Prehistory of Machine Learning in Call Centers

It All Started with IVR

Machine-assisted support emerged from the primordial soup of 1200 baud modems and cassette tape answering machines. Interactive voice response (IVR) started back in the dim, dark days of customer support with byzantine menus disconnected from databases of any kind. These menus rightly became objects of derision by consumers who plodded through innumerable queries answered with keypad activation. Usually, a response triggered a fresh menu to start the process over again. Later, voice-activated menus augmented keypads as the technology improved.

IVR never really caught on for users, but firms with large support appetites like telcos, banks, and airlines embraced the chance to make support self-service and trim down payroll. Often as not, customers dialed zero for the live agent option.

It was a dark, myopic era when cost reduction trumped customer experience.

The Case for Machine Support

If the technology didn’t pace with the aspiration, the idea still had merit beyond simple cost savings. Automated support doesn’t suffer from attrition, staffing shortfalls, or incompetent service. The idea that a person could call and quickly get to the right agent made sense. And, with skillful menu design, the IVR navigation could be relatively painless.

Maybe the service wasn’t all that great, but it was consistent, and it did siphon off some percentage of inquiries. The idea that you can hand off routine questions and focus on the heavy lifting of customer support is a holy grail in customer service circles. Unlike automated support’s evil twin, the annoying and intrusive robocalling, IVR was just a tiresome process of repeating numbers slow and loud, playing memory games — all before punting to speak to a live person.

Technology Improves, Self-service Matures, Expectations Rise

Machine Learning Takes Shape

Fast-forward to an Alexa-driven world where customers are well-versed in machine-assisted everything. Now, call center automation goes by the much sexier moniker of Artificial Intelligence.

Big BPOs promote automation because it’s a great story and because they have to. Major clients demand automation for their large programs. Intelligent systems taught through thousands of repetitions, tied to a customer database, and escalating to a cadre of super-agents are irresistible to vendor managers and CFOs. Where call center AI obtains is in captive centers. Here, cost gains could intervene to help companies avoid outsourcing altogether, keep native agents, and keep scaling if necessary.

And call center automation is viable and well-received (if executed well). Smart bots interface securely accessed customer data submitted through apps, websites, and phone calls. AI pioneered by big techs like Amazon, Google, IBM, Microsoft, and Apple nudges us ever closer to the singularity. Google has open-sourced its platform, Tensor Flow. Likewise, for IBM with Apache SystemML and Apple with Apple Core ML. Other variants abound.

Customer Support Got Better and Better

Another thing happened — consumers became accustomed to quality support across channels like smartphone apps and social media along with more traditional email, chat, text, and phone inquiries. Now we talk about customer support as customer experience (CX) and interactions triggering customer delight.

Those terms map prevailing expectations for excellent, seamless support. Keep customers happy to keep customers. Keep customers and harvest lifetime value. Companies like Zappos led the charge, famously espousing customer service as a core competency.

Self-service Becomes Smart-service

Simultaneously, cheap technology dropped barriers to entry and introduced a generation of nimble businesses zipping around their Jurassic counterparts. Those agile players focused on predictive, friendly, helpful, fast support. There wasn’t much choice once changing vendors became a matter of a few clicks.

The new rules mean customer service is the name of the game. Millennials and Generation Z-ers (Zoomers), steeped in online services, look for quick answers to questions. Robust self-service arrives post-Web 2.0, where data services and mobile access and security intersect.

Consumers want fast, brilliant support. Now, they expect to reserve live agent interactions for complex or exceptional scenarios like one-off requests, purchasing clarifications, complicated billing issues, or specific technical questions. Note Chatbots Magazine’s comment from way back in 2017, “67% of people expect to use messaging apps to talk to businesses….”

Pundits Agree

One industry report predicts call-center AI technology will more than triple in the coming years: from $800 million in 2019 to $2.8 billion by 2024. Forbes Research quoted the same statistic in an article highlighting Humana’s adoption of IBM’s platform where they observe, “AI-enabled conversational agents, for example, are expected to handle 20% of all customer service requests by 2022.”

Call Center AI, or robotic process automation (RPA), is built to handle extensive data processing since that’s how machine learning works. By processing documents, calls, texts, emails, or any other data set, the software can apply rules and identify patterns for suitable action. In essence, mountains of experiential data from different sources can be sifted into a single view.

As TechRepublic notes:

“customer service agents are beginning to get help with some of the intangibles of customer calls--like when a customer’s frustrations rise and they begin to raise their voice or when there are long pauses in the conversation that could indicate rising anger. Because the AI has been trained to operate in multilingual and cultural contexts, it can also be deployed in countries that have different linguistic and cultural styles that can influence whether anger or delight is being felt by a customer.”

