One of the fastest-growing parts of Amazon Web Services, by revenue, is its Connect offering, which helps customers run their call centers. Launched in 2017, Connect is run by AWS Vice President Pasquale DeMaio, and the companies using it to manage their voice, email, and chat support communications include Capital One, Saks Fifth Avenue, and Best Western Hotels & Resorts.
We recently spoke with DiMaio about building and integrating AI features into Connect. DiMaio emphasized the importance of focusing on business objectives throughout the process. Generative AI shouldn’t be treated as “some magic elixir that just fixes everything,” he said, “but instead part of a business goal and a customer outcome goal.” And, he adds, it needs to be safe, ethical, and “deliver disproportionately valuable outcomes.”
• • •
You’ve been involved in Amazon Connect since almost the beginning. How has the technological vision evolved from founding to now — and perhaps even into the future?
From day one, we knew that AI would play an important role. … We just released a new generation of Amazon Connect, and by default now we turn on every AI integration. Whether it be text or voice or email, we want to always make sure we’re engaging with the customer where they want to be met, and using AI to augment that and make it better every step of the way.
You mentioned something I want to tap into next. Amazon Connect recently announced unlimited AI capabilities as part of its standard pricing. Can you tell me more about this decision—what are the logistics of the new model and what led to the shift?
It has always been our belief that AI was massively beneficial to customers, but it also is one of those things where, until you can deliver it end to end, it’s hard to argue that we would want to turn it on by default. Now we believe that each part of it is going to actually be accretive to the other parts.
It’s not just that you have AI in one part — maybe I can automate a password reset or something — I mean, customers love that. But the thing that’s really magic for them is if I learn from what happened early in the conversation, I pass all that information on to the agent. Then, while the agent is interacting with the customer, I’m also suggesting solutions to the problem. The agent is not alt tabbing. After it’s over, we take all the information that we learned every step of the way and wrap that up into a nice bow to say, this is what really occurred with the customer. And then we can even look further down the line and say, what happened with this customer the next day, the next week, and bring all that together into one place.
With the new version of Amazon Connect, each part is working as a broader offering in a really cohesive way. It’s quite a bit different than having people go and buy each of the capabilities from another vendor. You might find someone who could check a lot of the boxes, but having it all in one solution means that it works together seamlessly. The data is understood the same way, the outcomes are measurable. That’s a big change.
The problem for customers has been implementation, quality and that the cost was just all over the place. With one fell swoop, we can make it incredibly easy to implement and configure. We can make it so the pricing becomes an afterthought, because we package it up in a very easy-to-consume way that is aggressively priced because we’re not tied to some weird legacy construct around seats plus allotments and blah, blah, blah. That takes out all these blockers.
Looking forward, how do you anticipate this shift will evolve over time, and how will it grow the business in the future?
My belief is a bunch of things are going to happen in the next two years that we want to usher in. One of them is that you’ll see generative AI just become AI again, and then it just becomes part of the scenery at some point.
Every part of Connect, because it’s just built in, you don’t have to do integrations and maintain them. This just becomes the default experience for someone who’s using Amazon Connect. I think folks will eventually catch up with that, but we wanted to be leading that charge.
Would you agree that the customer service workforce is seeing the biggest tangible change in this AI era?
It should be in theory… I don’t know that the advances are happening as fast as people want, because I think a lot of folks have to be very careful on how they implement it.
A lot of folks are saying, ‘Well, magic occurs, and AI does everything for you.’ I don’t think that’s how people who are running mission-critical businesses think about it.
With hallucinations, for example, if that happens in the middle of a conversation with an AI bot, it could be disastrous for your company. It could mislead the customer, transfer funds or do something crazy. A lot of folks are saying, “Well, magic occurs, and AI does everything for you.” I don’t think that’s how people who are running mission-critical businesses think about it. They think about their business and want to work back from the outcomes.
One of the big differences in the way we approach this is we let you use different types of AI and different implementations for different parts of the experience. The example I always like to use is, if you’re doing something like a big bank transfer, AI can make that a better experience. Like, “It sounds like you’re trying to do a bank transfer. Let me step you through that. First, tell me how much, tell me what account. Let me confirm that exact thing with you. I’m going to record everything we’ve done here, and before we go any further, please confirm this is exactly right.”
Or, “I’m starting a job next Monday, and I want to know what the policy on casual Friday is.” Now there’s a great Gen AI [use case].
In between that, there’s a whole bunch of opportunity to do things like use Gen AI to help understand intents. A lot of times, you’ll have an experience where someone will say, “I’m trying to do a balance transfer, but my address was wrong on the website, and when I tried to change it, the balance transfer wouldn’t go through.” That’s a very hard thing for legacy AI to understand, but you can use generative AI to interpret that.
The other thing we work really hard with customers to do is to get ahead of the problems before they even are problems. For example, if your flight’s been cancelled and you call up an airline and they say to you, “How can I help you?” Your answer is going to be, “Operator, operator.” You’re stressed out. [What if instead] it says, “I see your flight’s been canceled. Can I book you on the next one tonight?” Those proactive things are great.
