The New CCaaS Stack: How AI and Agentic AI Are Rewiring Customer Operations

In this episode of the Macro AI Podcast, Gary and Scott take a deep technical dive into how Contact Center as a Service (CCaaS) is being fundamentally transformed by AI—and why traditional definitions of the contact center are no longer relevant. 

What used to be a relatively straightforward evaluation—telephony, routing, and omnichannel—has evolved into something far more complex. Today’s leading CCaaS platforms are becoming AI-driven operating systems for customer operations, where voice, automation, enterprise systems, and real-time decisioning are orchestrated to not just answer questions, but actually resolve customer issues end-to-end.  

The discussion centers on the shift from conversational AI to agentic AI—systems that don’t just respond, but plan, execute, and adapt across enterprise workflows. Gary and Scott break down the modern CCaaS architecture, including interaction layers, AI runtimes, action layers, and control planes—giving business and technical leaders a framework for understanding how these systems actually work in production. 

They also walk through a real-world interaction, showing how AI can move from intent detection to full workflow execution—integrating with CRM, billing, and backend systems—while maintaining governance, observability, and human-in-the-loop controls. 

The episode provides a vendor-level perspective through an architectural lens, highlighting how leading providers like Genesys, NICE, 8x8, Zoom, Talkdesk, and IntelePeer are taking different approaches to orchestration, governance, infrastructure, and model strategy. 

Finally, the conversation ties everything back to business outcomes—exploring how AI-driven CCaaS is shifting key metrics toward resolution, speed, and customer experience, while introducing new challenges around implementation, data readiness, and governance. 

This episode is designed for CIOs, IT leaders, and business executives who want a clear, technical understanding of where the CCaaS market is heading—and how to evaluate platforms in an era where the contact center is becoming the front line of enterprise AI. 

Check out Macronet Services 8 Leading CCaaS Providers:  https://macronetservices.com/who-are-the-8-leading-contact-center-providers-and-what-they-offer/

About your AI Guides

Gary Sloper

https://www.linkedin.com/in/gsloper/

Scott Bryan

https://www.linkedin.com/in/scottjbryan/

Macro AI Website: 

https://www.macroaipodcast.com/

Macro AI LinkedIn Page:  

https://www.linkedin.com/company/macro-ai-podcast/

Gary's Free AI Readiness Assessment:

https://macronetservices.com/events/the-comprehensive-guide-to-ai-readiness

Scott's Content & Blog

https://www.macronomics.ai/blog

Transcript

00:57
Today we're going to deep dive into a category that almost every enterprise touches and one that has been getting a lot of inquiries about lately. And that is contact center as a service, or many of you know it as CCAS. Now, if you think  CCAS is just cloud telephony, call routing, and maybe some omni-channel capabilities layered on top.

01:25
You're already behind the eight ball here. That version of the market is effectively over and service providers are moving at AI speed to add capabilities. What we're seeing now is something very, very different. The leading platforms are evolving into what I would describe as operating systems for customer operations, where voice AI automation and enterprise systems are all orchestrated in real time to not just answer questions, but actually complete work.

01:53
And the big shift we're going to peel back a bit today is more from what used to be called conversational AI into what is now becoming agentic AI. We've talked about agentic AI quite a bit on this show. This is where things get interesting and frankly, where most of the market is still confused. Yeah, exactly. And I think that the confusion  is really why we wanted to put this episode together. And so today, if you listen to  most discussions about C-Cast now,

02:23
They're still framed around the usual features, chat bots, uh agent assist, maybe some summarization,  but that framing misses what's actually happening under the hood. And what's really happening is  that CCAS, Contact Center as a Service, is becoming a distributed AI system where the contact center is just the entry point into the system. And the real value is how these systems  interpret intent.

02:50
make decisions,  and interact with enterprise systems and ultimately resolve customer issues. And that's a lot more of an in-depth process than just routing a call. Right. And so let's anchor this with a simple but important shift in the market. If you look back, historically contact centers were optimized around containment. So how do we keep customers away from agents,  reduce call volume and lower costs?

