You Can’t Afford to Please Everyone
By applying the tools of probability, smart businesses can serve the right customers in the right ways.
Topics
Frontiers
The internet enables companies to serve customers faster than ever, but it has also made people more impatient, often expecting an instant response. With services representing an important sector of the U.S. and other Western economies, managing those demands and enhancing quality and efficiency will be an increasingly critical component to how businesses compete. But sometimes, the best way to handle ever-higher customer expectations is to under-deliver in some areas or to certain segments so you can be right on target where it matters most. That’s what Amy R. Ward, the Rothman Family Professor of Operations Management at the University of Chicago’s Booth School of Business, has found in her research on operational efficiency.
As a graduate student, Ward became intrigued by queueing — the methods companies use to manage customer wait time — and she went on to explore the topic in her doctoral thesis. Her interest in service systems has led her to examine how customer and employee characteristics can play into decisions about staffing and capacity. Although giving customers what they want — and as rapidly as possible — is certainly a worthy goal for service organizations, Ward notes that businesses can’t always afford to do this. She argues that companies can use probability to understand how best to align resources with customer demand.
MIT Sloan Management Review contributing correspondent Frieda Klotz recently spoke with Ward about her work. What follows is an edited and condensed version of their conversation.
MIT Sloan Management Review: How did you become interested in applied probability as a tool for managers?
Ward: As an undergraduate, I was influenced by two professors. One was using probability and statistics to do consulting work for Lockheed Martin, the aerospace defense company, on missiles. The other did work for pharmaceutical companies, assessing the shelf life of active ingredients in drugs in clinical trials. I was fascinated by how each of them relied on similar mathematical methodology to solve very different, real problems.
Every company has customers and relies on inputs, either raw materials or something closer to a finished good. Operations focuses on how to manage processes efficiently in terms of employees, cost, and quality. In real life, though, nothing goes according to plan. There’s randomness — and understanding probability can help companies make decisions in uncertain environments.
What are some of the biggest challenges organizations face in meeting the needs of customers?
Ward: Demand management is something many different types of businesses need to deal with. Call center managers need to figure out how many people to employ to get to the on-hold customers quickly enough (or, alternatively, to reduce the number of people they put on hold in the first place). In retail, when somebody walks into the store you want them to be able to buy a product, and you want to have a staff member who can serve them. You may want to cut down on the staffing costs, but scaling back staffing affects customer wait times. I like to think about what happens at the margins: If you have five versus six cashiers, how does that affect customer wait times? If it’s going to increase the wait by two minutes, it might not be a big problem. But if it’s going to increase the wait by 20 minutes, that’s a different story.
Over the past 50 years or so, service systems like call centers and their modern counterparts have become a big part of the U.S. economy, arguably even more important than manufacturing. In production systems where demand varies, you can often smooth things out and improve efficiency by building up finished-goods inventory. But you can’t do that with service systems — you have to serve customers whenever they come.
In one of your recent papers, you suggest that it sometimes makes sense to underinvest in capacity at a call center when things are busy because that will “trim” excess congestion and remove excess customers. This seems like an extreme approach.
Ward: I agree it might sound horrible, but look at it this way: If the cost of capacity — mostly, the staffing cost — is extremely high, then you probably won’t be able to invest in enough capacity to serve all customers quickly. This means some customers will leave. We want to provide an analytical framework to help managers decide when the cost of capacity is too high. If the cost of capacity is high compared with the cost of losing a customer due to excessive waiting, then investing in capacity doesn’t make sense. However, underinvesting in capacity — purposefully losing customers unwilling to wait to reduce excess congestion — can make sense.
I realize that this makes it sound like customers aren’t valued. There’s a nicer way to put it. Maybe it’s a service that offers discounted tickets for Broadway shows, and they’re saying that it’s OK for people to wait. It’s even OK for people to have to wait so long that some people won’t use the service at all. The business needs to focus on what it wants to provide. It can then reduce or minimize other service costs.
So, by investing in customer service, organizations might be putting money in the wrong place?
