Difference between revisions of "AMD Initial Discussion"
(Created page with "= Answering Machine Detection Initial Discussion = * For posterity, since we've had this discussion with clients many times. '''(Obligatory disclaimer)''' I Never recommend AMD...") |
(→3rd Party AMD =) |
||
| Line 17: | Line 17: | ||
To be clear: Our last test on US known answering machine numbers for a client showed better than 90% accuracy for both human and VM identification. It rarely identified a human as a machine, which is good. |
To be clear: Our last test on US known answering machine numbers for a client showed better than 90% accuracy for both human and VM identification. It rarely identified a human as a machine, which is good. |
||
| − | == 3rd Party AMD === |
+ | === 3rd Party AMD === |
When there '''were''' 3rd party AMD systems, they required installation of a full extra server through which you were required to pass all calls. That box would make the AMD determination, and send a special signal back to Vicidial (not Asterisk) because the Makers of that software made a special deal (and that deal was not with Digium/Asterisk, but with Vicidial). |
When there '''were''' 3rd party AMD systems, they required installation of a full extra server through which you were required to pass all calls. That box would make the AMD determination, and send a special signal back to Vicidial (not Asterisk) because the Makers of that software made a special deal (and that deal was not with Digium/Asterisk, but with Vicidial). |
||
Revision as of 17:50, 14 February 2018
Answering Machine Detection Initial Discussion
- For posterity, since we've had this discussion with clients many times.
(Obligatory disclaimer) I Never recommend AMD. Almost every client I've had who tried it, even those who got their statistics over 90% with a later installation of Asterisk, eventually abandoned it.
It always requires two seconds of dead air to "decide", and that two seconds of dead air added to the (approximately) 1.25 seconds of dead air naturally occurring in Vicidial pushes those receiving the calls beyond the realm of patience and into "!&%$! Call Center! Git off my phone!".
Even if they don't say it, they think it. Now the agent has to dig out from under this attitude that was totally unnecessary.
IMHO: Better to let agents routinely press "VM" or "A" and otherwise get Live Happy callers. Inconvenience for the agents who are being paid to push a button is much better on the pocketbook than depressed sales and Vicidial Agents dealing with significantly more "attitude" than they need to while losing sales as a result.
Accuracy
AMD in older Asterisk systems is notoriously bad. 70% is about right. But in the newer Asterisk systems, it has improved. In fact, those who used to sell 3rd party AMD products no longer sell them. I'm not sure if that's because they sucked, because the newer AMD in newer Asterisk put them out of business, or if they just didn't sell enough to maintain the product.
To be clear: Our last test on US known answering machine numbers for a client showed better than 90% accuracy for both human and VM identification. It rarely identified a human as a machine, which is good.
3rd Party AMD
When there were 3rd party AMD systems, they required installation of a full extra server through which you were required to pass all calls. That box would make the AMD determination, and send a special signal back to Vicidial (not Asterisk) because the Makers of that software made a special deal (and that deal was not with Digium/Asterisk, but with Vicidial).