03:18
As you would expect, that model began shifting very quickly with the rapid advancements in artificial intelligence. What enterprises actually care about now is resolution. And that  means how quickly and effectively can a customer issue be solved, regardless of whether a human or an AI agent handles that particular call. And that's where some people are calling  this more of the resolution economy. And  it's much more about

03:46
demanding uh part of that technical problem.  Yeah, exactly. Yeah, I think now because  answering a question is easy compared to actually completing a workflow.  And so now when you move into resolution, now the system has to  understand all the context around multiple interactions.  It then needs to decide what actions need to be taken, then follow through and execute those actions across systems like CRM.

04:15
billing logistics or what have you. And then if  something changes in the middle of that process, the system needs to understand how to adapt and then either complete the task or hand it off,  possibly hand it off to a human in the loop is that AI term that we keep coming back to. And it needs to hand it off with the full context preserved. And so that's not a chatbot problem. That goes a lot deeper into systems architecture. Yeah, that's a point. And I think...

04:44
You know, if we think about it like this, and this is a model that I think is, you know, every executive should understand it's, it's that modern CCAS platforms are really converging on a four layer architecture.  Um, and maybe we talk a little bit more about that architecture Scott, cause we see it a lot in our day to day lives, just working with customers, but it might be new to some of the executives here on on the show today. Yeah, sure.

05:10
Yeah, when you think of it as a layered model at the top, you'd have the  interaction layer and that's still what people are familiar with. That's the, the channels, uh, voice chat, SMS, email, social channels,  uh, into the,  interaction layer. And then right below that you would see now the AI runtime layer, which is where things really change that that includes speech recognition. And some of them are getting to be very good. You have the large language models.

05:39
retrieval systems, and now,  which is the latest development, you have the agentic planning engines. And we've talked a lot about agentic processes and other episodes. uh And then under that, you'd have the action layer, which is arguably now the most important piece. And that's where AI connects to real systems that you might have in place already. CRM platforms, billing engines, scheduling systems, knowledge bases. We did an episode on knowledge bases  and increasingly uh RPA tools. That's robotic process automation or RPA.

06:09
And then just below that, finally, there's the control plane.  And this is the layer that most vendors don't talk about enough, but really it's critical.  It includes observability, evaluation, governance, um and then other things like redaction, compliance, and auditability. And as we've  been talking about, those have become really important in the world of AI and agentic AI to understand what's happening in the system. Yeah. Those are good points. And I think what's interesting is

06:39
Once you see C-Cast through that lens,  the entire vendor landscape starts to make more sense because every vendor is effectively building towards the same architecture, but they're doing it in very different ways, right? ah Some vendors are stronger in orchestration, some are stronger in governance, others are stronger in real-time performance, but they're all converging on this idea of an AI-driven system that

07:06
ultimately sits between the customer and the enterprise. Yeah. And I think where they're, where they've been coming from is based on, their, their legacy product set, but they're converging, like you said, and that same idea. Yeah, exactly. All right. So let's talk about the terms that that's getting thrown around everywhere right now. Um, and that's agentic AI, because I think there's a lot of confusion about what that actually means in practice. Yeah, it definitely is.  I think the

07:35
The simplest way to define it would  be this, it would be agentic AI isn't just generating responses, ah it's actually planning and executing actions.  So a lot different from the legacy contact center as a service. So instead of saying, here's how you reset your password, the system actually will go and reset the password. And instead of saying,  here's your billing history, it'll actually retrieve the data.

08:03
identify anomalies and can then go forward and initiate a correction if needed. And then critically it can adapt. So if something changes in the middle of the process, like I mentioned, it can then go replant its steps that it needs to take to complete the action. Yeah. And that's a huge shift from what we saw even two or three years ago. Yeah, exactly. Moving fast. Yeah. Way back in the old days, two years ago, AI in the contact center was mostly

08:32
assistive summarization, sentiment analysis, maybe some  basic chat bot capabilities.  And there was a lot of flaws there. the sentiment analysis, was  state of the art. Yeah. Yeah. And yeah. And that wasn't that long ago, 12, 24 months ago.  And I think now the leading vendors are building systems that can orchestrate entire workflows across multiple systems. Often there's no human intervention at all. So

