The promises and perils of using AI in contact centers

4 min read
call center
Image credit: Depositphotos.com

Over the past four years, the number of enterprises implementing artificial intelligence technologies has increased by 270 percent, tripling in the past year alone. IDC is estimating the worldwide spend on cognitive and AI systems to reach $77.6 billion in 2022, more than three times what was forecast in 2018. There’s no doubt that interest in AI is not slowing down, and businesses are more willing to put money behind this than ever before.

Considering that 90 percent of IT and business leaders see AI as a crucial element in the digital transformation of their organization, and 94 percent recognize the ability of AI to transform the performance of their contact center, it’s not surprising to see this uptick in adoption. However, 47 percent of these IT and business leaders also feel that their organization is not able to effectively use AI, and 92 percent concede that their business has work to do to get the most of the AI solutions they already have in place.

It’s clear that while many want AI, the expectations for the technology are often falling short. In the contact center this is no exception. While we’ve all noticed how customer service has gotten smarter over the past few years, with chatbots, call scheduling and quicker verification processes, the claims  from vendors supplying this technology have not.

According to some, AI can quantify the tone of your voice, the sentiment of your words, and your personality, all within a couple of sentences.  To others, AI is a brand-new employee that can automate the entire job of a call center agent and learn everything it needs to know from a limited data set.

Enterprises looking to implement AI in their workplace and engage with new vendors need to exercise caution, and level-set the expectations of stakeholders to ensure AI’s impact is not being overhyped.

When bringing this technology into the contact center, agents can often be overlooked as part of the customer solution, with AI taking center stage. For example, we may know the caller visited the website for the new iPhone XR, XS, and XS Max before checking their upgrade eligibility. We know they have used iPhones for years and have an iPad and AppleWatch on their account. This information suggests a strong likelihood that this customer is not only interested in upgrading to a new device, but they’d be interested in an Apple device.

When this customer calls into most vendor’s AI-enabled contact center solutions, their call is queued up and delivered to the first available agent in the order in which it was received (First in, First Out or FIFO). But, what if the next available agent uses a Samsung Galaxy S10, and the agent right behind them owns an iPhone, AppleWatch, iPad and just about every other Apple device on the market?

How can an agent who is passionate about Android devices deliver the same experience to a customer who is just as passionate about their Apple devices? When given an agent with the wrong background, the customer may not appreciate the value of the more expensive iPhone XS compared to the XR, and as a result, end up with a device that doesn’t fit their needs over the longer term.

This interaction could potentially lead to a frustrated customer as they learn the limitations of the device. At the same time, the degraded experience will decrease revenue. When your infrastructure only focuses on delivering customers and their data sets to the next available agent in a FIFO model, you can’t create optimal pairing.

The truth is, AI is only as good as the data behind it. During a demo of IBM Watson I was struck by a statement made by our friends at IBM: “Most products perform best the day they’re taken out of the box and over the course of the product’s life-cycle the experience degrades through use. With artificial intelligence solutions, such as Watson, the opposite holds true. As more data is consumed the solutions become stronger and stronger.”

They were describing their approach for isolating data between Watson customers. Without a central repository of data, each customer’s instance of Watson must learn as it ages. Sharing a data repository would seem to add efficiencies but it creates a host of other challenges around privacy, competitive intelligence, and data ownership questions.

call center artificial intelligence

In the contact center this dynamic speaks volumes. The idea that AI could learn from customer requests, agent responses and overall outcomes is a powerful one, and with a central repository of information to mine, this hypothetical AI solution would be running at a high efficiency from day one of its implementation.

The reality, however, is far different.

While many may be sold on the power and intuitiveness of AI, these solutions will realistically need hundreds of customer-service cycle iterations to learn from before being able to behave in any autonomous function. When adopting this technology, many enterprises may be underwhelmed at first, and it’s important to set these expectations from the get-go, and not buy into the “hype” surrounding any particular product.

In its current state, AI is simply not mature enough to replace a human being, despite 39 percent of contact center IT and business leaders fearing that AI may replace their jobs, according to an Avaya VansonBourne Survey.  While speech response solutions, neuro-linguistic programming, machine learning, and other advancements continue to mature by orders of magnitude, it’s not often difficult to realize you’re interacting with an automated system.

The state of technology isn’t even the biggest threat, in my opinion, to mass adoption of human replacement AI—legislation is.

New legislation focused on the ethics of artificial intelligence, while good-intentioned, threatens to dull the impact of automated solutions. We know callers already avoid interacting with automated attendants and speech response systems, mashing the zero key or reciting phrases like “operator,” “receptionist,” “agent,” or whatever keyword ends the misery of using automated solutions. Assuming artificial intelligence solutions mature to the point of imperceptibility, an announcement prefixed to any human facing interaction would likely result in call disconnects or caller frustration trying to bypass the automated tool.

To borrow from the Avengers movies, I believe an AI strategy should focus less on building Ultron and more on creating J.A.R.V.I.S. In other words, focus more on augmentation and less on automation.

Too many businesses are looking at AI as the be all, end all, when it comes to optimizing operations and meeting customer expectations. The reality is that AI is not smart enough to act on its own to solve problems, it isn’t a new employee and shouldn’t be treated like one. As businesses look to implement AI into their operations, as they should in any digital transformation strategy, it should be treated as a new tool, something to work with, improve and refine as time goes on. AI may not make a huge impact the day you install it, but it might in a year or two’s time. The more data you feed it, the stronger it will become, not strong enough to take your job, but definitely smart enough to make it easier.

Advertisements

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.