Predictive dialers aren't necessarily just about raw dialing performance. In the eyes of many users, they are also about the ability of the dialer to do 'call progress detection' to screen out as many non-live calls as possible. In the US especially, this usually becomes a debate about how effective is answer machine detection (AMD). The truth is that provided the algorithms are well-tuned (which isn't that difficult), there isn't a lot of difference between manufacturers.

Watch out for the marketing handouts that claim over 90% (sometimes over 95%) detection rates for answering machines. One not insubstantial US vendor even claims, publicly, 100% accuracy. Watch out for 105% next!

Here is a typical example of the kind of artificial test that produces such high claims. The dialer gets one of two responses:

  1. Live answer of a single "Hello" followed by silence, or

  2. Answering machine greeting of continuous speech such as "Thank you for calling. We are not in at the moment. Please leave a message."

Any answering machine detection algorithm not being 100% accurate under such conditions deserves a good telling off!

But this is not real life. For example, what about the person who answers the phone, saying "Hello, John Smith here", and then when he doesn't get an immediate response starts saying "Hello...Hello..."?

In the UK, the telcoms regulator Ofcom has realised that because answering machines aren't that accurate, they sometimes hang up on live people in the belief that they are answering machines. Such calls are deemed false positives and they count towards the strict limit of 3% allowed for abandoned calls under Ofcom rules. For some time now there has been a rather silly debate in the UK as to whether the use of answering machine detection and thus false positives on the one hand and predictive dialing on the other are compatible. They simply aren't and one day Ofcom may have the courage to say so, stop the debate and allow us all to focus on more productive things.

But give Ofcom its due. At least they know about false positives. No other regulator appears to understand the issue and in countries such as the US, call centers are free to deploy answering machine detection without any fear of censure, through lack of accuracy.

So does unfettered use of answer machine detection make sense, when regulations don't get in the way? Ask yourself this – and it applies to the UK too. Why would you want to take all live calls and hold them all up for a number of seconds trying to figure out if they are a machine or not? How do you feel as a consumer when this happens to you? Many people would either hangup or just get annoyed. Imagine what that does to the quality of the call if a consumer stays on the line? And that's why savvy call centers don't go there because the loss in agent talk time per hour is more than compensated for by the improvement in call quality that an immediate connect brings. And there is another benefit: when the agent encounters an answering machine he can leave what is clearly a personalised message rather than a canned one.

The good thing about these ideas, even for the most dyed in the wool advocates of answering machine detection, is that you can test for the impact, by turning detection on and then off, and measuring the differential impact on bottom line performance over time.

We apologise for the unusually long blog this month, but these are serious matters. Sure we are opinionated (but based on very extensive experience in many countries) and in search of brevity have perhaps cut some corners. If you want more, or to debate any of our conclusions, just let us know.

Media Services

View Product »