Best practices in Predictive Dialing
Here are a few suggestions to help you get the best out of your CTArchitect predictive dialer.
The first thing to note is that there is a "trigger" that will give a signal to CTArchitect to start counting the number of seconds it takes for an agent to finish the interview. This trigger is key to the optimisation of predictive dialing.
1. Enable the trigger
- In your questionnaire routing. The system will start to count from the question where the trigger is set.
- Or you can let the system decide when next call must be dialled. In this case the trigger will be on the last screen.
The best way is to place the trigger routing questionnaire:
- On the screen where the estimated interview time will be higher than the ‘no answer time’ setting (see ‘dialling settings’ - menu).
- But not higher than one minute.
We recommend setting a trigger two or three questions before the end of the interview. But it will depend on the average amount of waiting time the system is working with. As an example, if the waiting time is 20 seconds and we estimate 10 seconds per question at the end of the survey, then setting a trigger two questions before the end sounds right. And this assumes that the trigger is set on a question that everyone sees.
If you have different exit points in your survey, we recommend setting multiple triggers (of course limited to one per exit point).
2. Configure the algorithm
The next part will depend on the number of agents that will be working and the limits you want to build in. For optimal results, we advise keeping the recommendations as described above. When different skills are defined on the task, it is better to place the skill variables in the quota tree. In this way, the contact distribution from the sample list on CCA towards CTArchitect will go faster. If the skills of a particular agent are also present in the quota tree, the best contacts available in the sample list for this particular agent can be calculated more quickly by CCA. But be aware that when using skills, calls cannot be distributed to all agents, since they may not have the same skills. The more agents with similar skills working at the same time, the better for the algorithm.
Default settings:
With this above configuration, the system is able to run a predictive ratio of +/- 1.5. The system can also slow down according to the level of nuisance and abandoned calls: when the level of the last 25 calls is tracking higher than the parameters that are configured, the rate will slow down.
Note: the limitation is calculated for double the number of agents, in this case 7 or 8 agents.
Aggressive settings:
Here the maximum call limit is removed. Thus, the system will take as many channels as it needs. The level of nuisance and dropped calls has been doubled compared to the default configuration.
Very aggressive settings:
NB: This configuration is not recommended as the maximum abandoned call level and likelihood of nuisance calls are very high. Your phone number risks being blocked by the telephony authorities.
3. Check the performance
- Firstly, the ‘list monitor’ functionality allows you to monitor the system in real time:
- The ‘nuisance calls’ and ‘abandoned by dialler’ figures are very useful.
- When the maximum percentages of these figures are reached, CTArchitect will automatically decrease the number of calls to dial to a level that can be compared to a progressive dialling speed. In other words: CTArchitect will launch one call at a time for each agent, thereby avoiding the risk to create additional nuisance or abandoned calls.
- The second monitor available is the ‘agent’ window within the Supervisor application:
- The average and maximum waiting time can be monitored when an agent has been waiting for the next call for too long, you can select the “no calls?” option on right click menu. The application will then try to present a possible reason for the long waiting time..
- The third monitor is the quota. Here you can see if a particular quota is met. When a quota line is full, there could be an effect on the dialling speed.
- Other data collection tips for better performance: a higher number of agents working in the same outbound group can result in better performance.
- Agents should try to keep as little variation in timings as possible between the dialling trigger and the end of the interview. For example:
- Agents should not remain on the last screen when they make a mistake, but instead should go on pause.
- Agents should not open the appointment window on Askia Client and close it several times during the same interview. This will make it easier for the dialler to calculate an accurate algorithm for each agent. VoIP CATI agents should have the “end call” button enabled during ready state in predictive dialling mode, and hear the ringing sound all the time while they are waiting for a respondent.
- When the “ringing” sound stops, the agent must start talking immediately (due to answering machine detection, the agent won’t hear the respondent pick up the phone and saying “hello”).