Vocalabs Newsletter: Quality Times

Issue
6

In This Issue:

  • Common Problems in Self-Service

Common Problems in Self-Service

By Peter Leppik
pleppik@vocalabs.com

In our experience, callers prefer to use self-service systems--when they work well. When they don't, however, callers tend to hate them.

There are several problems with automated self-service systems, both speech-recognition based and tone based, which we see over and over again. Some are more subtle than others, but all cost real dollars, through more calls handled by live agents, increased repeat calls, and reduced customer loyalty.

At their heart, most of these problems are caused by insufficient testing and/or poor design. Often, these creep into an application over time, through the accumulation of small, untested changes (we call this "application drift"). Fortunately, many problems can be found and fixed relatively easily, especially if proper testing is done during the design phase.

Here are some of the most common--and easiest to fix--problems we see with self-service systems:

1) Poorly Worded Menus and Messages

The problem we see is instructions which confuse callers, don't provide clear choices, or don't match with the reason for the call. It can be caused by using industry- or company-specific jargon, or messages which make sense from an insider's perspective but not a caller's.

A poorly-worded message can easily confuse 5% of callers, and we've seen messages which confuse over 90% of callers.

Impact:

Any caller who can't figure out how to use a self-service system is almost certain to go to a live agent, so this is an expensive problem. Even subtle problems can confuse 5% or more of the callers, and we've seen messages which confuse over 90% of callers.
Symptoms:

* "C" or "D" grade for call completion in a VocaLabs survey
* Callers often report choosing the wrong option
* Fewer than expected calls complete in the self-service system
* Improperly routed calls

Prescription:

  1. Using the Interactive Drill-Down in a VocaLabs survey, select panelists who could not complete a troublesome task.
  2. Listen to each individual call selected in step 1, and note where the panelist first made an incorrect choice, and which option was chosen.
  3. Re-record problematic options to be less confusing.
  4. Always include the option to go back to the previous menu.

ROI Impact:

Fixing these problems provides an almost immediate payback, since the only cost (beyond the testing itself) is re-recording some of the messages. Even small changes can yield several percentage points of improvement in calls handled through self-service.

2) Too Many Menu Choices

Especially in tone-based systems, providing too many menu choices can easily confuse callers, as they can forget which option they wanted, or select an option near the beginning of the menu before listening to all the choices. This tends to be less of a problem with speech-recognition based systems, since words or phrases are easier to remember than numerical choices.

Speech-recognition applications tend to have less of a problems with large menus.

Impact:

Calls will tend to be too long, and misrouted calls common, as callers choose the wrong option, or have to listen to the menu multiple times.

Symptoms:

* "C" or "D" grade for call completion in a VocaLabs survey
* Callers often report choosing the wrong option
* Many callers report they were given too many choices
* Callers often listen to the same menu multiple times
* Improperly routed calls

Prescription:

  1. Using the Interactive Drill-Down in a VocaLabs survey, select panelists who reported having too many choices.
  2. Listen to each individual call selected in step 1, and note which menus were repeated, or where callers chose incorrect options
  3. Combine similar menu options, and consider eliminating seldom-used choices.

Alternatively:

Speech-recognition applications tend to have less of a problem with large menus, so consider a move from tone-based to speech-based self-service. Application testing is extremely important for a successful speech deployment, so be sure the new application is properly tested before launch.
ROI Impact:

Fixing this problem will, at a minimum, require some application re-design, so it tends to be something best fixed before an application is live. A more streamlined self-service system is much more likely to be used by callers, so finding and fixing these problems can pay for itself quickly through increased usage of self-service.

3) Inconsistency

Self-service systems should work the same way everywhere, to be easy for both novice and experienced callers. For example, if the caller needs to end a string of digits with # in one place, it should not be * somewhere else. Inconsistency is often a design issue, and can occur when callers are switched from one self-service system to another. Fortunately, simple changes can often go a long way towards correcting this problem.

Inconsistency can be an expensive problem, as callers tend to abandon self-service when it doesn't work as expected.

Impact:

 

Like poorly worded messages, inconsistency can be an expensive problem. Callers will tend to abandon the self-service system when it doesn't work the way they expected it to work, or when they get confused.

Symptoms:

* "C" or "D" grade for call completion in a VocaLabs survey
* Frequent invalid account numbers, PINs, etc.
* Callers attempt to choose nonexistent options
* High error rates or out-of-grammar responses in speech recognition systems
* Improperly routed calls

Prescription:

  1. Using the Interactive Drill-Down in a VocaLabs survey, select panelists who reported the system was difficult to use.
  2. Listen to each individual call and browse free responses, and note where the system required different input for similar tasks, leading to caller mistakes.
  3. Choose a single, consistent method for data input and global options, and change any parts of the application which don't conform.

ROI Impact:

It may not be possible to correct all inconsistencies without a major redesign, but many problems can be fixed relatively easily with simple reprogramming and updating prompts. For these simple fixes, the ROI will be almost immediate, as the number of callers willing and able to use the self-service application improves.

4) Difficult or Impossible to Get Live Help

In our experience, we've found that callers prefer to use a well-designed self-service system over a live agent. Even the best self-service system can only do so much, and there will come a time when a caller needs to talk to a real person. Putting roadblocks in front of the live agents only serves to increase the frustration of an already-frustrated customer, and generally doesn't prevent them from getting to an agent anyway.

Hindering access to live help doesn't keep callers from reaching an agent--but it does make them more upset.

Impact:

 

Limiting access to live agents is often done as a cost-saving measure, but the real cost is in lost business because of frustrated and angry customers.
Symptoms:

* "C" or "D" grade for caller satisfaction in a VocaLabs survey
* Frequent attempts on common "operator" options, such as dialing "0" or saying "help"
* High error rates or out-of-grammar responses in speech recognition systems
* Improperly routed calls

Prescription:

  1. Determine an appropriate universal "operator" option, and implement everywhere.

ROI Impact:

The payback for including universal options for live assistance is measured in retained business and improved customer loyalty. As long as the self-service application is well-designed, most of the additional callers going to live agents will be people who would have otherwise given up (and quite likely have taken their business elsewhere). The exact impact will depend on the value of a customer to your particular business, and how much ability your customers have to go to a competitor.

5) Poorly Tuned Speech Recognition

Speech recognition is remarkable technology, and our research has shown that callers generally prefer it over tone-based applications, as long as it works properly. Speech still requires considerable expertise to implement well, from proper interface design to well-tuned grammars. Cutting corners can lead to a poor application, which is often far worse than a tone-based alternative.

An improperly tuned speech application essentially negates the entire investment in the application.

Impact:

 

An improperly tuned speech application essentially negates the entire investment in the application, as the end product will often be unusable, either because of poor recognition accuracy, or frequent out-of-grammar responses.
Symptoms:

* Recognition accuracy less than 90% overall
* Frequent invalid account numbers, PINs, etc.
* Callers often go to live agents
* Low call completion within the speech application

Prescription:

  1. Include a rigorous usability testing and speech tuning process as part of the application design.
  2. Retain a specialist in speech application design and/or tuning for post-deployment improvement.

ROI Impact:

The payback from tuning a speech application can be almost immediate, particularly if the application wasn't performing properly to begin with. We also advise anyone embarking upon a speech application to make sure the designers and programmers have specific speech experience, since expertise in speech applications is important to a successful project.

Newsletter Archives