Demystifying AI in the Contact Center


Demystifying AI in the Contact Center

Pop culture has, for better or worse, played an outsized role in how modern professionals understand artificial intelligence (AI). For those of us who grew up on “Star Wars,” the ever-present and droids C-3PO and R2D2 painted a picture of quasi-living computers as companions. More recent audiences, meanwhile, might look to Tony Stark’s JARVIS (Just A Rather Very Intelligent System) of the early Marvel Cinematic Universe and view AI as an omnipresent operating system, capable of advanced calculations in a matter of nanoseconds. In reality, these depictions are closer to fiction than they are science.

That doesn’t mean, however, that there isn’t some kernel of truth rooted in C-3PO or JARVIS. Across nearly every industry, especially customer service, AI is having a real and meaningful impact. For example, research firm IDC found that early adopters of customer experience (CX) centric AI have seen a 25% increase across core performance metrics such as customer satisfaction and cost of service. The key to capitalizing on the promise of AI is understanding that AI isn’t one monolithic piece of software, but instead represents a family of approaches and capabilities that serve distinct purposes. Different combinations of components like natural language processing (NLP), machine learning (ML) and automation come together to bring more common examples of AI-powered tools, like chatbots, to life.

If you’re thinking about deploying AI in your contact center, the best place to start is demystifying these distinct flavors of AI and determining which approach is needed to achieve your business goals. That way, you can make the most informed decision to strategically improve your contact center. The following is a brief primer that should get you started.

Natural Language Processing

NLP is one of the most common elements of AI. It’s a field of linguistic-based computer science that uses AI technology to help make language logical so it could be understood by computers in the same way our brains interpret words and phrases. NLP helps computers process and analyze text and speech and to respond accurately and naturally, breaking text into pieces and processing it to identify and tag attributes, such as signifying keywords, categories or sentiment.

Through this process, contact center tools like chatbots are able to engage, understand and respond to customers, and the most effective NLP also includes soft interpretations such as sarcasm, tone and emotions. For example, verbally, a simple statement like “great, thanks” can have multiple meanings depending on the tone used, which doesn’t necessarily translate via text. Highly tuned NLP uses context clues from throughout the conversation to indicate whether the statement is positive as written, or has another meaning entirely. This can be equally helpful in the area of live agent coaching, as AI assesses conversation in real time.

That said, just because a bot can understand direct questions and commands, relying on NLP alone can create problems when conversations go “off script.” When combined with procedural “if-then” scripting, NLP can be extremely useful in cases where the conversation is linear, such as automated FAQ (frequently asked questions) responses. However, when the conversation veers outside of its programmed responses, these simple bots can falter. Unless, of course, NLP is partnered with the capability to learn and adapt in real time.

Machine Learning

The near-human-like ability for an AI-powered virtual assistant to be truly conversational, understand nuances of human emotion, and successfully handle topics “off script” is underpinned by machine learning (ML). Machine learning involves consuming massive amounts of data with both good and bad outcomes to find patterns in data, using statistical models to predict outcomes without being explicitly programmed to do so. Through an iterative process, programs can learn from the data to improve accuracy of predictions over time. It’s the technological equivalent of early humans putting their hands above fire and realizing that they shouldn’t do that a second time.

Machine learning involves consuming massive amounts of data with both good and bad outcomes to find patterns in data, using statistical models to predict outcomes without being explicitly programmed to do so.

Machine learning provides the foundation for a truly conversational bot versus more scripted automated responses that follow a specific path. Through this flexibility and adaptation, AI can achieve both improvement and consistency, particularly in familiar tools such as the virtual agent or chatbot. In fact, it is an essential ingredient for agent assistance because of how it aggregates and understands context and topics that aren’t pre-scripted. In an environment where contact centers are automating the routine with self-service bots, there is increasing pressure for customer service agents to perform at higher levels while tackling ever more challenging issues. Using machine learning-powered AI that can understand human-to-human connections and can coach agents in real time on the behavioral soft skills that lead to happy customers is a true game changer.


This, at the core, represents a fundamental value proposition of AI: transitioning routine or mundane tasks to robots so that human staff can perform more specialized and complex tasks. In the contact center, this is known as robotic process automation (RPA), and is an example of automation where humans create or use computer programs to perform repetitive human tasks using rules or decision trees.

After-call or back-end contact center work is ripe for automation. RPAs can automate most after-contact agent work and intraday supervisor activities. Further, they can integrate data, applications and workflow from automating mouse clicks to filling in fields. From an agent management standpoint, this enables faster feedback aggregation and coaching, allowing for more meaningful interactions and activities. Depending on the specific use case, RPAs can be both attended by agents or unattended—i.e., set and forget—so contact centers can more effectively allocate resources. As a general rule, unattended RPAs are more scripted and procedural, while attended automation require at least some degree of machine learning to intelligently understand how best to help agents during their active conversations with customers.

Tip of the Iceberg

While these are just some examples, as a whole, AI is a deeply complex system that can have a transformational effect on how contact centers build stronger, more authentic connections with customers. When you have the right AI tools and know how to use them, AI looks a lot less like the unemotional supercomputer from 1983’s “WarGames,” and instead becomes an indispensable ally for both customers and agents to improve service and human-to-human connections. That’s a game where every player is a winner.