Genetic Algorithms in the Contact Center
Artificial Intelligence (AI) is a term we hear repeatedly in the context of innovative technologies. Underneath that overarching term, there are a myriad of ways in which AIs are developed and deployed, with varying degrees of human involvement. There are different approaches that can be taken to machine learning including the deployment of Automated Machine Learning (AutoML) technologies from companies such as Google, TPOT or H20, but these can be burdensome for midsize organizations to deploy and manage. They require specialist skills that may not be available outside larger enterprises. Some innovative research underway between Avaya and one of the largest dedicated AI and Analytics Research Centers in EMEA believe that genetic algorithms may offer a solution for smaller businesses and are testing that hypothesis.
Genetic algorithms are inspired by Charles Darwin’s theory of natural evolution. The idea is that selection of the fittest individuals in a population takes place for reproduction and produces the offspring of the next generation thus resulting in more successful individuals generation after generation.
Avaya is working closely with The Insight SFI Research Center for Data Analytics at National University of Ireland Galway, on innovative genetic algorithm research with the chosen data set being that of a typical contact center. The research was presented at a recent conference. The collection and analysis of data is expensive and time-consuming and Avaya is invested in exploring methods to automate as much of this as feasible thereby reducing costs which benefits all parties involved. Contact Centers generate vast amounts of customer data with many touchpoints, social channels and systems that contribute to the overall customer journey. We know what successful outcomes look like: for example a satisfied customer or a purchased product, and equally we know what an unsuccessful outcome looks like!
The Data Gene Pool
The genetic algorithm process starts with a population – a set of ways to look at solving the problem. In the context of this research, our “gene pool” consists of ways of analysing CRM data such as customer detail records and order or ticket information. Standard Contact Center data such as date and time of call, duration, time in queue, and the answering agent’s profile are also data points. Post call survey information is also leveraged such as satisfaction ratings, willingness to recommend and other feedback.
With a clear view on the correct (successful) outcome for the transactions, the genetic algorithms are put to work to make predictions on the outcomes of the interactions. In our starting population an “individual” is characterised by its set of parameters, or “genes”. Selection from generation to generation is based on a “fitness” function which is calculated based on success with regard to solving the problem i.e. making a correct prediction on the outcome of the customer interaction. Individuals with higher fitness levels (more accurate predictions) are more likely to be selected to reproduce. As each pair of parent algorithms are mated, a cross-over point is chosen at random from within the genes. Offspring are created through inheriting the genes of a parent until the crossover point is reached at which point they will inherit the genes from the other parent. A portion of the offspring algorithms will be subjected to a random mutation i.e. have portions of their bit-string flipped. These next generation of algorithms once again start the prediction process and are evaluated on how they perform. Selection is then applied to keep the most effective. Crossover and mutations are applied, and the next generation of algorithms are born and evaluated. This happens through iterative generations. Initial research on small samples has led to up to 20% improvement in just 5 generations. This is a different approach to classical machine learning.
Machine Learning for the Future
Machine Learning in customer experience environments can potentially predict call outcomes and automate processes or better route calls to agents for optimal results. It can learn customer trends, sentiment and areas for improved quality monitoring as well as identifying anomalies and targeting fraud. It can also be used to identify a company’s best agent behaviours to reward and identify new and upsell opportunities. Avaya is optimistic that discovering relationships in past datasets and applying optimal learning algorithms in an automated way could make the benefits of AI accessible to all sizes of organisations. For more information on how Avaya’s innovative AI solutions can transform business outcomes, explore the latest trending technologies on the Avaya AI Blogs and watch the video guides, case studies and more on Avaya Stream.