In This Issue:
- Customer Service Automation is Unique
Customer Service Automation is Unique
By Peter leppik
Automating customer service presents unique problems and opportunities. Unfortunately, many companies try to build customer service applications using the same mentality and processes they use for building web sites, enterprise applications, and other software with a graphical user interface (GUI). Both the customer service environment and the limitations of the voice user interface (VUI) are unique. Failing to work within these unique constraints often leads to failed projects, lost opportunities, and poor design.
Every Call Costs Money
One fact is unavoidable in customer service: every time a customer picks up the phone, it will cost the company money. It might cost a little, or it might cost a lot, but it will cost something. In exchange, the company has an opportunity to increase customer loyalty, sell more, and make a positive brand impression. Alternatively, the company might create ill-will, lose a customer, or even create an anti-evangelist who will actively tell friends and family about the bad experience.
The challenge is to minimize the cost while maximizing the positive impact of customer service. Automation is an enormously powerful tool for both these goals, but like a chainsaw or a forklift, it is a tool which needs to be used with care, since poorly designed automation can also cause great damage.
Bad Design is Very Expensive
If every call to customer service costs money, then it is easy to see how bad VUI design is expensive. Bad design will prevent some calls from being automated, annoy customers, lengthen calls, and can even be more expensive than having no automation at all.
We recommend that production automation systems be tested with the goal of finding most of the problems which will affect at least 1% of callers. Eliminating a 1% problem is extremely cost-effective, since the difference between automating a call and sending it to a live agent is often in the $5.00 to $10.00 per call range. 1% of a call center's total call volume, at a cost of several dollars each, quickly adds up to big money.
Contrast this to the cost of bad user interface design in the traditional software world, where design problems often have little direct financial impact, and software vendors therefore don't have much incentive to fix bad user interfaces.
Even a Little Automation Can Have Positive ROI
Unlike enterprise software installations or desktop applications, even a small automated application can pay for itself in the customer service environment very quickly.
Suppose that 5% of your calls are looking for some standard information--for example, details about a product announcement, store locations, or something else which can be handled by playing a recorded message. Simply providing callers with an option to listen to a recording can instantly cut your call volume, and pay for itself in a matter of days. As a bonus, customers often prefer to get routine information from a recording, and agents often prefer more intellectually challenging calls.
You can take advantage of this fact to design and build customer service applications incrementally, rather than as a single monolithic project. The result is more flexibility, less risk, and a better application.
Time is Of The Essence
In a customer service application, the caller is typically given only a few seconds to select each option or provide the information requested at each step. The caller is swept along by the VUI, without any opportunity to stop and study the options (and too often, no opportunity to back up either).
This is completely different from most other human-computer interfaces (other than video games), where people can generally pause, study the options, and make thoughtful choices about what might be most appropriate.
VUI designers need to work within the human limits of short-term memory and comprehension. This limits the number of choices you can give callers, and the amount of thought required at each step.
You (mostly) Can't Train Your Callers
In customer service applications, many--usually most--callers won't call often enough to become power users. This means that every element of the VUI needs to be completely intuitive.
But you still need to provide people with the ability to complete their tasks, and allow shortcuts for more frequent callers.
VUI testing must emphasize inexperienced callers. The value of a test participant plummets after completing a couple tasks, because he or she will have experience which typical callers do not.
Large-Scale Usability Testing is Realistic and Affordable
Testing a GUI can be expensive and time-consuming: often costing hundreds of dollars and an hour or more of lab time per participant. This limits the number of people who can be used to test an application.
In the world of customer service, however, the opposite is true. The most realistic test is one where each participant calls the application once or twice, from wherever he or she might normally call. This allows a highly automated data-collection process, and the cost per participant is in the range of ten to twenty dollars, making very large-scale tests affordable.
Large-scale testing ensures that even small design flaws--those impacting 1% of callers or fewer--can be discovered and corrected. It also allows VUI performance to be measured with a high degree of precision and statistical confidence.
Common Performance Metrics Can Be Misleading (or just plain wrong)
Measuring technical performance metrics like hold times and IVR containment is easy. Measuring customer satisfaction is relatively hard and contains many traps for the unwary.
It is very tempting to adopt technical performance metrics as proxies for customer satisfaction. The "logic" is straightforward: if hold times are getting shorter, then customers are happier. If calls remain "contained" in an automated application, then the company is saving money.
On the other hand, if hold times decrease because agents hang up on customers, then customer satisfaction goes down, not up. If IVR containment increases because options to connect to an agent have been removed or hidden, then callers are probably hanging up and calling back to talk to a live person.
Measuring customer service performance is not like measuring the performance of a database or web server. At some point, you have to get your hands dirty and actually ask callers what they think, and do it in a way which is meaningful.