Agentic AI For Customer Service

Agentic AI sounds like magic until you have to ship it inside a real enterprise contact center where every mistake becomes a compliance problem and every bad interaction hits customer trust. From Enterprise Connect, we sit down with Dialpad to unpack how they’re building agentic customer service on top of an end-to-end UCaaS and CCaaS platform, and why having AI deeply integrated into the communications stack matters more than flashy demos.

We dig into the hardest first step: figuring out what to automate. Dialpad explains “skill mining,” a way to analyze your past customer conversations and surface a practical hit list of workflows that are both common and realistically automatable. Instead of guessing which user flows will deliver ROI, you start with the data you already have, then keep refining as new conversations reveal new opportunities.

Then we get into the part that makes or breaks adoption in regulated industries like healthcare and financial services: safety. Dialpad walks through Guardian, a live conversation analyzer designed to protect agentic voice and chat with layers for security and compliance, scope and intent alignment, and frustration detection, plus a built-in path to hand off to human agents when the customer needs it. If you’re evaluating AI for contact centers, customer experience automation, or enterprise-ready agentic AI, this is a practical look at what “trust” really requires. Subscribe, share this with your ops team, and leave a review with the workflow you most want AI to handle next.

Transcript

What Dialpad Builds With AI

SPEAKER_01 0:01

And I'm at Enterprise Connect with Dialpad. Calvin, how are you? I'm very well, Evan. Thanks for having me. You guys are taking over the world. So much news and announcements here at Enterprise Connect and Beyond. Yeah. How do you describe Dialpad these days, those who may not be familiar?

SPEAKER_00 0:14

Yeah, so at Dialpad, we offer um an end-to-end UCAS and CCAS solution, right? And it's a complete platform. So our customer success platform is just built on top of our end-to-end unified communications platform. But since 2018, we've had AI running through it, right? So that helps you make the most out of your conversations. We have a lot of great tools that are AI-based. And this week we're launching, well, last week technically, but at the show here, we're launching our agentic product, which we're very excited about and is going to really change how customer service gets done and impact a lot of businesses. So we're very excited about it.

Getting Started With Agentic AI

SPEAKER_01 0:49

I'm excited to hear more. I have actually five agents working for me as we speak across different platforms. And for me and my small business as a professional sort of solopreneur, it's been astounding. It's changed everything. So I only can imagine in an enterprise environment the impact it'll have. How do you start with clients on the Gentec AI roadmap? What are the first steps and what's the ultimate vision?

Skill Mining Finds Best Automations

SPEAKER_00 1:16

Yeah, that's a perfect question. So we recognize that a lot of companies want to take this journey. They just don't know where to start, right? They're not confident in what user flows they want to automate. They're not confident that they're going to see the ROI. And so we've built our entire agentic platform around building trust that your agent is actually going to perform, do the task that it's set up to do, and uh achieve that ROI. So that starts with what we call skill mighting. So that's if I have no idea what flows I want to automate, what we're able to do is analyze, you know, a corpus of past conversations from your contact center, uh departments, whatever it may be, for a set period of time. You can dictate that. And what we do is we triangulate the use cases that are ripe for automation. So we do that kind of along two main dimensions. The first dimension is just the frequency with which that flow occurs. Um and then the second dimension is uh how automatable, if you will, uh that flow is, right? Um so yeah, prevalence. So these are things that are happening on a lot of conversations and taking human time to do these tasks. And so we detect if the flow is, you know, ripe for automation by how the human performs on it, right? Is that a simple repetitive task that the human's doing on you know 30% of their calls? Then that's ripe for automation. So using your own data, that will show you, and Dial Applies platform with skill mining, that'll show you where to start. And that will give you the hit list essentially of the use cases that are ripe for automation. And that evolves over time, right? As the agent is having live conversations in production, you can remine skills down the road and it will come to the fore and bring out new flows that can be automated. So um it's a really, really effective tool and a great way to start if you if you don't know where to start, but you know that there are some redundant flows that you'd like to automate.

SPEAKER_01 3:08

That's fantastic. You have so many customers who really love dial pad. Um, do you think agentic is right for all of them, or are there certain use cases or industries or verticals where you think you'll have the most traction with agentic AI?

SPEAKER_00 3:21

Yeah, so I I think that agentic is going to have an impact on any business. Wow. Even if it's a small one, you know, for a small business, even if it's just like a simple um, you know, uh AI receptionist, right, who's gonna route your call to the correct person. That's valuable for any business, right? Um, but uh we are going to be uh focusing throughout the rest of this year um on tailoring the product to really attack certain verticals that are ripe for you know this type of automation. And that's really come from uh a lot of the customers that we're already working with in the agentic space, in particular verticals. Um, like healthcare is one that uh that is very attractive to us, obviously. And we have a great customer um in the healthcare space that we're building agenc flows for. And so it's we're we're taking sort of the tack of um uh it's a generalized solution, but how it's deployed and what it's used for can be very industry specific, right? And so we're currently sussing out what those specificities are, and then we can bring to market for you know certain industries like healthcare, financial services, all you know, regulated industries are often the ones that are ripe for help here. Um and uh we can tailor this we will know how to tailor the solution for healthcare, for financial services, for retail, right? Um, and so uh yeah, universal technology, but I think it's gonna have a really big impact um in certain verticals, and that's that's what we're tailoring the product towards. Not so much in terms of like the features and functionality, they can probably be universal, but how the AI is deployed, what flows are being automated are gonna be industry specific.

