Lessons from the Supermarket Aisle
At most of the large suburban supermarkets I’ve shopped at, there’s a small army of check-out cashiers. Long lines can scare away customers, so supermarkets use lots of cashiers—as it turns out, overstaffing— so that individual lines won’t appear crowded.
When Whole Foods came to New York City, high-density apartment buildings and expensive real-estate posed a problem: how do you staff a space-constrained store to handle the hordes of Manhattan gourmets? Going against conventional wisdom, this popular organic grocer adopted bank-style queuing--a single line and a dispatcher that sent customers to the next free cashier. Sound familiar? It’s the call center model where the supermarket dispatcher is the ACD, the cashiers are agents, and waiting customers with their shopping carts are callers.
In fact, the science part of queuing—the classic Erlang C model-- tells us that Whole Food’s “ACD” approach achieves lower average waiting time with the same number of agents. Intuition behind the science: with a single line, the omniscient dispatcher always insures that cashiers are never idle, but when you pick the wrong line at a multi-line supermarket, you invariably end up watching a cashier open up…at the other end of the store. More advanced ACDs go one better than the supermarket dispatcher by predicting waiting times and then routing those callers who’ll have excessive waits to special reserve agents.
There’s more to customer experience than low average speed of answer. Psychologists tells us that occupied time feels shorter than unoccupied time. Translation: time flies when your mind is active. My solution to reducing perceived waiting time on a slow-moving supermarket line is to reach over and grab one of the conveniently located cooking magazines. For news starved customers, several chains have even added kiosks running CNN feeds. Keeping waiting customers mentally involved improves customer experience with no extra staffing.
Back at the call center, music-on-hold helps to engage waiting callers while location information—“you’re number three in the queue”—removes wait uncertainty. There’s research that shows that customer who are given an expected wait time will remain on the line 25% longer. Makes sense: when you know service is in progress, you’re more likely to stay put.
On the horizon, blended reality and IVVR technology has the potential to make waiting on your iPhone for an agent more like watching previews at the movies. Pushed music, video, and other content will keep us entertained and better informed, and virtual interactions with other waiting customers will add the human element. With all this stimulation, it’s just possible that we may look forward to being number ten in the ACD queue.
Posted by Andy Green at 18:13 on January 17, 2008
ivar said..
Posted at 20:54 on April 21, 2008
You make an excellent point about the Wholefoods experience and I agree with you that when you know that things are progressing you are likely to be more patient. It's like calling customer service and being told you are the 10th person in the queue or your estimated wait time is 5 minutes, instead of just putting you on hold with annoying music playing in the background! I wish all customer support lines would implement this. Delayed certainty is better than short uncertainty!
Andy Green said..
Posted at 13:35 on April 24, 2008
As humans, we're definitely hard-wired to crave certainty. Since it's impossible to remove variability from the system--or more accurately, expensive to limit it--customer support just does the next best thing: gives you good about information about your expected wait. Delayed certainty with some feedback is a nice compromise.
Andy Green said..
Posted at 09:46 on February 05, 2008
The Erlang model and the Erlang C calculator mentioned in my post are a good starting point to get a feel for call center staffing. Of course, there’s a lot more to customer experience than average speed of answer, and the Erlang model is an approximation of reality, but you can get some nice insights even from the basics.
The single queue versus multi-queue decision is easy to resolve in the Erlang world. You have two basic inputs, an arrival rate of customers, and a processing rate –the rate at which customer are serviced.
Suppose for our supermarket example, we have 20 customer arriving per minute (on average) and an average handling rate of 5 per minute—that works out to 12 seconds handling or service time. And for the sake of argument, we have 5 cashiers :-). So, plugging our numbers into the calculator, we get an average speed of answer or average waiting time on the queue of only 6.6 seconds. Not bad!
That’s the single queue result. What about having separate multiple queues? The service rate or handling time hasn’t changed, so that stays at 12 seconds. But we adjust the number of agents to one, since we’re looking to learn the waiting time on an individual queue. So far, so good. The trick is that we now use a new arrival rate. Since the customers are picking cashier lines randomly, we can assume that an individual cashier sees 4 customers per minute – 20 divided by the number of cashiers(5).
Plugging our numbers in, we get 48 seconds average waiting time on the queue. The single queue model does far better. In fact, it beats the waiting time of the multiple queues by more than a factor of 5! In other words, consolidating 5 separate shopping lines into 1 gives more than a 5 fold reduction in waiting time! An incredible improvement, and maybe not what you’d expect.