Centaur Solutions For Customer Support: Amplifying AI Efficiency with Human Empathy
While renewed interest in AI has driven extreme excitement around applying artificial intelligence to everything, the more practical applications which would dramatically enhance our everyday lives right now are often overlooked.
I had a recent customer support experience (nightmare?) which vividly illustrates the need for implementing more centaur solutions: automation/AI enhanced by humans, and vice versa. AI can help both the humans offering customer support and the people receiving the customer support. Here’s how:
When an automated system fails, the first thing everyone looks for is a human.
This June, I found myself fresh off a plane from New York to London. Jet-lagged and groggy, I was on my way to a conference on automation in Germany, which had me go through the connecting flights section of London Heathrow Airport. When I got to the connecting flight zone (now completely automated with ticket scanning machines that scan phone QR codes and paper tickets), my ticket wouldn’t scan. I couldn’t go through the gate.
I was stuck. Because I didn’t have a ticket, I couldn’t move forward through security, but I also couldn’t exit the airport. I was confined to Heathrow’s Connections Hallway (a liminal space betwixt and between the here and the there).
There was a chair to sit on, which was a relief, because I was about to be stuck there for a long time.
When I finally got signal on my phone, I found the problem. My connecting flight on British Airways was canceled. When I asked one of the connections agents, they couldn’t help me get a new ticket. “How can I get one then?”, I asked. They told me to follow the instructions from the airline website.
The wifi and signal in the Connecting Flight Zone was barely present, making it almost impossible to get the email from my phone for the instructions on the flight cancellation. After waiting a few minutes for it to load, I was told to pick a new flight, and given a link to click on.
This link was so filled with javascript and other heavy items that it wouldn’t load fully. Each page took a couple minutes to process. The same issue happened with my laptop, which I had to drag out of my carry-on bag. After ten minutes and a lot of clicking, the ticket screen finally loaded.
The website informed me that I couldn’t process my request online, and that I needed to call the airline’s customer support.
Oh no.
Trying to keep my composure, I called the airline’s helpline. After a brief wait, an automated message told me something to the effect of: “British Airways isn’t able to offer support over the phone at this time, please use the web Interface”.
I was caught in an ouroboros. I couldn’t call or use the website. And I was confined to a small chair in the middle of a transit zone. People whooshed past. I sat still, bags at my side, laptop balanced on my knees.
I was trapped in a problem similar to the one Nick Carr warned us nearly a decade ago: The Great Forgetting, where human knowledge is gradually lost in the morass of automated systems. But in this case, it was human care that was lost.
I had been sitting for twenty minutes at this point, and the checkpoint staff were beginning to get concerned. They knew I was stuck, but they didn’t know quite how stuck I was. I explained the loop to them, with as much grace as I could muster.
Finally, one of the connecting agents told me, “I’m not supposed to do this, but I’m going to help you”. After another 15 minutes, he brought me a boarding pass that would allow me to get through the connecting flights system.
It was only then that I realized that the flight he booked me on was 12 hours later, at another airport, with a 3 hour layover in Amsterdam, arriving at Midnight in Germany. I didn’t have a choice. My trip to an automation conference was ironically delayed by an automation system.
The trouble with cutting corners around customer support
At heart, the problem I experienced has to do with gaps and failures in customer service support systems. We experience this almost every day in travel, health care, technical support, and internal documentation within large organizations. They all depend on massive support structures that have often existed for decades, but many of them don’t have a way to send information to each other, ensure redundancy — i.e., that there is at least one backup system to use in case the primary system doesn’t work, like a phone call or text message service.
And these customer support requests grow larger each year. At the same time, management is constantly looking for ways to cut costs in this area. So the customer service support staff for publicly traded companies, who are responsible for growing the company, year over year, often take the hit.
This is precisely the kind of problem artificial intelligence was originally designed to solve — or as I put it last year, nerdy automation tasks. But ask the average AI enthusiast what the most powerful use case of AI could be, and it’s unlikely they’ll include something so unsexy as “using AI to optimize legacy knowledge support systems to smooth the interactions between desperate travelers and airlines”.
But I’d argue that it’s among the very sexiest use cases of artificial intelligence: Leveraging the technology to give us more time to enjoy our human lives. AI can help both the humans offering customer support and the people receiving the customer support.
Here’s how:
Enter the Centaur in the context of Artificial Intelligence
The term Centaur was first used by chess master Garry Kasparov, who argued that a medium-skilled chess player with a good AI system could beat an expert chess player.
In Ralph Losey’s wonderful article about AI and human interaction, he writes that Centaurs:
Symbolize a union of human intellect and animal strength. In AI technology, Centaurs refers to a type of hybrid usage of generative AI that combines human and AI capabilities. It does so by maintaining a clear division of labor between the two, like a centaur’s divided body.
