A dialing journey to AI - In recent years, the IT airwaves have been lit up by new computing techniques and buzzwords presenting themselves as Artificial Intelligence solutions using techniques among others such as machine learning and deep learning. The general idea is that you have a collection of data, you invoke some algorithmic processes and then throw massive computing resources at it, and you end up with useful insights and results, which otherwise would just not have been achievable.
This is not seismic change but evolution. For, AI as described here, has been around for many years and Sytel itself has been making use of AI techniques in its predictive dialer, since the 1990s. We developed a massive simulation engine for calculating optimum dialing rates, the engine in the dialer, known the Virtual Event Machine (VEM®), which simulates at a rate in excess of 50 million calls a second for each campaign it is running!
So how does it work? VEM® is a classic example of machine learning as the dialer takes in information relating to a dialing campaign and figures out how to dial. One key difference, from other AI products, is that VEM® does not try to build up snapshots of model dialing conditions and use this to generate new simulations in other campaigns. It is an interesting idea and we looked at it very carefully but the science was too inexact for us and still required massive computing power.
Diagram showing the relationship between gain in agent talk time against abandoned calls for the Sytel AI Dialer.
Note that the Dialing Protector typically provides substantial performance improvements whilst keeping abandoned calls to 1% of live calls.
The approach we take is to treat each campaign as unique, and not allow VEM® to use any approximations derived from other dialled campaigns. The sheer enormous power of VEM® is such that it can very quickly calculate dialing rates in just milliseconds. And this is vital since VEM® may have to support many different campaigns simultaneously, where dialing conditions are constantly changing, often abruptly.
What this means in practice is that users of the dialer decide on a couple settings, such as the abandoned call target, start a campaign and then leave VEM® alone! There is no need for human intervention in the dialing process, nor does VEM® allow it. Dialing supervisors can be redeployed!
New features – The culmination of all the work we have done with VEM® has taken the dialer to a new level with the release of the Sytel AI Dialer, and enables two key new features, both industry firsts:
Dialing Protector: The first of the new features is validation of the dialer's ability to provide excellent performance whilst dialing under compliance at a level of 1% abandoned calls, a feature which is invaluable to those markets such as the UK, which stipulate compliance. This is unique in the industry and gives users peace of mind, knowing that the dialer will deliver excellent performance, ensure that they stay out of trouble from current or future legislation and protect consumers from nuisance calls.
Multi-Context Dialing: The second is extending the work that VEM® does to allow it to recognise and work with different record types, for example call-backs, with inherently different answering properties from other phone numbers. VEM® runs simulations for different record types and combines results dynamically to match the dialing targets set for a campaign. This again is a world first and means that call centers don't need to worry about mixing data with different calling characteristics on the same campaign, whether knowingly or inadvertently.