OR/MS Today, August 1997
Ringing Up Big Business
By Vijay Mehrotra
The Call Center industry is growing so fast that not even the experts can say for sure how big it is. A look inside this ubiquitous, dynamic, yet little-known industry reveals an arena rich with opportunities for quantitative methods and OR/MS professionals needed to solve its complex organizational problems.
I remember my first call center. I was a 17-year-old college freshman desperate to make a little money to keep up my pinball, Pac Man and textbook habits. The sign on the board said, "Phone Reps Wanted." I answered the call. A few days later I found myself in a crowded room with 20 others, on the phone with strangers for hours on end.
Little did I know that much of my professional career would be just one call center after another.
What is a call center? My definition: "Any group whose principal business is talking on the telephone to customers or prospects." This group may be centralized in a single site, distributed across multiple sites or "virtualized" with agents in individual offices. Typically, agents in a call center share a common automatic call distribution (ACD) telephone switch and a common set of PC and desktop resources.
"Everybody Has a Call Center"
You've probably thought about call centers, at least a few times. You may have scowled at the electronic voice at your favorite airline that informs you that, "All of our agents are busy. Please stay on the line, and your call will be answered in the order received," or groaned at the agent who wasn't fast enough. You might have wondered about the woman on the phone at 3 a.m. at your bank, thankful that she could verify that you do have overdraft protection. You may have called around until you found the cheapest rental car or hotel room. Ever call to activate your Visa card? Did you buy all of last year's Christmas presents through a mail-order catalog over the phone on Dec. 23? Or call FedEx or UPS to check on your shipment on Dec. 24?
You get my point. This call center business is enormous, ubiquitous and growing fast. In fact, like the man who was so rich he didn't know how much money he had, even industry insiders don't have a clear sense of how big this thing actually is.
A few numbers to get your eyes popping. The number of call centers in this country is estimated at anywhere from 20,000 [Dawson] to 350,000 [AT&T], and employs anywhere from 4 million [Perkins] to 6.5 million people [AT&T]. Telephone agents at the top 50 call center "service agencies" alone spent 1.3 billion minutes on the phone with customers in 1996 [TCCS]. Annual expenditures are estimated to be between $100 billion and $300 billion, with anywhere from 50-75 percent on labor.
(Note: The numbers listed above are domestic estimates, but call centers are an international business. European unity, business growth in East Asia and an increasingly global economy are all helping drive the call center industry's expansion abroad.)
Call centers are also becoming increasingly complicated to run. This complexity, I believe, is driven largely by a business phenomenon known as "Mass Customization" [Pine, Davis] which results in more and more specialized products and services being created, developed and sold to increasingly value-focused customers.
This has a significant impact on call centers. For example, where service calls were once received on a standard line and queued on a FIFO basis, now there may be a variety of service contracts that determine when, by whom, how quickly and at what cost each request is handled. Where once a customer calling to order a widget got a standard price for a standard item, now that customer receives a price based on widget preferences, purchase history, advertising reference and the bundle of products being purchased.
Finally, as competition between firms grows increasingly fierce, senior executives are beginning to wake up to the impact of their call center on the customers' perception of quality. In addition, as customer transactions take place over the phone, firms are beginning to realize that the call center is their most direct line to the customer. As a result, sales and service call centers &emdash; once an organizational backwater &emdash; are today more commonly viewed as a strategic asset.
Call centers. A big, growing, complicated and increasingly important part of the business landscape. Lots of data being gathered in telephone switches and call tracking databases. Fertile ground for OR/MS.
Call Forecasting
To run its business, there are some fundamental questions that every inbound center must address:
- How many calls will we get?
- How many people do we need on staff?
- When/how should these agents be hired, trained and scheduled?
- What will this cost?
However, call center management has historically been an "up-through-the-ranks" world that, with some exceptions, has neither possessed nor sought the skills needed to tackle these core analysis problems. A widespread perception has been that industry experience is more relevant than any particular training. As a result, forecasting and scheduling have in many cases been handled through best guesses, back-of-the-envelope calculations or black box software packages.
Have you ever called to make an airline reservation, been put on hold, waited a minute and then hung up? In the call center parlance, this is known as "abandonment." Along with traditional forecasting issues such as data availability, data integrity, seasonality and non-stationary randomness, abandonment makes call forecasting more challenging than simply fitting a regression model to historical call volumes, particularly since there is usually no way to tell if an abandoned call led to a call back later on.
Traditionally, time series models have been the most popular approach to forecasting telephone calls, predicting calls based on historical data while taking into account day-of-week and month-of-year factors. In addition, other forecasting models have also been successfully applied in some call center environments. For example, sales centers may forecast calls based on shipments or advertisements to customer segments; software technical support call centers may use forecasting models based on new sales and product upgrades; and reservation centers may utilize forecasting models that reflect the impact of increasingly common "price wars."