09:01
So the key enabler here is the combination of three things.  You have large language models for reasoning, tool calling to interact with systems,  and persistent state. So the system remembers what it's doing. So that combination is what turns artificial intelligence from a feature into an execution layer. Yeah. Perfect. Yeah. So let's, let's walk through a real interaction to make it a little bit more tangible for the listeners. uh

09:31
So we mentioned a billing issue. So if a customer calls in with a  billing issue, the first thing that happens is  real-time transcription. So speech is converted into text almost instantly. And some of those tools to do that are becoming, they're ranking very high in the performance capabilities there. And at the same time, the system is analyzing intent and sentiment. It's trying to understand not just what the customer is saying, but how urgent or what their emotional state is, what the situation is.

10:00
And then you have a decision point. The system will determine whether this can be handled through retrieval. ah So pulling information or whether it requires  action. ah If it requires action, that's where the agentic layer and that layered approach that we talked about, that's where the agentic layer kicks in. The system might  call a CRM API,  retrieve account data, identify a duplicate charge,  and maybe even go and initiate a refund process.

10:30
all agentically. If something doesn't line up, maybe there's  a compliance risk or the system isn't confident, it'll then escalate again to a human in the loop. But I think here's the key, the human doesn't just then go pick it up and start from scratch. They will receive a fully structured handoff. So that'll have summary actions taken, recommended next steps, all the context that that human would need  to jump right into that interaction.

10:59
And behind the scenes, the entire interaction is logged, traced, and evaluated so the system can improve over time. Yeah. And I think those are really good points because imagine if a human was trying to  explain in that handoff the summary action items recommended next steps. It's  it's going to happen. You're going to miss something, forget something. Yeah. Write something down. Exactly.  And I think that's...

11:24
You know, the part that most people underestimate the value isn't just in the interaction and it's in the feedback loop because over time, these systems  get better at recognizing patterns, predicting outcomes, and  really optimizing the workflows. So that's what turns a contact center into a continuously improving system  for customers and for employees. Yeah. Continuously improving system. That's a good way to put it. Yeah.

11:55
Okay, so let's shift into vendors. We talked a little bit about this a little while ago ah because I think this is where people tend to get, I would say overwhelmed. ah And for full disclosure, my firm represents a lot of these leading contact centers as service suppliers. Scott, yours does as well. uh So we're very, I ah would say experienced in this area. uh There are a lot of players each with their own advantages, but if you look at them  really through,

12:23
the architectural lens  we just discussed, the differences really become clearer. Yeah, exactly. we can, uh, Gary, think we can  probably quickly tag team through a few of them. So listeners can get a flavor for the difference in the capabilities. Sure. I'll start with, uh, I'll just start with Genesis for example. And so Genesis is really leaning into orchestration, which with what they call their large action models.

12:51
Uh, their, focus is on enabling systems that can not only interpret intent, but actually decide which actions to take and then execute on them. So they're, they're also pretty early in exploring agent to agent communication. And we talked about that last summer in episode 28, a two a or agent to agent makes sense. If you, if you think about a world where AI systems are then interacting with each other. So you're using a contact center as a service and that contact center as a service is interacting with other agents.

13:21
complete tasks. Yeah. And then you could look at a vendor like Nice, which is taking a very different approach. They're focused heavily on governance and memory,  persistent context  across interactions, strong compliance controls, and the ability to  simulate and test AI behavior before it goes live. And that's,  I would say, very appealing for large enterprises that really need predictability and control in their business. Yep. Yep.

13:50
Exactly.  And  I'll take a look at, eight by eight,  they're an interesting one been up in this space for a long time, originally a unified communications as a service provider and then built on the CCAS capabilities.  And they're  interesting because they pushed AI down into the infrastructure layer.  And what that means in practice is that AI isn't  sitting off to the side waiting for

14:17
transcripts, it's actually operating in line with the voice stream. So processing audio in real time as packets move through the system. And that  architecture has a few big implications. So first,  in  a lot of cases, you'd be very concerned about latency and voice latency and processing latency.  You're not dealing with the typical delays you see when audio has to be sent out, processed, and returned.

