AI Genesys

As organisations increasingly seek to differentiate their brand from competitors, the ability to provide unique and personalised experiences to their customers is critical. Through its market leading cloud contact center solution, Genesys Cloud CX, Genesys unifies customer and agent experiences across phone, email, chat, text and social channels.

The challenge:  Routing customers to the right service agent
While most organisations recognise that connecting their customers with the right agent first time can dramatically impact business outcomes and customer satisfaction, to date this has required bespoke and complex orchestration, using static business rules. Changes to workflows and business processes can be brittle and error prone.

This is an area well suited to the application of machine learning, to automatically identify best customer-agent matches based on historic data, reducing the need to configure complex business rules. However, solutions in this space have to date been bespoke custom projects requiring integration of many disparate business applications and a team of data engineers and data scientists. Such projects are difficult to manage and maintain. Degradation or drift in AI model performance requires extensive analysis, often leading to re-evaluation of the models due to changes in the underlying data.

Using our experience of such projects, the Genesys design team sought a new approach: Could we automate all aspects of the AI lifecycle? Could we use Design Thinking to create an intuitive user interface so that a business user without any training or AI knowledge can deploy AI-based routing, thus allowing organisations of any size and capability to benefit from machine learning?

Our solution: AI-powered Predictive Routing
In summer 2021, we achieved our goal with the introduction of Genesys Predictive Routing. Predictive Routing matches customers to agents in real-time by using AI to curate hundreds of features and create machine learning models to predict outcomes of potential matches. Selecting the ‘best fit’ between customers and agents improves critical business KPIs such as reduction of call handling time, or reducing the number of times a customer is transferred from one agent to another.

To create scalable, automated AI solutions requires consistency in the data used in machine learning models. In support of this, Genesys created its own Feature Store to store and access documented, curated features to train machine learning models and make predictions. Within the Genesys Feature
Store, we standardise the engineered features and schemas of common entities (customer profiles, agent profiles, interaction profiles), thus allowing Genesys to create repeatable machine learning models accessible to all of Genesys Cloud CX’s 2000+ customers at the click of a button. Various data pipelines, specifically created to engineer machine learning features from data in the Genesys platform and from external data sources, ensure that models can automatically be re-trained on a regular basis. Each model is an ensemble of N decision trees. This implementation was specifically chosen as it provides the flexibility to support the various KPIs that our customers are seeking to improve, it provides excellent performance with smaller datasets and has a robust implementation in XGBoost.

An intuitive user experience is critical to the adoption of our AI products, due to our customers’ relative inexperience with AI. The first step for the user is to launch, at the click of a button, an assessment of their contact center’s queues, to validate that there is sufficient data volume and variance to create an
effective predictive model for each queue. Where this assessment produces a positive result, users can proceed to conduct a controlled test of the AI routing engine. Recognising that trust in the performance of AI can be a concern for customers, the solution allows for A/B testing of predictive routing’s performance against the existing routing configuration, and customers are provided with a report delivering a transparent view of the impact of predictive routing on their key KPIs.

Genesys’ customers are finding that predictive routing is allowing them to gain efficiencies within their operations while providing a better experience to their own customers. Customers adopting predictive routing have seen immediate success. A North American finance company saw a 10.6% reduction in handle time for callers to their contact center. A travel and leisure company in Australia saw a 14% reduction in their handle time over a 6-month period with a return on investment of 880%. These results, made possible through machine learning, have greatly improved on those previously possible through the manual configuration of routing business rules, and were achieved at a fraction of the operational effort to the business.