Ward: That’s right. To express it a little differently, delivering excellent service is difficult and expensive, and people really appreciate it. But a lot of organizations don’t do it very well. To deliver quality service, you have to be very clear about your priorities: Focus on doing important things well, and don’t promise more. For example, if you deliver cheap discounted tickets, you should be clear with people that they will have to wait in line. Rather than investing in everything, focus on areas that improve the part of the service that’s most valuable to your customers. Ideally, that will improve things for everybody.
How often do companies find the sweet spot?
Ward: Let me turn this question back to you: Think about the last time you encountered poor service. Delivering excellent service is very easy when you don’t have to worry about resources. But most organizations need to make choices about the number of people they employ and how much equipment, and this requires having very clear priorities.
One company that does this quite well is Southwest Airlines. Its turnaround times — the time it takes for an arriving aircraft to get ready for the next flight — are among the fastest in the industry. How do they do it? For one thing, they don’t have first-class or business class seating — there’s a philosophy that customers are equal. So when they load the plane, it’s quick. Also, they mainly use one type of aircraft, which means that their pilots, copilots, and flight crew can pretty much fly any plane.
Is it good or bad service? Some people might not like the fact that there’s no first class. But they’re delivering very high-quality service to a particular market segment. For any business, the key is to make sure that your operational capacity is in line with the segment of customers you’re trying to serve — and what’s important to them.
You’ve studied customer wait times and different ways of prioritizing customers based on factors such as employee costs and a business’s optimal operating regime. Are there other sectors where the idea of waiting is key?
Ward: Wait times continue to be an important area of research for many organizations. In settings like hospital emergency departments, the wait time can be a matter of life or death. In the past 10 or 20 years, [emergency] departments across the country have been getting more and more crowded and wait times have been increasing. Studies by other researchers have correlated wait times with deaths. And even if waiting doesn’t lead to death, it can be dangerous and costly: If the patient is seen within five minutes, treatment can be easy, but if it takes a couple of hours and the patient’s condition worsens, it can result in a hospital stay.
Research on this topic can help hospital managers think about patient prioritization and wait times. If you’re a lower-risk patient and there are X doctors in the emergency room, are you waiting one hour or two? Is the variability such that people typically wait two hours but sometimes as long as six hours? Hospitals already have guidelines, of course, but many want to use more analytical tools too.
Wait times and prioritization can be very important in manufacturing as well. Whose request do we work on next? What lead times do we promise customers? As with hospitals, the decisions can have a first-order impact on how long different customers wait to receive their goods. The customers whose purchases get top priority have a short wait. Others might have to wait weeks or months.
Email Updates on the Future of Work
Monthly research-based updates on what the future of work means for your workplace, teams, and culture.
Please enter a valid email address
Thank you for signing up
Looking down the road, where do you see this sort of applied probability research going? Where does machine learning fit in?
Ward: Waiting formulas are extremely helpful at predicting how average wait times vary with capacity. Every MBA student who has taken an operations course knows that wait times increase linearly when capacity exceeds demand, and they increase exponentially as demand gets close to capacity. But providing customers with an exact wait time is extremely difficult given all the system dynamics. There may be an opportunity to combine the predictive power of machine and deep learning with the explanatory power of analytic formulas.
Take ride sharing, which is an exciting and evolving area. If you really want to get a car to a customer in three minutes, you need to have enough capacity (i.e., enough cars) to make a rough-cut estimate using a basic analytic formula. But where should the cars be, and how do you match them with riders? On the surface, it might make sense for the company to assign the driver who’s closest to the rider. But that won’t always be great for the company. It may work better to skip over the driver closest to the rider and assign someone who is four minutes away instead of three.
Ideally, you want to be able to plan ahead — to anticipate future demand. That means being able to assign drivers not just in response to what you’re seeing now, but also in response to the demand that might materialize due to future events like bad weather or people leaving a Beyoncé concert. Where the drivers are located spatially affects the system’s efficiency. You may not be able to tell the drivers exactly where to go, but you can certainly nudge them in certain directions. It may mean that a few passengers have to wait a few minutes longer, but as a coauthor and I showed in a recent paper, the delay can create flexibility for later on. The best solutions will likely combine both traditional analysis and machine learning.