28 May 2015

Background:

One of the most difficult things for any automated calling system is to determine if the party which answers is  voicemail or an actual human.

Because of the many different types of carriers, handsets, voicemail vendors, and voice delivery technologies which make up today’s telephone networks, there is no reliable way to determine whether a call has gone to voicemail or if an actual person has answered your call (sometimes even a carrier will mistakenly think a call is answered when it is not, such as in the example of an old-fashioned telephone answering machine. These devices actually answer the call to play a recorded message). Even if we ignore this small percentage of people with old-fashioned answering machines, it’s unlikely carriers would open up their networks to share this information with a large number of outside businesses on a per call basis.

Decision Criteria:

Since we can’t be 100% sure that a call has been answered by a machine or a human based on the network technology, we must base our decision of Human vs Machine elsewhere.

It turns out, people exhibit strong tendencies to how they answer the telephone or how they record voicemail greetings. So strong, in fact, that estimates suggest we can be as high as 90-95% certain that a particular call was answered by a human or a machine by analyzing the way the call was answered. The following tendencies are the foundation for this analysis:

1. Generally speaking, humans tend to answer telephone with a short greeting such as “Hello”, “Hi”, or “This is Nathan”.
2. Generally speaking, voicemail messages tend to be much longer, such as “I’m unable to come to the phone right now, please leave your name and number…” or “I’m sorry to have missed your call, but I will be unavailable from 1PM to…”

So, if we have an analyzer listen to the initial greeting, we can do a pretty solid job of determining if a live person is on the other end… or if you’ve reached a machine.

Drawbacks and Limitations:

Of course, there are a number of nuances and exceptions that bring down our confidence in our determination. Here are a few:

1. Live person with long greetings. Retail businesses tend to require employees to answer the telephone with much longer greetings to announce their slogan or their specials, such as ”Thank you for calling Joe’s Pizza, would you like to try our triple-pepperoni and sausage special for $9.99?” or “Dave’s Car Sales, where everyday is a great day to buy a car! This is Pete.” Some people may also answer their phone in a long-winded way. These events will cause the analyzer to mistakenly judge these calls to be answered by a machine.
2. Background noise may be the biggest hindrance, especially in today’s cell-phone world. The analyzer may think that background noise is part of the greeting, and consider a short greeting to be a very long one. This is probably the biggest reason why a human-answered call might be judged to be a machine-answered call.
3. Delays prior to the initial greeting or slow speaking greetings may throw off the analyzer, since it may determine that a greeting was over when it actually wasn’t finished.

But in general, these are the exceptions rather than the rules. Especially if you’re less likely to be automating calls to the retail line for many types of businesses or calling home numbers instead of cell phones, these may not have a significant impact in your confidence levels in determining Human vs Machine.

Conclusion:

Overally, this method for determining Human vs Machine is a fairly reliable method, and its impact to your broadcast is likely small.

This method does have some bearing on the quality of the voice messaging being played. Specifically, the amount of silence that is played prior to the message might be a direct result of the greeting analyzer’s configuration. We cover this topic more detail in another discussion.