Catching up with AWS’s Pasquale DeMaio on the Next-Gen of Amazon ConnectCatching up with AWS’s Pasquale DeMaio on the Next-Gen of Amazon Connect
Gen AI and Agentic AI are tools companies can, and should, use to improve outcomes for customers—but they’re really just about providing what customers need.
On Monday, March 17, 2025, AWS announced the next generation of Amazon Connect, as well as the general availability of Salesforce Contact Center with Amazon Connect, global integration of the AWS Telephony Network and Amazon Connect’s integration with ChromeOS devices.
The next-gen of Amazon Connect AWS AI offers several new features, including native integration of AWS AI so that an organization can 'click to add' AI capabilities such as self-service, agent assistance, analytics, post-contact evaluation, and automated follow-up. These features are delivered across all channels and include 'future-proof continuous AI updates with all-you-can-eat pricing.'
“Organizations are no longer stuck with price or technology-based tradeoffs. Now they can use whatever technology is the right one because we're giving them all ‘you can eat’ whether that’s automation, agent assist or call wrap-up assist.” said Pasquale DeMaio, VP of Customer Experience Services, AWS. “This is a game changer since companies no longer have to ask questions like how much will it cost, how hard is the integration, will it work, etc. It is just one system, one solution.”
No Jitter spoke with DeMaio at Enterprise Connect about his perspective on how and when generative AI should be used in contact centers. (NJ spoke with DeMaio back in 2023 before Connect introduced Gen AI.)
No Jitter (NJ): What is your overall approach to incorporating generative AI into Amazon Connect?
Pasquale DeMaio (DeMaio): It's a revolutionary technology that will have an evolutionary impact to everything we do. I want generative AI for some hard questions, but I still want to have some [non-Gen AI for] basic interactions. For example, I'd rather type my social security number in with a keypad than use generative AI to deliver that in a conversation. Also, I don't necessarily want generative AI to do a balance transfer for me – writing Python scripts to drive an outcome based on the APIs that are available for doing balance transfers. What I really want is a structured, incredibly reliable, easily observable tool that shows me what happened, why we did it, has the information of what the customer instructed us on, and why the way [it responded was] the right choice.
When I think about more complex or interesting [use cases], I can have it [provide] responses to questions like: ‘I'm starting a new job, what is the dress code for casual Friday?’ You wouldn’t want a structured AI bot to do that.
The same technology we built for agent assistance to help [agents], we’re now expanding to handle self-service experiences. So, for example, if you build a [self-service] experience to be great up front, you'll get customers interacting more with these automated experiences. They'll have more natural conversations, so even if they say something that's a little meandering the generative AI system can understand them and respond and then maybe put the customer back into a structured flow [if that is appropriate].
We just released a new feature where you can add guardrails so that the AI doesn’t say something you don’t want to an end customer. And, at the end of the conversation, we’re eliminating every piece of wrap-up work so that the agent doesn’t have to navigate all of the dropdown menus or write notes, etc. You can get rid of all that and the computer will probably do a better job than a person will. Agents are under tremendous time pressure, and they don’t necessarily think the same way as a supervisor would think when categorizing a call. It's not what you hire them for in the first place. But they're probably really good at empathy and solving customer problems.
NJ: In these examples, are you talking about implementing agentic AI?
DeMaio: I’ve never seen a crisp definition of ‘agentic’ that everyone agrees is true. In some sense, an IVR is ‘agentic’ when it works really well. But that’s not what we're talking about here. What we're talking about is really: “Am I acquiring capabilities on your behalf to take actions?’
From my standpoint, customers will want to get to that over time – and we will, too. We also care about structure and observability in the really important things. Because of the way we built Connect, we can transition from that ‘press one’ [approach] all the way to full agentic experiences. We are working with customers to expose some agentic capabilities, but most of the time it's not a technology question, it’s a ‘is that the right outcome for the customer’ question.
Agentic must be provable [in the sense that] we're confident when it takes an end customer action – you have to remember how much fraud, for example, can come into play. It's not a question of ‘can I get agentic AI to move my balance over?’ I could do that yesterday. That’s not the problem. The problem is: Do we trust that the agentic AI is the right way to do that, given the many complexities outside of the experience of just the AI.
NJ: One of the terms I’ve seen used lately, particularly in relation to Zendesk buying Local Measure – both of which are AWS customers – is the ‘renaissance of voice.’ Any thoughts on that trend – on voice again becoming important with respect to contact centers, AI, etc.?
DeMaio: I think that's a great term, but it's probably not accurate in that voice never really dropped. Companies tried to get rid of voice, and they failed for the most part, because they were doing it for the wrong reasons – trying to save money and [versus optimizing] customer outcomes. Voice is one of the best ways to interact and certain things are much easier to convey in a sentence than typing into a text box. There are also times when a text experience or a combination is way better.
You're going to engage in the way that helps customers, and often that's a proactive engagement. When your plane is delayed and you're wondering if you’re going to lose your rental car, the rental car company can be very aware of the flight you're on, know what's happened, and reach out to say, ‘No, don't worry, your car is still waiting for you.’ That way you're not frantically trying to call them.
In that scenario, I just gave you the information using the best means available. There are lots of jargony terms for that, but it’s really just understanding the things your customers are likely to need and want in the moment, and then proactively delivering those things to them.
About the Author
You May Also Like