14:43
With 8x8, you can get sub 100 millisecond response times, which is what makes a voice interactions feel natural instead of robotic.  You don't have that pause in between. And then secondly, just to add onto that kind of a strategic part is it allows 8x8 to unify context across both UCAS and CCAS environments.  And because they control the underlying infrastructure and the data layer, their AI can access signals not just from

15:13
customer interaction, but from internal communications, presence, which is important, and also collaboration tools as well. So just to kind of wrap that up, instead of optimizing just the contacts interaction, they also look to optimize the entire communication graph for the enterprise.

15:32
And I think there's a couple others here too. uh A lot of people are probably not familiar that they can offer this, but they're familiar with the platform and that's Zoom. Zoom is taking a different approach with what's essentially a federated AI architecture. And many people are familiar with Zoom more for meetings, but they don't realize that they have a CCAS option.  So instead of relying on a single model, they can dynamically route tasks across multiple models. So think of models like

16:01
open AI, entropic or their own based on the use case.  And that matters because not every task requires the same level of intelligence.  Real-time voice needs low latency.  Summarization needs cost efficiency and complex workflows  really need stronger reasoning. So Zoom is effectively introducing a decision layer at the model level, selecting the right model in the right time.

16:28
based on performance, cost and latency. And I think that gives them three advantages here. So they have better cost control, more resilience if a provider changes or fails. And then there's the last component, long-term flexibility as models continue to evolve, which we see on a daily basis. So while most vendors are building vertically integrated AI stacks, Zoom is betting on a model agnostic orchestration layer.

16:56
which is a very different strategy compared to some of the others here that we've talked about. Yeah. It'll be able to plug in whatever models down the road as well. Right. Yep.  I'll just take on another one. That's a talk desk. um Probably more familiar to uh mid enterprise users.  The talk desk platform is built around a,  what we call a coordinated system of AI agents. So what they're positioning

17:26
through autopilot, copilot, and their CXA framework is where different agents are responsible for different parts of the workflow. So instead of a  single bot trying to do everything, uh you have essentially a lead agent that's orchestrating specialized sub-agents  for things like identity, verification, billing, logistics, or  compliance.

17:50
And they've put a lot of emphasis on  real time observability and control, is  a pretty critical point for things like production deployments. Yeah, yeah, those are good points there.  The last one that I'll mention, and there's certainly others, uh but for today is a company called IntelliPeer. So IntelliPeer is essentially building an AI-driven orchestration layer that sits between communications, business systems. So think of it as their

18:19
focused on automating real-time customer interactions without requiring a full CCAS replacement. So technically they sit closer to  probably a  communications platform as a service or known as CPAS.  Yeah. Plus AI orchestration model. So with strong integrations into contact centers, CRMs and uh know, backend systems. So that makes them particularly effective for enterprises that want to layer AI driven automation on

18:49
hop of existing infrastructure that they've already purchased. Yeah. I think that's kind of a good overview of some suppliers. uh But I think  that just looking back on it, one thing that's important to understand is that most of these platforms, they're not building everything from scratch. They're sitting on top of a broader AI ecosystem that they're tapping into. So you have the foundational models from companies like OpenAI, Anthropic, Google, Microsoft, and you have

19:18
specialized providers  or their own platform that, will be doing things like speech recognition and voice processing. So there are all these different components.  And then you have automation layers,  like RPA platforms that, that will actually then go on execute tasks. Yeah. So in many ways, C-Cast is becoming the orchestration layer that ties all that together. Yeah. Yep.  And the orchestration layer is the environment where these technologies come together to solve real customer problems. So it's.

19:47
really, really interesting, especially how this has evolved. Like we said in two years time or less, it's completely flipped. Yeah. Yeah. A lot of, a lot of movement here. Um, so let's just kind of, uh, bring it back to the business impact is what have a lot of our listeners kind of tune into, um, really at the end of the day, none of this matters if it, if it doesn't make, you know, move the CX metrics. And I think it's, it's the metrics that are, that actually matter.