Now let me take another step and say, now you’re worried that your rental car is going to be given away. You’re now thinking, I’ve got to call the rental car place and make sure the car is still there. What if instead, the rental car company just sees that your flight’s been delayed, which, by the way, they can know very easily, and says, “Your car is still waiting for you. You can pick it up when you land. We know you’re running late.” That just fundamentally changes the experience in a way that is better for everyone involved. There’s no more stressing out. There’s no waiting on hold. There’s no expense of having an agent just to say everything is okay. All of that just goes away. We want to make sure we’re bringing value at every step of the way, far beyond just being reactive. We actually include artificial intelligence to help predict some of these things and to create segmentation, so people understand customers before they even run into a challenge.
We’re not just trying to say ‘Gen AI, Gen AI, Gen AI,’ but instead, let’s bring you a solution that really works for what your customer goals are.
…But I think overall, the industry has not caught up to saying, “We need to use this not as some magic elixir that just fixes everything, but instead part of a business goal and a customer outcome goal.” Now, we’re bringing this to a place where people who are actually trying to achieve goals can do so in ways that are valuable, safe, ethical and deliver disproportionately valuable outcomes. We’re not just trying to say ‘Gen AI, Gen AI, Gen AI,’ but instead, let’s bring you a solution that really works for what your customer goals are.
Amazon Connect is one of the fastest growing AWS services by revenue. Can you tell me how technological innovation helps you maintain momentum for continued growth?
AI has been an important part of the story since [the beginning]. From day one, it still provided real-world value, but it was limited to mostly automating basic tasks for people and saving some people some time.
At this point, the game-changing aspect of having it at every interaction means that if you don’t start thinking about this now, in two years, you’re going to look back and you’re going to be two years behind the curve of where you want to be, where your competitors are with how they are leveraging it to make automated experiences better, make agents superhuman and help them be proficient from day one. These are the kinds of things that you just can’t afford to not be great at, because people’s expectations for customer service rise every year.
There was one word you said in there that I jotted down, and that was “superhuman.” You’re placing that emphasis back on the human. With AI optimizing the workflow for these agents, how are you building into your products the ability for them to retain agency over what goes on?
Almost every person you talk to who became an agent didn’t do it just to make a fast buck. They actually really enjoy empathizing. I’ve watched some of these folks, and it’s unbelievable to see them interact with people, but they can’t do a good job of that if they’re busy fighting the technology. A huge part of what we do with Connect is try to make the technology receive and offer them things when they need it, and they can choose to leverage it. They can choose to change it if they want to. We use Gen AI to make the information more human-readable. They could read it aloud to the customer if they wanted to. I don’t necessarily encourage that, but it helps them walk through that in a way that is much more natural. If it’s your first day on the job, it’s great to have that. If it’s your 200th day on the job, or fifth year on the job, it’s nice to have the technology working on your behalf so you’re not spending all your time alt-tabbing, but instead focusing on the human on the other end of the phone.
If you are a company, this may be the only time you interact with this human being in a real way. They might be using your web page for everything they do, and at that point, do you have a relationship with the person? Many of my customers work really hard to say, “This is a moment in time where it can make the difference between the person feeling like this is a great company to do business with, or this company doesn’t care about me, and it’s just trying to do things as cheaply and as fast as possible, with no regard to our to our relationship.”
As someone who was double-charged on my phone bill last month and had to spend 45 minutes resolving that, I can definitely relate to that. You’re really talking about an increase in the quality of these interactions. While you know your technology contributes to productivity — a 60 percent reduction in call volume and 50 percent reduction in agent training time — how do you help measure quality?
A couple things I encourage my customers to do are to take a step back and not look at an individual contact or interaction as the end-all-be-all of what happened, but instead say, “What was the value to this customer after that?” If you called up to cancel a product, and then the person convinces you not to, because they make you feel bad about it, you’re still going to call back the next day and cancel. You’re not going to do it with that person. All you really did was make it twice as expensive to let this person cancel. Let’s take a step back and decide what your real business outcomes you want to measure are.
In the case where something goes wrong, the business outcome you want to measure is: Did you feel listened to? Do you feel like you got it fixed right away? But also, do you feel like you want to keep doing business there?
From the one outcome, they would say, “We succeeded. We got the person off the phone, and they’re still a customer.” But now we want to start predicting, is this indicative of something? Are customers more likely to leave when they have these experiences? We should be reaching out to you and helping make that right in ways to make you feel valued. Maybe we give you a discount.
By the way, they don’t want anyone waiting on the phone for 45 minutes either. We can give better customer service, we can do it cheaper, and we can do it with better outcomes if we just take a step back and learn to use the information we have.
Can you talk about your approach to building new features? How does the team actually go from ideation to external-facing product?
Our feature development starts with customer obsession: we listen deeply to customers and work backwards from there. We’re lucky to have a large cohort of internal customers (such as Amazon Customer Service, Audible, or AWS support) and external customers, (like Doordash, Hilton, or CapitalOne), who are eager to partner with us to innovate on behalf of real customers.
While we’re enthusiastic about AI’s potential, we’re also disciplined about its application, incorporating it where it delivers meaningful value. If a solution works efficiently without AI, we don’t try to shoehorn it in.
