Posted by: Malcolm Jarvis
Picture the following situation. You’ve a query about a recent online purchase, and you call the company’s phone number which you eventually find on their website (it’s not as easy as it used to be, is it?). After choosing an option from the menu, you’re swiftly placed into a queue as “all of our agents are busy serving other customers just now”.
At this point, let’s take our imaginary contact centre experience in a few different directions and consider how they’d each alter our perception of this company’s customer service.
Scenario 1
After the “everybody’s busy right now” announcement, you go straight to hold music. Chances are it’ll be poorly chosen, but we’ll put that aside as I already dealt with that in an earlier blog post. Once every 30 seconds you’re asked to continue to hold as “your call is important”, but that’s your lot as far as queue information goes. After 10 minutes of holding, you’re put through to an agent.
Scenario 2
As you join the queue an automated announcement announces that the current queue time is 20 minutes. You decide to stay in the queue and hold and after only 10 minutes your call is answered.
Scenario 3
On joining the queue you’re given an automated announcement stating the queue time is 20 minutes. You need to go out in 15 minutes so you decide to hang up and call back another time.
Scenario 4
As in the previous two scenarios, you’re given an announcement that the queue time is 20 minutes. You hold, but 30 minutes later your call still hasn’t been answered and you hang up.
In terms of customer satisfaction, Scenario 2 clearly provides the best customer experience (under-promise and over-deliver - always a winner). Scenarios 1 and 3 are both OK depending on how important the call is and how pressed for time you are. While you’ve not actually managed to get through to anyone in Scenario 3, at least you’ve not wasted much time.
Scenario 4, however, is not only going to be remembered as a bad experience, it’s also likely to reduce the chances of you choosing this particular company’s services in the future. Such is the danger of over-promising followed by under-delivering.
So what do these scenarios tell us about how we can manage our inbound queues and, in particular, our estimated wait times, more effectively?
Queuing Blindfolded
When approaching a real-life queue, e.g. at an airport, supermarket or post office, you get lots of information about how long you’re likely to wait if you join the queue. You can clearly see how long the queue is, how many people are serving, how quickly people are getting served and so on. If your estimated queuing time is less than you’re prepared to wait then you’ll join the queue. If you guess that queuing will take too long then you leave. In queuing theory this is known as balking.
When we elect to join a queue but are then stuck in the queue longer than we’re prepared to wait, we eventually get fed up and leave the queue without being served. Leaving a queue after joining, but before getting served, is known as reneging.
We’ll be using these two terms a lot, so just to make sure we’re on the same page:
Balking: When a caller decides not to join a queue and hangs up the call immediately.
Reneging: When a caller joins a queue but hangs up before getting through to an agent.
Got it? Excellent!
From Scenario 4 above, it’s easy to see why reneging is so much worse than balking. People’s time is valuable. When we waste their time, we lose their confidence and, sooner or later, their custom.
When you’re placed into a phone system queue with no delay announcement, you’re effectively joining the queue blindfolded. You have no ability to gather your own information to help you decide whether to queue or to balk or, after queueing, to renege. Maybe you’ve joined a queue in a highly efficient 200 seat contact centre. Maybe you’re ringing a single lonely telephone in an empty office. All you know is what the phone system tells you.
Even when you do get a delay announcement, you still have no actual visibility of how many agents are taking calls, how many callers are ahead in the queue, or how long it usually takes to serve each caller. Each caller is still effectively blindfolded and they’re completely dependent on the content of announcements when deciding to queue or balk.
The question is: is it always in our best interest, and indeed our callers’ best interest, to give them wholly accurate information?
Closest Without Going Over
Do you remember the TV game show, The Price Is Right? The one where contestants guess the price tag of a range of consumer goods and then whoever got “closest without going over” won the toaster, camera, Aston Martin or whatever was on the show?
When setting up our inbound queue announcements, there are important decisions relating to estimating expected wait times that work the same way. Callers renege when the actual hold time exceeds the announced hold time, which is strongly perceived as bad customer service. We ideally want to give the most accurate expected delays, but we need to avoid people waiting longer than our estimates.