SPEAKER_01 5:00

Fantastic. So you mentioned large regulated industries, um, of course, guardrails and compliance and data protection, yeah. Something you know a thing or two about, but they really come to the fore with agentic. Uh what are some of the philosophies there around uh of those topics? Yep, how to protect them with my agents? I mean, I'm sort of a solopreneur, so I can have fun, I can set up OpenCloud, I can do all sorts of cool things, but an enterprise can't necessarily do that. So, how do you enable this innovation while protecting the the enterprise as well? Yeah, very important question.

Guardrails With The Guardian Layer

SPEAKER_00 5:37

So um we've been doing AI features since 2018, um, when Dialpad acquired a company called TalkIQ that did uh NLP, ASR, um speech to text, right? Um you know, uh very performant, uh benchmark against industry standards, like the best in the business, right? And so we've been building on that foundation since 2018. And so we have that compliance uh and security posture sort of baked into how we build because we've been doing it for a long time, right? So we know that um, you know, if you're using a transcript to uh run processes, we need to strip out the PII before processing, right? Um so that artifact as it moves through the system uh is clean, right? So those types of things are just sort of baked into the way that we build our product, and we pass that you know level of security on to our customers, along with you know the official you know security standards um uh that you can see on our website, right? Uh trust at dialpad.com. Um and so that's that's kind of like uh at the base, but we're also very conscious of protecting those agentic conversations live in production as well. And we do that with a feature called Guardian. So what Guardian is, is it's a conversation analyzer that uh monitors all of your agentic conversations, uh both chat and voice. Um, and there's three main like uh elements of analysis that it does to ensure that it's a safe and secure conversation, right? So the first level of analysis is um probably what we're all thinking, right? The security and compliance layer, right? So that will do things like um, you know, scrub PII, um ensure that the agent isn't releasing any data that it's not meant to. Um it protects against things like prompt injection, the like that. Um, you know, things like hate speech, uh, it'll sort of try and redirect the conversation or hand it off to a human if if if those types of things are are uh occurring in a conversation. So um it's built on Google Model Armor, um, which is the industry standard at this point. And so um, in terms of all the data compliance and regulatory compliance, that's that layer. Secondly, we run a conversation analyzer or conversational alignment analyzer as well, right? So um that will ensure that the session is actually within the scope of what the agent can handle, right? So we have really robust intent detection, and that can change throughout the course of the conversation. But the moment that the customer starts, you know, asking about something that's out of scope for that agent, um, we can ensure that that's handed off to a human who can actually uh help the customer, right? Um and so then the third analyzer that we run is a frustration analyzer. Um and so it's built on uh our customer sentiment model that we've been iterating on since 2018. Um but then it also has a second layer of analysis wherein it ensures that that frustration is actually directed at the agent before suggesting a handoff to a human representative, right? Um so the caller can be incredibly frustrated about the product, um, but still you know receiving help from the agent uh readily and it won't you know terminate that session and hand it off to a human, it'll it'll you know do what's right for that for that session, right? And so that's a way that we can both help ensure uh security and compliance, but particularly with the latter two analyzers, it really enhances the customer experience, right? Because at Dialpad, because we're an end-to-end solution, because that conversation is being handed off directly to a contact center in the dial pad ecosystem, um it's never an issue for us to do that handoff, right? And so the second that the customer experience starts to dip, then we can save that session and hand it off to a human, right? It's it's never um it's a feature of the dial pad agenc platform, uh, not so much a bug, right? Where some of our competitors, um, they try to keep that agentic session going, hell or high water, right? Because that's a business model, right? With us, because it's soup to nuts and it's built on the contact setter platform, um, we have no problem, you know, escalating that conversation to a human agent.

Why Agentic AI Is An Inflection Point

SPEAKER_01 9:41

Fantastic. Wow. Well, so much progress. You guys are really on a chair. Uh, speaking of which, you you've been with Dialpad for quite a few years. Uh you're still a founder-led company, Craig. Yes, sir. Craig is at the helm. Yep. You're Silicon Valley based. How does it feel these days? The inside, uh, looking out uh within Dialpad, you must you must have a lot of uh enthusiasm and excitement around the possibilities and uh where you're headed.

SPEAKER_00 10:06

Yeah, yeah, absolutely. I I really think um AI in particular, you know, now with agentic, um, we're really at an inflection point in the industry. Um and I think even outside of the customer uh support ecosystem, I think as consumers become more comfortable speaking with AI and interacting with systems agentically, um, we're gonna really see that type of interaction take off. And it's really gonna be the new sort of operating system for how people deal with software, right? Or how people interact with software. Um, and so we think it's gonna be really transformative for businesses, for all of our customers. Um, again, in any industry, big, small, um, you know, there's there's something that's right for automation there. And so, yeah, we're we're incredibly excited. Um, it's definitely an inflection point that we are going full seam ahead on.

SPEAKER_01 11:01

On a mission. Well, congratulations on all the success so far. Thank you. And onwards and upwards. All right, thanks so much, Evan.