In reality, call forecasting is probably equal parts art and science, and, therefore, in order to build the right forecasting model, it is important for an OR/MS professional to ask a lot of questions, to listen carefully and to build a good understanding of the business of the call center being modeled. My experience is that this interaction is important not only in enabling you to build the right model, but also in building credibility with the call center staff that will be using the forecasts, and for understanding very clearly how these forecasts are being used in the business operations. Often, your experience and clear thinking will provide tremendous value in helping them figure out how to use forecasts.
Agent Scheduling
For call center managers, a key performance metric is "service level." A call center's service level is defined as the percentage of customers who wait less than some target (for example, 30 seconds) before reaching an agent.
From queueing theory, we know that service level is a function of several variables, including call arrivals, call handling time, and the number of agents on staff. Call center managers, faced with tight budget constraints and specific service level targets, wrestle constantly with the trade-off between more agents (more costs) and lower service levels (lower customer satisfaction).
Like call forecasting, agent scheduling has traditionally not been done in the most management-scientific manner. Prior to the PC revolution, agent scheduling was an extremely labor-intensive task that produced a decidedly non-optimal result.
However, over the last 10-15 years, a number of commercially-available software packages have been introduced that try to schedule agents cost-effectively. The general logic underlying these tools is:
- Call forecasts. Determine call forecasts for a particular time period (typically a day, a month or a week).
- Call patterns. Treating this time period as a set of "time blocks" (typically 30 minutes long), use historical call arrival patterns to distribute the forecasted calls across these individual time blocks.
- Required number of agents in seats. For each time block, based on the number of forecasted calls and some forecasted call handling time (usually based on historical data), use Erlang tables to determine the number of agents needed on duty to meet the service level target for that time block.
- Assign agents to schedules. Based on operational constraints such as available shifts (including lunches, breaks and meetings), agent preferences and agent costs, use some type of algorithm, in some cases a linear program, to assign specific agents to specific schedules.
In addition, these tools are often used to support medium- and long-term call center planning. However, these tools generate solutions that depend on key underlying assumptions:
- Every call is of the same type.
- Every agent can handle calls equally fast.
- Calls are queued on a first-in-first-out basis.
- Call abandonment rates are known and independent of the time a customer spends waiting.
Additionally, the actual abandonment rate in a call center cannot be assumed as an input, even from historical averages, for the abandonment will be heavily influenced by the amount of time that customers spend waiting, which is, in turn, driven by the scheduling and skill of agents.
Despite these shortcomings, the workforce management systems described above are still considered the state of the art for agent scheduling, often accompanied by manual steps to modify schedules. While leaving something to be desired, these workforce management systems remain a superior alternative to the traditional pencil-and-paper methods.
However, when planning for the future, call center managers are relying increasingly on simulation models as the source of answers to key "what-if" questions, and as the right way to design and/or modify different aspects of their call centers.
Call Center Simulation
As discussed earlier, today's call centers are increasingly sophisticated systems. Over the last 10 years there have been major advances in how calls can be routed and queued to different groups of agents and how different agents can be "skilled" to handle different calls with different priorities.
Along with the dynamics of call abandonment, this increased queueing complexity creates havoc for traditional FCFS queueing models. As these complicating factors become more and more common in the industry, simulation models are becoming more and more prevalent as management tools for analyzing call center behavior.
In fact, over the last few years, commercially available simulation packages (such as Call$im from Systems Modeling Corporation and CallLab from Bard Technology) have been developed specifically for call centers, enabling managers and analysts without extensive mathematical or programming skills to quickly build and run call center models.
How are simulation models used in call centers? Most typically for answering planning questions in the early stages, typically three months to a year prior to implementation. For example, simulation has been used to tackle the following types of issues:
- "I've got my staffing budget for the next fiscal year, but I don't know how many people I need to make service levels, what shifts to hire for or what skills to train my workers on."
- "Service levels look pretty good right now, but our peak season is coming up. What I don't know is how badly our speed of answer and abandonment rates will suffer if we staff as planned and our call forecasts turn out to be too low."
- "Marketing has come up with a new program giving our 'preferred customers' a special priority when they call us with questions. What I'm worried about is how this new program will effect the waiting times that the rest of our customers experience."
Call Content Analysis
While forecasting, scheduling and simulation are classical OR/MS applications, call content analysis is part of a new breed of problems for which OR/MS professionals are well-equipped. What these problems require is a combination of mathematical modeling skills, database aptitude and organizational/business process savvy.
A phone call is a two-way transaction. In conversation, the customer and the agent both give and receive information. In almost all call centers, some data about each phone call is tracked by the agent, including who the customer is and what was discussed and resolved in the phone call. Additional information, such as which agent handled the call and how long the call was, is tracked automatically.
