An automated dialer, like all powerful tools, must be used carefully. This article is focussed on avoiding issues with dropped (silent, lost, abandoned ...) calls and related issues to do with using answering machine detection. We'll dig into the metrics and the thresholds applied in guidelines and by some regulatory bodies.
The undesirable things to avoid
The particular dangers of using a dialer are:
- connecting calls to people and having no employee to take the call
- connecting calls to people and incorrectly automatically categorizing as an answering machine
Both cases result in the call connecting briefly to a real person and then the call ending. In the call center, direct marketing and sales areas, these undesirable occurrences are sometimes called "dropped calls" and sometimes "abandoned calls" or "lost calls". The term "silent call" is also used, since the person who takes the call hears nothing and then the call ends.
We usually use the term "dropped calls". We shall refer to them as "dropped live answered calls" when we want to differentiate them from dropped calls where an answering machine is correctly detected. So "live answered" means that a real person actually accepted the call.
How tolerances are laid down in regulation and guidelines
There are regulations in many territories that aim to limit the occurrence of dropped calls and the impact on consumers. There are also best practices recommended by industry bodies and consumer organizations among others.
In general regulatory bodies like Ofcom in the UK, FCC in the USA and similar organizations elsewhere publish research on the topic and also document what a company using a dialer must measure and what tolerances are allowed for the frequency of undesirable dropped calls.
These tolerances and the related regulations vary quite a lot. For example, in the UK Ofcom published their tolerance level as not more than 3 abandoned live or falsely detected answering machine abandoned calls should occur out of 100 live answered calls. A similar measure in the USA had an acceptable tolerance of 2% but did not originally include falsely detected answering machine count.
The babelforce position and goal
Our position at babelforce is that the occurrences should be even lower than these tolerances. We see it as essential to good practice to drive down these undesirable side effects as much as possible.
So we see it as part of our mission to help businesses to achieve best possible metrics for these numbers.
How to measure the right things
It is actually not that trivial to measure all aspects accurately directly from the data that is available as a matter of course from a dialer. As with anything we need to achieve, it is first vital to understand exactly what needs to be measured to decide what success will look like.
One of the more complex problems is related to the use of Answering Machine Detection (AMD). So we will divide up the two cases.
Answering machine detection is OFF
In this case, one of the accepted directly measurable metrics is defined as follows
Since these are determined based on the outcomes that an agent labels the call with and since these may vary, babelforce as standard delivers the following metric which we call "ratio of dropped to live answered":
Just to get a feel for how this ratio works, let's do an example calculation. Let's say we had 100 calls so far today where an agent labelled the outcome so that we know they talked to a live person. So [livePersonToAgent] = 100. Since AMD is OFF, the number [allCallsDropped] should be small, i.e. if your campaign and staffing are right then we can only have a dropped call, if an agent is not available for an outbound call that is answered. For the sake of the example, let's say we had 2 so far. Then the automatic babelforce metric will be 2%.
Now if for a moment, we ignore the number of calls where an agent is connected to an answering machine, then the first metric above gives us 2 divided by ( 2 + 100 ), i.e. damn close to 2% (1.96..% to be precise).
Let's say, that so far agents connected to 20 answering machines. Then the first metric gives us 2 / (2 + 100 + 20) = 1.6%
So the babelforce standard ratio of "ratio of dropped to live answered" always overestimates one of the standard metrics. So attempting to minimise the babelforce ratio will definitely keep you on the good side of good.
Answering machine detection is ON
Since all answering machine detection (AMD) will definitely have "false positives", i.e. there will definitely be a number of occurrences where the AMD algorithm will determine that the call was connected to a machine but in fact a real live person took the call.
These occurrences are likely to be greater in number than the normal dropped calls, since as we noted above, a properly operated campaign related to proper staffing can avoid connecting to callees when no employee is available.
For this reason, regulators and industry bodies often recommend a way to factor in these false positives. Here is a common way to get an estimate of the rate needed
The formula looks scary, right? But actually only the top has changed. Basically the frequency of occurrence of false detections of answering machines is multipled by the calls connected to agents.
Here is the tricky part. The FalseDetectRate can only be determined by sampling the data for your campaign with your audience. So it is not possible to give a standard calculation of this automatically just from data available directly from your use of a campaign.
IMPORTANT: if you use answering machine detection, then in many territories it is your responsibility as the business using the dialer to assess that false positive rate for answering machine detection as well as the dropped rates above. Penalities can be applied to businesses who cause silent, dropped, abandoned calls and you are required to be able to provide data on how your use of campaigns with answering machine detection operate.
DO NOT USE AMD in any situation where it can reasonably be avoided.
For example, if your employees are detecting and labelling answering machine connections in a few seconds, then the impact of that is far lower than the impact of a number of your customers waiting for a few seconds on a silent call that then ends.
Note also that AMD will always introduce a delay in the call connecting to an agent, even when everything goes perfectly. So there is a customer experience cost anyway even in best cases.
The vast majority of the benefit of using a dialer is achieved not from AMD, but rather from the automation of integrated processes, the reduction of manual steps, the standardization of status codes and recycling/rescheduling rules, the reduction in data entry and the automated discovery of unclean data.
Secondly, if you do use AMD in your campaigns, then make sure to do sampling to detect the false positive rate. This will allow you to determine the impact of AMD more completely and is the only way to ensure you are within reasonable bounds or within acceptable thresholds for territories where they apply.
In short, consider very carefully the perceived pros of using AMD compared to the likely cons.
Some links to other resources
In each territory where you operate a dialer and in each territory where your customers live, you should consult the relevant latest regulations and guidelines.
Here are a few organizations and links that might be useful to you:
Ofcom - UK communications regulatory body: https://www.ofcom.org.uk/about-ofcom
German Call Center Verband Branchenkodex: https://callcenter-verband.de/verband/branchenkodex/
Bundesnetzagentur - German communications regulatory body: https://www.bundesnetzagentur.de
Information to misuse from the German Bundesnetzagentur: https://www.bundesnetzagentur.de/cln_1421/DE/Sachgebiete/Telekommunikation/Verbraucher/Rufnummernmissbrauch/Missbrauchsfaelle/missbrauchsfaelle-node.html
FCC Federal Communications Commission - USA regulatory body: https://www.fcc.gov/general/telemarketing-and-robocalls