Optimizing Service Capacity: Balancing Cost and Wait Times

Introduction

In our last lesson, we delved into the M/M/1 queueing model. We learned how to calculate key performance metrics like average wait time ($W_q$) and queue length ($L_q$). Most importantly, we uncovered the critical non-linear relationship between server utilization and wait times, highlighting the operational dangers of pushing a system with variability toward 100% capacity. This is a principle you've likely encountered when managing high-load systems, where provisioning extra capacity is the price for maintaining low latency and meeting SLAs.

Today, we will build directly on that analytical foundation to address the learning outcome: Use queueing analysis to evaluate the trade-off between service capacity cost and customer waiting cost.

We will move from simply describing a queue's performance to actively making business decisions about it. You'll learn how to model the costs associated with providing a service and the costs incurred by customers who are forced to wait. By combining these, we can determine the optimal level of service capacity that minimizes total cost.

1. The Fundamental Cost Trade-Off

At the heart of any capacity decision in a queueing environment is a fundamental economic trade-off. On one hand, increasing service capacity—by hiring more staff, buying faster equipment, or improving training—incurs direct costs. On the other hand, insufficient service capacity leads to longer queues and wait times, which have their own costs, such as lost customer goodwill, reduced employee productivity, or even direct financial penalties.

The objective for a manager is to find the "sweet spot" that balances these two opposing costs.

As the video illustrates, the total cost of a queueing system can be expressed as:

$\text{Total Cost}=\text{Service Cost}+\text{Waiting Cost}$

• Service Cost: Increases with service capacity.
• Waiting Cost: Decreases with service capacity.

Our goal is to find the service capacity that minimizes this total cost function.

2. Optimizing a Single-Server (M/M/1) System

Let's formalize this trade-off for the M/M/1 model we studied previously. This is a common scenario where you can't just add another server, but you can invest in technology, training, or process improvements to make the single existing server faster. This means we are treating the service rate, $\mu$, as our decision variable.

To do this, we need to define our cost parameters:

• $C_s$: The cost of providing service capacity, per unit of service rate, per unit of time (e.g., dollars per hour for each customer/hour of capacity).
• $C_w$: The waiting cost, per customer in the system, per unit of time (e.g., dollars per hour for each customer waiting or being served).

The total cost function, $TC$, as a function of the service rate $\mu$, is:

$TC(\mu )=(\text{Service Cost})+(\text{Waiting Cost})$
$TC(\mu )=C_s\cdot \mu +C_w\cdot L_s$

Recalling from our last lesson that for an M/M/1 system, the average number of customers in the system is $L_s=\frac{\lambda }{\mu -\lambda }$, we can write the full cost function:

$TC(\mu )=C_s\mu +C_w\left(\frac{\lambda }{\mu -\lambda }\right)$

Given your background in applied mathematics, you'll recognize that we can find the optimal service rate, ${\mu }^{\ast }$, that minimizes this function by taking the derivative with respect to $\mu$, setting it to zero, and solving for $\mu$.

The following reading walks through a concrete example and performs this exact optimization.

As the article derives, the optimal service rate ${\mu }^{\ast }$ that minimizes the total cost is given by the elegant formula:

${\mu }^{\ast }=\lambda +\sqrt{\frac{C_w\lambda }{C_s}}$

Let's apply this to the article's tech support desk example:

• Arrival Rate ($\lambda$): 8 requests/hour
• Service Cost ($C_s$): $25 per hour for each unit of service rate
• Waiting Cost ($C_w$): $40 per hour for each customer in the system

Plugging these into the formula:
${\mu }^{\ast }=8+\sqrt{\frac{40\times 8}{25}}=8+\sqrt{\frac{320}{25}}=8+\sqrt{12.8}\approx 8+3.58=11.58\text{ requests/hour}$

This result tells the manager, Sarah, that the most cost-effective service rate is approximately 11.6 requests per hour. She can now evaluate whether the investment in tools and training required to achieve this rate is justified by the total cost savings.