20:16
is what's changing. It's not about, not just about reducing cost fall anymore, which used to be something that was a metric that was monitored. Now you're looking at things like uh resolution rate, time to resolution, obviously customer satisfaction, that's table stakes, and increasingly  the overall, overall cost to serve. And what we're seeing is  that when these systems are implemented correctly, you get

20:44
improvements across the board. So faster resolution, fewer transfers, better customer experience,  also known as CX. So this investment should pay off  tenfold just based on the enhancements  that are coming out. And who knows what will come out here in the next six or 12 months. Yeah, agreed. And just you mentioned implementation is obviously key. uh So keeping in mind that the design and the sourcing process

21:14
Uh, when you're, when you're going through the design and sourcing process, you also need to be thinking about implementation, uh, because these systems are only as good as the data, uh, the integrations and the governance around them. Like we've talked about in other AI systems in the show. Well, 100 % implementation planning is, critical. It's it, it shouldn't be taken lightly. Um, especially because AI works great in demos. Uh, you get on there with your account team and everything looks great, but production environments are messy.

21:44
especially with legacy tech and debt that might've accumulated over the years. You also could have things like hallucinations, broken integrations, data privacy concerns, compliance requirements, that could keep going on and on. And that's where the control plane really becomes critical. You need visibility into what the system is doing, the ability to evaluate its performance and controls to ensure it's operating within policy. And I think that's really paramount here for, you know, kind of this next iteration of CCAS.

22:15
Definitely. Yeah. Couldn't agree more with that. I think that the companies that get it right are the ones that  treat AI as a system that needs to be monitored and managed and not just deployed. Yes, agreed. And so hopefully this was a good overview for all the listeners today. And based on popular demand, we will follow up with more episodes around  how AI is making a real impact in the customer experience, the CX space.  And maybe what we should do, Scott, we talked about this before the show.

22:45
close  with a few predictions. Yeah, sure. Sounds good. ah Obviously, the CCAS space will move at lightning speed  in pace with AI capabilities. And we saw a huge acceleration just a few years back, having seen CCAS kind of slowly develop for years  and now rapidly accelerating.  I think over the next two years, we're going to see uh autonomous resolution is going to be the default expectation.  And based on what we're seeing with some of these uh

23:14
most recent models,  uh, and some of the scoring that's coming out. think that you're going to see a voice AI that reaches  near human quality. not, not the, uh,  dialing the toll free number of the old days, but you're going to be, it's going to seem like you're really talking to a human and that's already coming out. Yeah, no doubt.  and suppliers enterprises will roll out personal AI agents interacting with enterprise systems. Um, also I think

23:41
You know, pricing models for Ccast shift towards outcomes and observability becomes the primary differentiator here. And maybe the biggest takeaway from this entire discussion is really this. The winners in this space  are not going to be the companies with the most features.  And I know that that's probably not what anybody out there that works in product management wants to hear. ah But the reality is they're going to be the ones with the best systems, systems that can reason, that can act.

24:11
over time. Those are the companies that probably win this race. Agreed. Agreed. And for anyone out there that's evaluating these platforms, think the question that you should be asking is not what AI features do you have. ah You definitely need to dig a little bit deeper and understand how does your system actually work? What solutions are native to your platform and which solutions are bolt-ons?

24:37
And, um, and also we mentioned implementation. What does your implementation model look like for taking AI from pilot into production specifically around integrations, data readiness, observability, and ongoing optimization? Otherwise kind of like how pilots are failing out there. CCAS implementation can fail. don't want that to happen. Um, so I mean, previously the CCAS evaluation process was, relatively straightforward. You know, you had telephony routing, routing logic, and,  and some omnichannel capabilities.

25:06
But now you're effectively evaluating a distributed AI system, including model strategy, ah orchestration, observability,  and then obviously integration into your core business systems. Yeah. And I think that's a good place for us  to wrap up this episode of the Macro AI podcast and also a good lead into some more conversations on the topic of contact center as a service, also known as CCAS  on our future episodes.

25:35
If you're a business or technical leader, this isn't just a contact center decision anymore. It's an AI architecture decision. And the companies that get this right are going to move faster, operate smarter, and deliver a fundamentally better customer experience.  We'll keep breaking it down and thank you for listening.