1) What is the minimum number of interviewers we should have on a project when we run in predictive dialing mode?
Ten is a reasonable number to get the best performance from the PRO-T-S predictive algorithm. We estimate that a minimum of five (5) interviewers should be waiting in the queue for reasonable performance.
2) We were recently working with a friendly competitor on a project and they told us that they run projects with 2 people and 500 numbers on the predictive dialer and it works great. They said that MSG has told them that the dialer is smart and is constantly making adjustments to how many numbers it is dialing based on hit rates, how many interviewers are waiting in the queue, etc.
Dialing predictively with a small number of people will help some, but it won’t be ideal in terms of optimizing PRO-T-S performance. However, there is no down side to running at that low level, but be mindful that virtually all dialings will be proactive.
3) Should we be changing any settings when working on a project with 2 people versus a project with 20 people OR a project with 50,000 numbers versus a project with 2,000 numbers?
No.
4) Does it matter if we have one project running with 2 people and another project running with 20 people?
No. PRO-T-S looks at the projects separately and adjusts the predictive algorithm accordingly.
5) Should we be doing predictive when quotas are near full?
At that point, it may make sense to switch to proactive dialing, but if quotas are that close, you’ve probably already reduced the number of interviewers, which means that more dialing will be in proactive mode anyway. You need to judge whether you will spend more effort trying to control quotas compared to what a few extra interviews might cost.
Also, when nearing quota, one could tighten up what the sample manager releases to PRO-T-S when a number request is made. Also, if using CfMC Version 7.7 or greater, we have introduced support for a Number_Recall message. This can be invoked by a SurvSupr command - the basic syntax is: number_recall <study> <all/phone_number>. When PRO-T-S receives a "number recall" message from CfMC, it searches the queues; if the number has not been dialed, it is removed from the queue and returned to CfMC as not dialed.
6) What is the minimum amount of sample required to run predictive?
5000 would be a starting point. Take the number of interviewers to be working a job, multiply that by 75 numbers and then multiply by the number of hours. Other things to take into consideration are the length of the survey (short vs. long) and the connect rate.
7) What is the best way to dial clustered sample?
There is no easy answer to this question. This is something that would need to be controlled on the CATI side when sample is loaded. That being said, MSG would need to know what type of clustering was being used, what is the normal process and call rules for handling sample for these types of projects.
8) Is there a way to control how many numbers are sent to PRO-T-S in a batch?
Typically, the amount of sample requested by the dialer is 50 at a time. Some CATI systems allow numbers to be hidden to help control the flow of sample to the dialer, so as quotas fill, the sample associated with those quotas should be taken out of the mix. Sample sitting at the dialer will still be dialed, but additional sample will not be added for dialing. If you are using CfMC 7.7 or higher, sample can be recalled through CfMC’s SurvSupr.
9) It has been said it can take up to 45 minutes to dial all the sample left over. Should we be dialing it, or can the sample be “returned” to the CATI system – or simply flushed.
You should be able to stop the interviewers and studies once they get to a stopping point. PRO-T-S automatically flushes any numbers it has for the CfMC study when the study is stopped by CfMC at the PRO-T-S system. CfMC maintains a record of all unresolved sample at the dialer so that it can properly handle the unresolved sample when the study is closed. You should not have to dial all the "left over" sample on PRO-T-S. Additionally, if you are using CfMC 7.7 or higher, sample can be recalled through CfMC’s SurvSupr.
10) When calling into California, do we have to keep our abandonment rate at 1%?
The 1% rule applies only to calls both originating and terminating within California.
11) What abandonment rate should we use?
The industry standard seems to be hovering around 3%.
12) What is the definition of abandonment rate?
The calculation for abandonment rate is 100% x [abandoned calls / (abandoned calls + connected calls)].
13) What type of projects is it best to use PRO-T-S on?
PRO-T-S can be used on any study, there is always upside, but lower incidence, low contact rate and short interview studies will yield the highest productivity gains.
14) What about daytime calling?
Even though you won’t see a large productivity lift, still use proactive dialing for daytime. You will get accurate dialing and a pacing of the interviewer.
15) How many rings should we allow on a No Answer?
Set rings to four (4) - most answering machines will pick up after four rings. It will be more efficient overall to allow PRO-T-S to handle the call, terminating before the answering machine picks up, than having the interviewer deal with the answering machine and have to make a manual disposition.
16) Some have had trouble waiting for sample in the past.
Complex quotas can be a problem. Some have had trouble in the past with a slow CATI server.
17) What productivity gains can be expected?
For proactive dialing, you should see at least 24%. Predictive dialing should take that up to at least 30%. These are very conservative estimates. We have seen a tripling of productivity on some very low incidence projects.