3. Application: Optimizing the Number of Servers (M/M/s)

A more common business decision is not about fine-tuning a single server's speed, but deciding how many servers (e.g., cashiers, support agents, processing nodes) to deploy. This moves us from the M/M/1 model to the multi-server M/M/s model.

While the underlying formulas for M/M/s performance metrics ($L_s,W_q$, etc.) are more complex, the optimization logic remains the same: calculate the total cost for different numbers of servers ($s$) and choose the value of $s$ that minimizes it.

The total cost function for an M/M/s system is:

$TC(s)=(\text{Cost of Servers})+(\text{Cost of Waiting})$
$TC(s)=C_s\cdot s+C_w\cdot L_s(s)$

Here, $C_s$ is the cost per server per unit of time (e.g., an employee's hourly wage), and $L_s(s)$ is the average number of customers in the system, which is a function of the number of servers, $s$. Since $s$ must be an integer, we don't use calculus. Instead, we simply calculate $TC(s)$ for $s=1,2,3,\dots$ and find the minimum.

Practice Exercise: Pediatric Clinic Staffing

Let's apply this to a realistic business problem. The following exercise is adapted from a textbook case study about a pediatric practice's telephone system.

Scenario:
A pediatric clinic is analyzing staffing levels for its receptionists who answer phone calls. During the busiest part of the day (9 AM to 4 PM), calls arrive randomly but at an average rate of $\lambda =42$ calls per hour. A receptionist can handle an average of $\mu =8$ calls per hour. The system functions as an M/M/s queue.

The clinic has determined the following costs:

• Server Cost ($C_s$): The total cost for one receptionist (salary, benefits, overhead) is $26.31 per hour.
• Waiting Cost ($C_w$): The cost of a customer waiting (dissatisfaction, potential for lost business) is estimated at $50 per hour for each customer in the system.

The clinic currently employs 6 receptionists during this period. Is this the optimal number? Let's find out.

Your Task:
Calculate the total hourly cost for the current staffing level ($s=6$) and then for $s=7$ and $s=8$ to find the cost-minimizing number of receptionists.

To save you from the complex M/M/s formulas, here are the pre-calculated values for the average number of customers in the system ($L_s$) for each scenario:

• For $s=6$, $L_s\approx 9.04$ customers
• For $s=7$, $L_s\approx 2.05$ customers
• For $s=8$, $L_s\approx 1.07$ customers
• For $s=9$, $L_s\approx 0.81$ customers

Use the total cost formula $TC(s)=C_s\cdot s+C_w\cdot L_s(s)$ to find the optimal s.

This analysis provides a clear, data-driven recommendation for a critical management decision. The full case study, which you can find as Exercise 14.5 in the resource "QUEUING MODELS AND CAPACITY PLANNING", explores this for different times of the day.

Conclusion

Key Takeaways:

• Capacity planning in queueing systems involves a core trade-off between service costs and waiting costs.
• The managerial goal is to find the service capacity that minimizes the total cost (the sum of service and waiting costs).
• For an M/M/1 system where the service rate $\mu$ is the variable, we can use a calculus-derived formula to find the optimal rate ${\mu }^{\ast }$.
• For an M/M/s system where the number of servers $s$ is the variable, we find the optimal number $s^{\ast }$ by iteratively calculating the total cost for each integer value of $s$ and identifying the minimum.
• This analytical approach transforms a complex staffing or investment decision into a structured, quantitative problem.

Next Steps

In this and the previous lesson, we have focused on managing service operations where customers or tasks form queues. In our next lesson, we will shift our focus to managing physical goods. We will explore the Economic Order Quantity (EOQ) model, a foundational concept in inventory management. You will find that it involves a similar type of cost trade-off analysis, this time balancing the cost of ordering inventory against the cost of holding it.