The “Knowledge Economy” and “Age of Information” dominate all aspects of our society today. Evidence of this is demonstrated through the huge volumes of data and which are now easily accessible through big data type technologies and the use of artificial intelligence (AI).
Contact centers have certainly not been insulated to the use of data and its associated outcomes, and in fact have always embraced the use of data to better serve their customers. But the emphasis on data and its use continues to grow as all industries become much more data-driven.
In this unique environment, the question arises of how should contact centers respond?
The Amex Experience
Before conveying some ideas on an overall approach, my early experience can shed some light on what leading-edge institutions such as American Express did in the early 1990s with their contact centers.
My role at that time at Amex was the development of credit risk and fraud models. But one of my key deliverables, given my prior experience in developing marketing behavior models, was the development of profitability models at the customer level.
Once these models were developed, our challenge was to socialize the use of these models in our contact center. We had to recognize and improve the understanding of our work and its utility to enhance decision-making within the contact center.
So, collaborating with the director of operations, we conducted informal lunch and learn sessions every Friday to groups of 10 people comprised of contact center managers.
Our objectives at these sessions were twofold. The first one was to educate the field staff about these models and specifically the business objectives behind them.
In other words, what were the key business elements that comprised a profit score, which would have included broad elements related to customer spend and customer risk.
At the same time, we would have discussed how the information was relayed on the screens. Besides profit scores displayed, we segmented customers into three profit tiers (high, medium, and low) to present the information in a more meaningful way.
Our second objective was to discuss how the added information might be used. Rather than displace existing information, we conveyed the fact that these profit scores or profit tiers could be used to augment existing information. For example, two people with the same payment history but with different profit tiers would require distinct types of communication from the same contact person.
The ML-NLP Empowerment
Recent advancements in data science, in particular natural language processing (NLP), can empower contact centers to leverage what people are saying to the point where the machines i.e., the computers, now responds to the customers.
Let’s look at chat boxes. It is a familiar business practice and used by most organizations today. The difference for many of them is the level of sophistication that is being used to analyze this text.
In many cases, the chat boxes are quite simple where the machines ask prescribed questions, which are dictated by the responses of the customers. But the determination of these questions is programmatically determined through business rules.
The interaction between the chat box and the human is limited by the business rules which have already been pre-defined. But with machine learning (ML), this limitation is no longer there.
Computers, through techniques such as NLP and the advent of big data and AI, has accelerated the use of ML, where the machines learn on historical text and utilizes that knowledge to generate responses that are specific to the questions. The applications of Siri and Alexa in our own home devices are manifestations of these current capabilities.
Although the benefits of this approach for companies are the reduced expenses associated with labor, along with shorter and potentially more satisfying interactions, the corollary is that all too often there is increased customer frustration with automated computer responses.
These negative customer experience causes customers to call and insist on speaking to live agents: which offsets the staffing savings and lowers the ROI from the technology investments.
The Hybrid Role
But is there a hybrid role here for the machine and the human? Could we connect technologies, such as NLP and ML, with the agents’ skills and knowledge to provide even better service to the consumer?
For example, analytics as observed in the Amex case could be used to identify key behaviors based on what customers do. Like “are they likely to go bad, how profitable are they, and are they likely to leave us?”
ML-based predictive models, which represent common business practices, are the tools underpinning the access to this type of information.
But the competitive advantage is to utilize both types of tools which allow the use of ML to predict consumer behavior and its resultant prescriptive outcomes, complemented with what the consumers have said, and to then determine the appropriate forms of communication.
Could we connect technologies, such as NLP and ML, with the agents’ skills and knowledge to provide even better service to the consumer?
Here is where the agent supersedes the machine through this type of complementation.
Let us look at one example. Suppose I call up the customer service center at the Royal Bank of Canada. The agent receives my identity-related information and knows that I am a high-value customer, but one who is likely to defect based on ML models.
Yet my previous conversations with the bank now provide the agent with information through NLP and text mining techniques on how to communicate with the customer, like myself, more effectively.
For example, the agent may determine that this high value, likely to defect customer through previous dialogs may have specific interest in wealth products. Using this information, the agent might emphasize the unique features and benefits of this organization’s wealth products.
In another example, a customer is more likely to be higher credit-risk using ML models. Yet using text mining and NLP, previous communication dialogs indicate interest in more lending-type products. Once again, armed with this knowledge, the agent might suggest a plan to pay down debt with the goal of having access to those types of products.
Building the Framework
Contact centers exist in an ocean of data that is readily available for access and use by the agents. Yet companies are often resistant to the use of additional humans in their contact center strategy as it is easier to defer to the use of automated chat boxes because of their immediate cost advantages.
The more challenging aspects of evaluating longer-term profitability through this hybrid approach are not readily adopted by many organizations.
The establishment of a measurement framework is what is really required to look at what is truly incremental with regards to the hybrid approach versus the sole use of chat boxes.
Yet building such frameworks is not easy as there needs to be collaboration amongst all the key business stakeholders on those elements which are incremental, as well as the time periods for these evaluations. At the same time, the design of measurement templates would also need to be agreed upon by all the business stakeholders.
Data scientists and ML practitioners with any degree of business experience have always advocated the use of both the human and the machine in any of their analytical efforts.
It is this collaborative hybrid approach, connecting technology with people, which is the key to success for contact centers. The simple short-term myopic perspective of cost-efficient chat boxes must be given less focus as organizations opt for the hybrid approach and its longer-term goal of greater ROI.