Humans, however, are awkward creatures, and there’s no way for a computer to precisely evaluate how long each caller will take once they’re connected to an agent. Calls may be a complaint that will take an hour to resolve or simply a quick question that could have been answered in 20 seconds using the company’s website. You can head them off at your IVR and funnel calls more strategically, but this is only effective if your callers are playing along.
Ultimately the best a contact centre system can do is estimate likely wait times based on average serving times for calls to that queue. Larger inbound teams that receive more calls make this calculation more reliable as call durations average out better across larger volumes of calls, but smaller teams aren’t so lucky. In order to account for the potential for long calls to throw out the system’s estimated waiting times, it’s therefore a good idea to make the estimates more cautious to give the system a bit of leeway.
Let’s go back and look at our scenarios from earlier and think about what options this gives us in our own contact centre environments.
In Scenario 2, where you were told your wait would be 20 minutes but you got through to an agent in half that time, we can see how a contact centre’s happiest customers are those who have had their expectations set and then either met or exceeded. This suggests we’re better off erring on the side of caution and setting our queuing system to generously pad our estimates of how long we think hold times will be, that is, to make our estimates more cautious.
However, in Scenario 3, when you were told that the hold time would be 20 minutes and then hung up in response, we can see how making wait time estimates too cautious increases balking.
In fact, sometimes it may be better to give no delay information at all. In Scenario 3, with no delay announcement, instead of balking you’d probably have waited and your call would have been answered after 10 minutes. In Scenario 4, where you ended up waiting for half an hour, with no delay announcement at least you couldn’t complain you were misled.
This suggests we can alter customer behaviour by adjusting how cautious or optimistic our expected wait time announcements are. Happily, there’s a research paper that gives evidence to suggest exactly how this might be done.
Are We All Paying Attention?
In the research paper Call Centers with Delay Information: Models and Insights, the authors Jouini et al, go to impressive lengths to model various aspects of queuing behaviour relating to expected waiting time announcements. For example, they model and examine ways in which changing how cautious or optimistic our expected waiting times are can be used to keep the proportion of customers reneging below 3%.
This was all done mathematically, which given the range of factors considered, was impressive stuff. Here’s a brief summary of what I took from it:
1. Optimistic expected waiting times can be bad for customer service
Providing optimistic expected wait times will encourage more callers to join a queue. Conversely, to discourage callers from queuing we can make our estimated waiting times more cautious (i.e. longer) to encourage callers to balk.
Naturally, we want to serve as many callers as possible, but as people will renege if their hold time much exceeds the expected waiting time they were given, we can use more cautious wait times during busy periods as an effective means of reducing reneging at the cost of balking. This also causes fewer non-urgent callers to join the queue, thus reducing queuing times for urgent callers. This may not be ideal, but during busy times it’s better for your brand to have more callers balking than reneging as a result of unachievable expected waiting times.
2. Using more cautious expected wait times to reduce reneging eventually results in more balking
As a queue hits a busy spell, where the number of new entrants to the queue outstrips your agents’ ability to serve customers, you can use more cautious wait time estimates to encourage customers to join the queue without consequently increasing the proportion of customers reneging.
However, this only works up to a point as once your delay estimates become too cautious the benefits will sharply start to reduce. Not because customers are reneging, but because more will start balking when told how long they’re likely to have to wait.
The graph above is an excerpt from the graph on page 22 of Jouini et al's research paper. This is based on a 10 agent system which is doing everything it can to keep the rate of callers reneging below 3%. The graph illustrates how making our estimated wait times more accurate (by making them more cautious) will work to a point, in this case around 50% more callers than the system is designed to handle. After this point, cautious wait times become less necessary as more and more callers will simply balk on hearing the normal expected wait times.
As increased balking is preferable to telling customers an expected wait time that is less than their actual wait time, this can be an effective and acceptable compromise.
