2019 Will See A 'Gold Rush' In Customer AI Adoption2019 Will See A 'Gold Rush' In Customer AI Adoption
2019 Will See A 'Gold Rush' In Customer AI Adoption
December 5, 2018
SAN FRANCISCO - In the digital era, customer expectations are shifting fast. Companies are facing greater demand than ever before for fast, intuitive service across a multiplicity of platforms. This is understandable, considering that the average person can expect to spend up to 43 days on hold with call centres across their lifetime.
Abinash Tripathy is the founder and Chief Strategy Officer for Helpshift, an intelligent customer service platform which helps businesses and enterprises alike deliver seamless, AI-powered customer messaging services. Ahead of The AI Summit New York, we sat down with Abinash to discuss the challenges of customer service today - and how AI bots can really help.
What does building a superior customer experience mean in the context of AI?
If you look at the customer experience today, a lot of it boils down to how customers are interacting with the brand. Customer service contact centers happen to be the number one place these conversations are happening, with humans talking to other humans.
This isn’t a very scalable model, particularly in B2C businesses with millions of customers. You can’t just put millions of people in call centers to talk to these people. Consequently, communications with brands is hugely frustrating to modern consumers because it’s slow. Everything is being processed by human beings and there’s very little self-service, which is why the cost scale of customer service is so high. Brands don’t want to invest in building large, human-powered contact centers, so consumers are subjected to waits and holds and all sorts of frustrating.
What we are trying to do is move all these conversations into a highly dynamic medium which all of us today are used to. We largely communicate with one another through channels like iMessage or WhatsApp. Messaging is therefore the first place I believe AI will proliferate in the enterprise, because once the medium moves to chat, you’re able to deploy bots and it becomes the vehicle for all kinds of intelligent algorithms. From there, you can scale the conversations between consumers and brands using technology—and AI happens to be a significant part of that stack.
Obviously, there’s problems with overreliance on human-run contact centers. But are there any limitations specific to scaling conversational AI?
Most of the first generation of conversational AI technology has not really worked. Progress hasn’t been tempered by the reality of where the technology is at, and I think people got over-enthusiastic. AI cannot sustain the kind of long conversation that you and I are having right now—it just has a very limited understanding of the world, of context, and of human emotions. It’s therefore unreasonable to expect an AI to have a free-flowing conversation with the human being.
However, AI can do some things really well. For example, it can try to extract the intent of what the customer is saying. We can use AI very practically to do simple things like that using an automated bot capability that might not be that intelligent, but is text literate.
What are the bottlenecks in improving that experience then? Do you see any frustration from users?
If you look at customer service as a vertical, a good 70 – 80% of customer transactions in most industries are repetitive questions about things that have been solved before. AI is perfectly suited to scaling that. You can serve queries where the answers are well-known through AI. It can play a role in servicing well-known answers then passing on things that are new and unknown to human beings—that’s what AI is good at.
Companies should then focus on the things that drive most of the call volumes and train AI specifically to handle those so that human operators are not burdened by those common, repetitive questions. Then, anything that is new, unknown, or complex should be passed off immediately to be processed by human staff.
Why can’t bots handle those complex queries? What is needed to get to that level?
If you really think about it, AI needs some prior knowledge in order to be successful. Most complex problems are new and those problems haven’t been heard before, so the AIs aren’t trained to handle them. The very fact that we call it machine learning shows you need a corpus of knowledge that the AI can operate on and in most of these cases that knowledge doesn’t exist—hence the need for human involvement.
"Why do we force human efficiencies into any system in the world? We don’t need to."
I think we will see more and more conversational traffic going towards AIs, but you will always need AI to be backed up by humans. The role of those humans might be very different—they will be specialists that, in solving these problems, will be training the AI.
What does a strong ROI look like for these technologies in a customer context?
Call centres worldwide today rely on large outsourcing outfits in the Phillipines and India in order to handle very low-level enquiries. The first place AI will have an impact will be automating the low level enquiries—the easy ones. These companies can now get rid of their outsourcing partners in Manila and India, i.e. places where cheap labour is used to basically do what AI can do. 30% of contact centre volume is driven off of these low-cost outsourced operations.
Once these are automated, what you’ll see is that the workforce captive to the brand will also start to reduce quite a bit. As the AIs get better, I’m expecting a 30-40% reduction in contact centre workforce. If you look at our deployment this year across our customer base, our customers are already seeing the potential to reduce their human workforce by 10 to 30 percent.
Do you think that’s a good thing?
It’s debatable. I’m personally for it, because when you think about it, why do we force human efficiencies into any system in the world? We don’t need to. Humans can be relieved of the mundane and work on creative pursuits. There’ll be a large portion of the population that will not have work because AI will start to automate most of what humans are doing.
It’s happened before in the industrial era—most of the factories today are just run by robots and the humans are supervisors. Just look at Ford factories in the 1930s versus today—today, they’re heavily automated, and there’s just so few humans working on those assembly lines.
So what does this look like in action for a company moving from a legacy contact center to making AI operational? What advice would you give to people starting out on this journey?
For a company based in Silicon Valley like us, we can hire data scientists and AI engineers and put solutions together. For most companies out there, it’s going to be really hard to build that capability in-house. My advice is that if your core competence is not technology, then you want to work with the people who have solved that problem many times over and over for other people.
Today, what I see is that brands are confused, they’re trying to boil the ocean. That’s the wrong approach. The best thing to do is to look at things that can be easily automated, the things that are driving volume, automate those away. Even if it’s with a simple bot technology, they can start to see a lot of ROI in the short term.
How do you see the landscape changing in 2019?
I think there’ll be wide recognition that AI is going to have a lot of impact on the top and bottom lines. There’s going to be a goldrush where companies start seeing practical use of AI. Then you’re going to start seeing lots of success stories and case studies come out of it—and a rush by brands to try and adopt.