If the tech press and market researchers agree on the rise of call center and BPO Artificial Intelligence, they do not agree on its implication for bastions of calls center work like the Philippines. In March 2021, Bloomberg published the apocalyptically named; Empathetic Robots Are Killing Off the World’s Call-Center Industry. Its authors predict a 9% decline in human call center work as a “bottom line” conclusion.

There is plenty of hubbub about upskilling workers to move up the customer service chain as they retreat from more mundane inquiries. Often the discussion descends to the assertion that machine-assisted technologies or empathetic robots can’t match empathetic humans for empathy and engagement.

Conversely, I’m sure everyone can recall a support call where human empathy didn’t quite measure up to a quality customer experience or a satisfying outcome. No doubt, large programs with multiple queues will subtract scores of agents with automation. But that’s not the whole picture.

Democratizing Automation & Exploding Business Models

Referencing Moore’s Law, AI technologies for data processing and categorization will only become better and more accessible. The unrelenting momentum toward cheaper and faster processing drove the necessary conditions for AI development in the first place.

The same thing happened with telecommunications platforms in the call center industry in the 2010s.

Seat-based pricing with incremental VoIP fees replaced larger outbound and customer service platforms that required significant investments in local data centers, dedicated bandwidth, and specialized IT personnel.

For a slight performance sacrifice, boutique call centers and BPOs (like Lambent) could offer all the services associated with those large telco platforms or dialers. Skills-based routing, granular reporting, predictive and preview dialing moved from costly premium services managed by an arcane class of database and VoIP technicians to a monthly pay-as-you-go service managed by operations personnel.

It looks like AI is on the same path. And that opens up automation and augmented agents to small and medium-sized businesses. These companies leverage managed back-office services where an offshore subject matter expert, working behind the scenes like the Mighty Oz, can field escalated support queries, verify resolutions, and keep building the self-service support features for small brands.

This trend closely tacks alongside new web and app services built on similar trends in other technologies (and machine learning). Even better, machine learning offers blended agent automation solutions for different industries.

Online Travel Platform Automation

Referencing this excellent explanation of AI and machine learning applied to call centers in the Medium post, AI and Machine Learning to Improve Customer Contact Experience, by Ricardo Balduino, let’s develop a scenario for online travel support.

This is especially compelling since online travel agencies or OTAs operate at slim margins, and they often cut corners on quality customer experience. There are a couple of scenarios worth noting.

Outbound Notifications for Travelers

Travel plans can change drastically based on airline changes and travel rules to specific destinations. Automation can accomplish proactive notifications for travelers for these changes and offer structured options for rebooking or further information from a live operator.

Travel businesses can offer this service a premium add-on or simply bundle it with a reservation to improve the overall customer experience.

Outbound notifications keep the traveler informed in near-real-time and head off additional calls for rebooking information and cancellation work. More importantly, this keeps the traveler informed and improves the entire experience — even when unexpected changes arise.

Inbound Support for Travelers

On the flip side, inbound support offers all the benefits we ascribe to artificial intelligence in the call center. People call for routine inquiries or to request a process. Once callers confirm their identity, speech recognition, trained across hours of processing to recognize most natural language queries, handles their request.

If the AI platform encounters a scenario that it cannot address, like an unauthorized cancellation or a refund request, it simply places the call, chat, or email into the queue for a live agent.

Machine Learning Shifts the BPO and Call Center Relationship with Data and Process

The benefits extend beyond smoother, shorter handle times and condensed payroll for automate-able tasks. Machine learning fundamentally changes a business’s relationship with data. As Benedict Evans observed on his podcast,'Digital Transformation' - beyond the silly slogan, machine learning allows you to ask questions about unstructured data.

From a call center perspective, unstructured data, like customer sentiment, can be added to call data and associated with call transcripts to give customer service managers greater insight into the call dynamics. Essentially, agents handle escalations and help train the platform for critical metrics.

The other aspect of call center AI is the level of detail required to deploy solutions. A program launch means subjecting the associated process flow to a deeper level of analysis. This roughly mirrors the shift to data understanding that arose from the profusion of data points that large telephony platforms enabled through predictive dialing and inbound support linked to customer databases.

Inbound call scenarios now need to be assessed based on the complexity of the call intention. Customer service teams then develop process flows around natural language input and inflection points where customers need a live agent on the phone, behind the chat interface, and responding to a help desk ticket. Moreover, automated data collection is tagged with additional information using human input to accelerate the machine learning process.