3. Where customers are not patient, delay announcements need to be extra cautious
If callers are typically impatient, in that they're unlikely to be tolerant of your estimated wait times being wrong, you need to provide the most cautious estimates. Patience in this case relates to how important it is that callers get to talk to an agent. An individual may not normally be very patient, but if they’re calling to register the winning lottery ticket you can bet they’ll hold on all day and all night if they have to. This would make them a very patient customer, if only for that one call.
The research estimates that when customers are least patient, you need to get your delay announcements correct more than 90% of the time in order to prevent more than 3% of customers reneging regardless of the size of the team. Understandably this means that your balk rate will be highest in this situation, but at least you’re not creating unnecessarily dissatisfied customers by providing them with estimated wait times you can’t stick to.
On the other hand, if you’re running a contact centre for an essential service that people call when they absolutely need to speak with one of your agents, such as our lottery ticket hotline, you can be a lot more optimistic with your estimates. This reduces your balk rate while taking advantage of your callers' patience to keep your reneging rate down too.
The graph below shows this relationship alongside the effect of larger system sizes.
4. Larger inbound teams allow more optimistic wait times than small teams
Regardless of how patient customers are prepared to be, in larger systems where there are more agents dealing with more customers, outcomes are consistently better than in smaller systems where there are less agents dealing with a smaller number of customers.
This graph, taken from the information on page 19 of Jouini et al's research, shows the benefits of larger systems (more agents available to handle more calls), and more patient customers. The top line, showing the most impatient customers, is the one example where larger system sizes make little difference - your wait time estimates need to be accurate (i.e. cautious) to prevent this type of caller reneging no matter how big the system is.
With more patient callers you can maintain reneging levels with more optimistic expected wait times, which gets far easier as the system size increases. This is because in large systems the small number of customers with requests that are abnormally difficult and time-consuming don’t prevent the majority of agents carrying on as normal. If there are only a handful of agents taking calls it doesn’t take many awkward customers to cause the queue to grind to a standstill while their awkward requests are being dealt with.
5. Small systems cannot keep reneging down without a delay announcement
The authors of the research also noted that without a delay announcement, it’s impossible to keep reneging below 3% in smaller systems (in their example, the cut-off was somewhere between 20 and 50 agents serving the queue). This is because only a small number of customers balk when entering a queue with no announcement (in the model, it’s estimated at 5%), but very quickly the customers that would have balked had you told them how long the wait was going to be become impatient and start to renege. This means that having delay announcements is far more important in small queuing systems than large ones.
So, what we’ve learned is that it’s far better that customers balk having been informed of an expected wait time you can stick to, rather than having them renege after being misinformed. We can also use more cautious estimated wait time announcements to discourage callers from joining the queue during busy periods. This may mean they need to call back later, but at least they’re not sitting on hold for a long time before finally reneging in a bad mood.
For small inbound teams, providing cautious expected wait times is much more important than for large teams. This means that wait time announcements in smaller teams (around five agents or fewer) should be highly pessimistic, doubly so if customers are likely to be impatient while queuing.
Of course, contact centres these days have many tools at their disposal to help customers make informed queuing decisions. One of the suggestions in Jouini et al’s research is to give customers a minimum and maximum wait time as opposed to a single duration, e.g. “queue times are currently between 10 and 20 minutes”, in order to achieve something approximating the best of both worlds.
Contact centres also have the option of self-service call backs. These have many benefits including balancing agent activity across busier and quieter inbound call periods. This also allows urgent callers to advance more quickly up the queue while those not willing to wait will be called back once things have quietened down a bit.
One industry trend that is becoming increasingly popular is to simply do as much as possible to keep people as far away from the contact centre queues as possible. This includes approaches such as self-service troubleshooting guides, frequently asked questions (FAQs) online or as part of the inbound calling system, or even just creating better products and services that don’t generate as many calls to your contact centre in the first place.
Or, if you’re not too bothered about keeping customers happy, you could simply hide the company telephone number deep within the company website where only the most determined callers will ever find it. Job done!