Maximizing Efficiency and Customer Experience: A Case Study in Call Center Optimization

Introduction

For many businesses, the call center is a critical customer touchpoint that has a major impact on satisfaction, loyalty and ultimately revenue and profits. Customers today expect fast response times and helpful, knowledgeable service when they call. At the same time, call centers are a significant cost center, so businesses need to carefully manage staffing levels and agent productivity.

This presents a complex optimization problem – how can call centers minimize costs while still delivering a great customer experience? Understaffing leads to long wait times and frustrated customers, while overstaffing wastes money with agents sitting idle.

In this case study, we‘ll walk through how advanced analytics and simulation techniques can help call center managers optimally balance service levels and staffing. We‘ll build a model to predict the minimum number of agents needed to meet wait time targets, and show how to turn those results into an actionable staffing plan.

While every call center is different, the analytical approach demonstrated here can be customized to your business‘s unique circumstances and goals. We hope this case study gives you a practical framework to optimize your own call center operations.

The Business Problem

Let‘s consider a typical scenario. Imagine you manage a medium-sized call center for an ecommerce company. Your agents handle a variety of customer inquiries – everything from order status and returns to billing issues and product questions.

As call volumes have grown, you‘ve noticed average wait times and abandonment rates starting to climb. You know you likely need to hire additional agents, but it‘s not clear exactly how many. Overhiring will hurt profitability, but you can‘t risk delivering a poor experience to your customers.

Specifically, your goals are:

1. Determine the minimum number of agents needed to ensure no customer ever waits on hold (zero wait time)
2. Determine the minimum number of agents needed to keep the maximum customer wait time under 30 minutes

To meet these targets in a cost-effective way, you‘ll need a data-driven staffing plan that aligns agent schedules with predicted call arrival patterns. Rather than guessing or relying on trial and error, you can use historical call data to build a predictive model.

Data Overview

To build our staffing optimization model, we‘ll use a historical dataset of all calls received by the call center over a typical 24 hour period. This includes:

• Call ID
• Call arrival timestamp
• Call duration in minutes

Here‘s a sample of what the raw data looks like:

Call ID | Call Timestamp | Call Duration
1 | 2022-03-01 08:23:00 | 4.2
2 | 2022-03-01 08:25:00 | 6.1
3 | 2022-03-01 08:25:00 | 3.8

Before jumping into the modeling, it‘s always a good idea to explore the data visually to get a sense for the key patterns and relationships.

Let‘s start by looking at the distribution of call durations:

We can see that call lengths are typically between 3 to 10 minutes, with an average around 5-6 minutes. It will be important to consider this distribution of call durations, not just the average, when building our staffing model.

Next let‘s visualize the volume of calls by time of day:

There are clear peaks in call arrivals in the late morning and mid-afternoon, with volume dropping off substantially overnight. This intraday pattern is quite common for call centers serving consumers. Our staffing levels will need to adjust accordingly to meet service level targets during both peak and off-peak hours.

Simulation Modeling Approach

Now that we understand the basic patterns, we can build a model to predict the number of agents needed to hit our wait time goals. We‘ll take a simulation approach – essentially recreating the call center in software to analyze various scenarios.

The basic logic is:

For each 15-minute interval in a day (96 total):

1. Estimate call volume based on historical arrival rates
2. Determine number of agents available based on staff schedule
3. Assign incoming calls to available agents
4. If no agents are free, calls wait in a queue
5. Once an agent finishes a call, they take the next call from the queue
6. Calculate metrics like average wait time and % of calls answered within target

By running this simulation with different staffing schedules, we can zero in on the optimal number of agents needed to meet our wait time service level targets.

Here‘s some sample Python code demonstrating a basic version of the simulation logic:

def simulate_day(staff_schedule, call_arrivals):

  agents = []
  queue = []

  # 15-minute intervals in a day  
  for i in range(96): 

    # Calls arriving in this interval
    new_calls = call_arrivals[i]
    queue.extend(new_calls)

    # Agents finishing calls and available  
    for agent in agents:
      if agent.call_end_time == i:
        if queue:
          agent.take_call(queue.pop(0))
        else:
          agent.wait_for_call()

    # Assign new calls to any idle agents    
    for _ in range(staff_schedule[i]):
      if queue:
        agent = Agent()  
        agent.take_call(queue.pop(0))
        agents.append(agent)

  # Calculate final metrics
  wait_times = [call.wait_time for call in call_log]   
  service_level = sum(wait < SLA_TARGET for wait in wait_times) / len(wait_times)

  return {
    ‘avg_wait_time‘: sum(wait_times) / len(wait_times),
    ‘max_wait_time‘: max(wait_times),
    ‘service_level‘: 100.0 * service_level
  }

Of course, there are many additional details we could incorporate into our simulation to make it more realistic and precise – shift schedules, breaks, after-call work time, call abandonment, etc. But the core logic of matching incoming calls against available agents over time will be the same.

With this model in hand, we can test out different staffing scenarios and analyze the results to find the best solution to our optimization problem. We‘ll look to minimize total staffing costs while ensuring we‘re meeting service level targets.

