Re=Post | In Case you missed it | Using Asterisk as an inexpensive Answering Machine Detection (AMD) appliance
The following is Part One of a Blog Post that Mike Mastro (Present and CEO of Primas) put together last year. I am re-posting just in case someone out there needs this ;-)!
Using Asterisk as an inexpensive Answering Machine Detection (AMD) appliance
May 7, 2018 // Asterisk, Avaya, IVR, Press Releases, Primas Blog
After all these years… my very first blog post! Hello and welcome to Primas’ blog. For most of you that do not know us, we have been in the industry for more than 24 years creating custom solutions for Call Centers from fortune 100 to small centers, generally as a third-party developer behind a system integrator that is well known. In the past few years we have moved to product company and therefore need to create awareness of Primas, hence I was instructed to share some of our experiences and secrets through our blog to expose our creativeness. I hope you find some of these facts useful! Our first post will be:
Using Asterisk as an inexpensive Answering Machine Detection (AMD) appliance
Situation:
Our current customer has an Avaya switch with a Genesys Campaign Manager. Their current solution utilizes Genesys to initiate the campaign call, and Avaya to provide the Answer Supervision of the call to determine if it was answered or not. When the call gets answered, the call is then transferred to the agent queue for processing.
Problem:
Answered calls can be either A) Real Person or B) Answering Machine. So, as you would guess, a majority of the calls answered by the agents are answering machines, which creates very low productivity because these calls can be automated for leaving messages by an IVR or another announcement device.
Root Problem:
In order to do “Real” Call Progress Detection, a combination of signaling messages as well as “Media” Classification needs to happen. The last part of this formula to do Media Classification requires special technology (and Licenses) that performs the analysis either in hardware or software which “listens” to the media. For those interested, the technology listens for noise, not speech recognition. A person is a short burst of noise, followed by silence (“hello………..”), and an answering machine has a lot of noise (“hello you have reached Mike Mastro and I am not here, blah blah”). It is that simple.
For our current customer, this feature was never purchased due to the extreme cost of the solution, so the expense was passed on to the operation.
Solution:
When this problem was discussed in general conversation, I suggested that this classification can be done inexpensively using Asterisk Open Source software, with no licenses and no hardware costs = saving 100’s of thousands of dollars.
The concept was to use the asterisk software to do the classification and then transfer real calls to the ultimate agent queue or remove the Answering Machines and send them to an announcement, thereby removing them from the agent queue.
Now, having implemented the concept, the campaign sends the call to an Asterisk VDN via a SIP trunk, which then classifies the call and immediately transfers the call back to 1 of 2 VDNS depending on classification. If done correctly, the Genesys system sees the call coming back and continues processing as usual.
Our goal was to achieve >80% and initial customer tests showed >95%. We think this is a bit high for real production, but nonetheless, the test was successful and the customer is proceeding to production.
In Part 2 I will discuss some of the pitfalls we hit during the way.
Happy trails or trials?
Mike
Mike Mastro | President | CEO | The Primas Group
Learn more about Primas at www.primas.net