The centaur myth is an invaluable reference when thinking about AI applications. Just as the centaur has both the intelligence and hands of a human combined with the strength and speed of a horse, it’s not about AI “taking over” from humans, but enhancing our capabilities with the power of computing. To use a phrase from Calm Tech, “Technology should amplify the best of technology and the best of humanity. Machines shouldn’t act like humans, and humans shouldn’t act like machines. Amplify the best part of each.”
How might we make a centaur-powered support system for an airline?
At ticket purchase, customers could be given the option of text updates for their flight. These texts, using less bandwidth, would allow stranded travelers to do more with limited connectivity.
When customers text about airline issues, AI could initially handle responses and store context. If no suggested answer fits, the conversation should quickly and seamlessly transfers to a human agent. Provided with full context, this agent continues the text dialogue, ensuring uninterrupted customer support.
Another thing an AI can do well is translation and context. If a customer is super angry and trying to be mean to support staff, the AI can blur the offending words and phrases. It’s very important that the information is not translated or removed. It’s not about censoring — it’s about giving the support team choice. A support agent should always be able to click to uncover words to gain the full context. In this way, customer service agents can choose how much of it they want to see instead of being bombarded with the full text.
The AI could provide an interpretation of the problem, with some suggested phrases and internal system links the support person can use to help the customer. The AI could also bubble up helpful links for the Customer Service agent to use or forward onto the customer. This could save customer service agents time by giving them the link they need, turning customer service agents into more super powered assistants, capable of doing far more than they’re currently allowed.
I imagine customer support agents having their own chatbots for internal use, where if they need a resource or an answer to a problem, they can ask the airline’s own internal knowledge system. This would be especially helpful for new agents who are just learning where the knowledge resources are located.
This system would be helpful for uncovering gaps in support. If no solution is available, customer support agents could be make one, for peer review (much like a wiki). This would help make a system that’s fresh, ever-evolving, and learnable.
By allowing humans and automation to work side by side, automated systems aren’t left to deal with angry customers who really need human support, and humans aren’t bearing the full brunt of all of the customer support calls, especially the angry ones.
On the back end, the system would do four things:
- Check for the most appropriate answer in the database.
- Overflow to human staff when the answer is not present or good enough.
- Capture the human staff’s answer when they produce something novel, and add it to the system.
- Capture the customer’s issues and thoughts for improving the system, so that it can be improved over time
This approach performs “impedance matching” between the asker and the answerer, taking two different systems that need to communicate, constituting a permeable barrier for efficient information flow. It’s calm if and only if it’s good at managing the information flows between these two regimes.
How is a Centaur Support System Calm?
For this technology to be “calm,” it needs to be highly effective at managing information flows, ensuring that the user receives just the right amount of information at the right time and in the right format. This means reducing noise, filtering out unnecessary details, and presenting information in a manner that is easily understood and processed by the human user without causing stress or cognitive overload.
This approach shifts focus from technology as a tool in itself to the interactions between human and tool, using context and fitness to context. It emphasizes the regulation of information flowing across boundaries. This is where the principles of calm technology meet AI.
I’ve seen similar systems built for knowledge management in customer support for web hosting. A good knowledge management system allows users to type in their questions and immediately text-based matches — or if there’s no appropriate entry for “custom domain”, the system can set a flag as a necessary gap in the system. From there, staff can write a support article and publish it to the knowledge base. This process dramatically reduces the time that both humans and customer support agents spend on responding to basic requests, allowing them to focus on more complex issues.
To scale this up for larger organizations, designers could also implement:
- Multiple expert contributors.
- A tiered response system: automated –> first-line support –> expert
- Collaborative editing and peer review processes
- Integration with customer relationship management (CRM) systems
This system is an archetypal example of the “centaur” form of human-AI collaboration, leveraging both human expertise and machine efficiency to create a constantly improving knowledge base.
As for my delayed flight from Heathrow, that particular knowledge support system failure required a 2 hour tube ride to the other side of London to a completely different airport, a 12 hour wait for another plane — and a 6 hour flight with a 3 hour layover. But if I’d been able to load and cancel my flight from text, I could’ve gotten this taken care of in far less time, all while waiting for the plane to taxi at London Heathrow. And once I got off the plane, I would’ve had my new boarding pass in hand, ready for the next leg of the journey.
Support systems should work, even when they fail
People in flight don’t always have access to a phone, laptop, wifi, or cell signal. And when those systems don’t work, it’s always best to have in multiple support solutions available. All airlines have the opportunity to make things better, but no airline should ever support a single resolution system.
When technology focuses on supporting both us and the people who support us, we’ll all have more time for interesting discussions about how technology is changing the world.
Isn’t that the original purpose of technology after all?
—
These are the kinds of problems we’re helping clients to solve at the Calm Tech Institute If you are interested in workshops, consulting, or just running your problem by us, visit us at https://www.calmtech.institute/
