We are now beta testing our brand new math algorithm for predictive dialer to be used in Aurus Outbound product. The algorithm was recently developed by our engineers in collaboration with the Institute of Mathematics and the first test results make a killer impression.
Here are some figures…
We performed 3 synthetic tests with the following input parameters:
- client’s time to answer: varies from 15s to 25s;
- The success rate (percentage of calls answered by live person): 80%;
- the number of agents involved in the campaign: 50;
- the campaign duration: 58 min;
- call duration range:
- test 1: 40-60s – to emulate campaigns with short call time, for example address verification, appointment reminders etc,
- test 2: 60-120s – for campaigns like subscription renewals, mass mailing follow-ups, “welcome calls” and so on,
- test 3: 500-600s – for sales campaigns, surveys, market research and so on.
The agent occupancy calculated in the tests above:Predictive vs Progressive – Agent Occupancy
As expected, the agent occupancy in progressive campaigns depends on the average call duration – the more the talk time is the less failed calls influence the total agent utilization. However, when using predictive dialer the agent occupancy is over 90% in all tests.
Another important metric is the call abandonment rate which indicates the percentage of “nuisance” calls – the ones where the client answers but no agent is immediately available to talk to him.
In our tests the abandonment rates are:Predictive Dialer Abandonment Rates
The highest average abandonment rate is achieved in the 3rd test because of the highest (500-600 sec) and the most spread out (100 sec) call duration.
Having high call duration variance we may get too high abandonment rate which is unacceptable due to regulations like Ofcom and FCC (Federal Communications Commission). This can be illustrated by 2 others tests:
- the call duration is 400-600 (variance - 200s)
- the call duration is 400-700 (variance - 300s).
To avoid this our new predictive algorithm allows to setup the max abandoned rate input parameter. So, when the abandonment rate of the active campaign gets closer to the max value specified, the algorithm reduces the number of calls.
Predictive Dialer in Low-Volume Campaigns
It has been argued that predictive approach is only effective when used in high-volume (30+ agents) and long-lasting campaigns to get enough statistics for proper prediction.
We performed a couple of tests with low-volume campaign parameters and higher variance:
- the campaign duration: 10 min (in opposite to 58 min in the tests above);
- the number of agents: 10;
- call duration:
- test 1: 60-120 sec
- test 2: 60-180 sec
As we can see the effectiveness of predictive algorithm in low-volume campaigns reduced by 10%, but still it provides extra 10% of agent utilization when comparing to progressive approach.
The new predictive algorithm will be included into the nearest release of Aurus Outbound (Jan 2016), so you’re welcome to try it.