In talking to predictive dialer users around the world, we come across many misconceptions about how predictive dialers actually work, and how to get the most from them. Here are the top 5, along with some suggestions for better practice.
- The longer I set my Ring No Answer (RNA) time, the greater my productivity will be.
RNA is the duration an initiated call rings at the destination before being killed as a No Answer. The reality is that 95% of consumers answer the phone within 18 seconds, and setting an RNA longer than that just pushes line costs up and agent productivity down, not up! Not sure why? Just ask us.
- Answering machine detection (AMD) must be beneficial because it cuts down the number of non-live calls connected to my agents.
Yes, it can help, but 85% detection accuracy is as good performance as you are likely to get in most cases, and even then there is a price to be paid. Firstly, you will be hanging up on live callers thinking that they are answering machines, putting you at risk of trouble with the regulators. Secondly, you are likely to be keeping the consumer waiting for 2-3 seconds before putting the call through, which just annoys people and lowers the quality of the call. Any attempt to go above 85% will make these two effects worse. But ask for a copy of our paper on whether to use AMD at all.
- To gain the maximum benefit from my calling list, I should pass through it once, then call all the answer machines, no answers and busies again.
Batching a group of previous non-connects will probably lead to more non-connects, and agent left waiting for a live call. It is better to keep the connect rate steady, firstly by combining fresh list data with smart recycling of individual outcomes (e.g. calling answering machines at a different time the following day), and secondly, by preventing supervisors from cherry-picking data at the expense of overall campaign performance.
- Predictive dialers need to be managed according to observation,
e.g.a. slowing the dialing rate when there are too many abandoned calls
b. dialing at say the reciprocal of the average connect rate, or according to average talk times, or at ‘x trunks per agent’The problem is that during any campaign, conditions such as connect rate, agent numbers, long/ short talk times, are liable to roller coaster. If the dialing rate is either fixed or under the control of a manager, it can lead to agent thumb-twiddling on one hand, and silent calls on the other. It is better to use a dialer that reacts immediately and automatically to these changing conditions.
- Call blending is unproductive and should not be used.
The problem here is that agents tend to be good at either outbound processes, or inbound, but not both. Deploying agents to work outside their skill area degrades call center performance.The answer is firstly to restrict blending to those (rare) agents with both inbound and outbound skills, and secondly to blend not only voice traffic but multiple media types as well. And where do you find multi-disciplinary agents to deal with email, chat, sms, social media, etc? Anybody with teenage children will see how naturally young people interact with multiple media sources. Your multi-disciplinary staffing issue could be a solution for youth unemployment!
If these myths are familiar to you, you are not alone. We hope the above suggestions will improve your use of existing call center software. If you would like more detail, please ask us (firstname.lastname@example.org) for the white papers we have produced on these subjects.