le c_x \le 120

D. 80 \le c_x \le 160

Correct Answer: B. 40 \le c_x \le 80

Question 10

What is the least-cost path value from A to E in the illustrated network?

A. 8

B. 9

C. 10

D. 11

Correct Answer: C. 10