Don’t overload short-term memory
If you've had any exposure to IVR design, you've probably heard references to Miller’s 1956 article on short-term memory that found that humans on average can hold seven chunks of information in working memory, plus or minus two chunks. And you've probably heard it applied to IVR menus. It doesn't apply. We're not asking people to remember all the options, just pick the right one. We go into this in great detail in the sections on menu design.
That said, if your application is presenting information the caller needs to remember, keep in mind the limitations of memory. An IVR interface is transitory, that is, once we've said something, it's gone. There is often no easy way for the caller to back up, so they are relying on memory.
A good design will likely not require the caller to remember a lot of things (if any) during the course of the call. But when detailed information is presented, callers are not likely to remember all of it after the call.
Another short-term memory issue to factor into menu design is the primacy-recency effect. Studies have demonstrated that when participants are presented lists of words, they tend to remember the first few and last few words and forget words in the middle of the list. Therefore, the first option and the last in any menu are the ones most likely to stick with a caller.
Generally speaking, input requests should consist of options that make sense to the caller, and this guideline includes how the options are grouped together. Preferably, the caller will not have a “which of these things is not like the other” response to an option, because it causes distraction and hesitation, and leads to timeouts. Where possible, however, it is advisable to use the fewest number of questions to shorten the caller experience, so collapsing different types of questions or input into a single request may be done judiciously.
An example that works is as follows:
In this example, the application is asking the caller directly to provide the information that most callers will want to provide, but giving him an alternative option. Even though the prompt presents to different ideas, both are methods for account lookup with which most callers are familiar today, so it's OK to combine them.. The caller who is comfortable or able to provide a phone number (the majority) is able to just get on with it. The caller who’d rather go the address route is provided an out.
What should be avoided are multiple options mapping to a single input, particularly if the options are wordy or don’t logically belong together. A counterexample might include,
How a designer combines multiple options into a single group with a single label usually requires input from true end-users of the system. Conducting a card sort is often helpful to establish meaningful groups of tasks, and accompanying labels of groups of tasks.
Very long phrases that the caller is expected to say back as a menu choice cause problems with cognitive load, among other things. See the section on phrasing menu options for more information.
To play back a long numeric or alphanumeric string to the caller, something like a confirmation number, playing it in chunks help them process the data. Chunking simply means to break it into pieces with pauses in between. The same thing is true for reading as with hearing. Look at the following numbers.
The strings are identical, but the one on the right is much easier to read, and likewise much easier to listen to, assuming the reader puts pauses where the spaces are.
When presenting information with a familiar pattern like a North American phone number or a US Social Security number, it’s important to chunk the information the way people expect to hear it. Confirming a 9-digit Social Security number as three chunks of three digits will throw the caller off.
Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. The Psychological Review, 63, 81-97.