Simulation Results

After running simulations across a wide range of staffing levels, we can visualize the key outputs:

This chart shows how average wait times decline as the number of agents increases. We can see that we‘d need around 50 agents to get average wait time near zero. However, the marginal impact of adding agents diminishes quickly after around 40.

We get a similar story when we look at the longest wait time any customer experiences:

To guarantee no wait over 30 minutes, we‘d need 47 agents. But even with 45 agents, the worst case wait is only about 35 minutes.

So based on these results, we can go back to the original business questions:

1. To ensure no customer ever waits on hold, we‘d need 50 agents
2. To keep maximum wait times under 30 minutes, we‘d need 47 agents

However, we need to consider the tradeoffs carefully. Going from 45 to 47 agents reduces the worst-case wait time from ~35 minutes to under 30 minutes, but it means paying for an extra 192 agent hours per day (2 agents x 24 hours x 4 shifts). That‘s a substantial cost for a relatively small improvement in service level. Staffing to a zero wait time target is likely prohibitively expensive.

This is where the simulation model is so valuable – it allows us to quantify the service level impact and costs of different staffing scenarios so we can make a well-informed decision aligned with broader business priorities. The right staffing level for one company or time period could be quite different than another.

Translating Model Results to a Staffing Plan

Once we‘ve determined the optimal overall staffing level, we still need to translate that into an actionable scheduling plan. We know we need 45-50 agents on hand, but how should they be distributed across the day? When should shifts start and end to align with call volumes?

To answer those questions, we can leverage the same data on call arrival patterns we visualized earlier. We know we‘ll need more agents on hand during the late morning and mid-afternoon peaks than we will overnight, for example.

A common approach is to break the day into shifts that align with agent availability and demand patterns, with some overlap between shifts to handle the busiest times. For example:

Shift 1: 8am – 4pm
Shift 2: 9am – 5pm
Shift 3: 11am – 7pm
Shift 4: 2pm – 10pm
Shift 5: 4pm – 12am
Shift 6: 7pm – 3am

The exact shift times and agent allocations would be optimized based on the intraday call volume predictions and simulation results. During lower-volume periods like the early morning, 1-2 agents may be sufficient to meet service level targets. But during the busiest times, all agents may need to be on hand.

Layered on top of the core shifts, call centers typically have flexible arrangements like part-time shifts, split shifts, and overtime hours that can be used to fine tune staffing levels as needed. Breaks also need to be staggered across agents to ensure continuous coverage.

Scheduling is a complex optimization problem in itself, but with solid forecasts of call volume and handle time patterns, a staffing plan can be developed to efficiently meet service level goals over time. The best call centers are continually analyzing their staffing and scheduling approach to identify opportunities for improvement.

Additional Call Center Optimization Levers

While staffing is typically the most impactful driver of call center performance, it‘s not the only one. As you‘re optimizing your staffing levels, it‘s worth also considering opportunities in a few other areas:

1. Routing and skills-based matching. Leverage skills-based routing rules to automatically direct certain call types to the agents best equipped to handle them. This can reduce handle times and increase first contact resolution.

2. Call deflection and self-service. Look for opportunities to resolve common, simple customer issues via self-service tools like knowledge bases, chatbots and IVRs. This reduces the volume of calls that need to be handled by agents.

3. Agent training and performance management. Identify the behaviors that separate top performing agents from the rest and build training programs to share those best practices across the team. Real-time call monitoring, call recording reviews and agent scorecards can reinforce the right behaviors.

4. Technology enablement. Equip agents with modern, integrated telephony and CRM systems that put customer information at their fingertips. AI-powered agent assist and robotic process automation (RPA) solutions can guide agents through calls and automate repetitive after-call work to save time.

The best call center optimization approach will ultimately depend on your unique business context and goals. But a combination of robust analytics, efficient staffing, comprehensive training, and smart technologies can enable consistently great customer service in a cost-effective way.

Conclusion

Managing a call center is a constant balancing act between delivering service quality and controlling labor costs. But with the right analytical approach, you can take the guesswork out of key decisions like how many agents to hire and when to schedule them.

By leveraging historical data to build a discrete event simulation model of your call center, you can quantify the impact of staffing changes on key metrics like wait times, service levels and costs. Visualizing the relationships between these variables enables you to align on the optimal solution with executives and other stakeholders.

Of course, simulation is just one piece of a comprehensive call center optimization strategy. Workforce management software can automate data analysis and scheduling tasks. Quality monitoring programs and agent scorecards can improve individual performance. And new AI-powered technologies can provide real-time tips and guidance to agents during customer conversations.

The dynamics of every call center are slightly different, so it‘s important to customize your optimization approach to your specific situation. Make sure you deeply understand the patterns in your customer interaction data before drawing conclusions or making policy changes.

Test new strategies on a small scale first before rolling them out broadly. And continually monitor your KPIs to track the impact and adjust your approach over time. Your customers and agents both stand to benefit immensely as you hone in on the most efficient ways to deliver great service at scale. We hope this case study has given you some ideas for getting started. Good luck optimizing your call center!