Stop Calling It a Chatbot: Rewriting the Rules of Virtual Agent AI - Diabolocom

Diabolocom’s AI Product Manager Tom Azernour on why orchestration, not automation, is the real future of contact center AI

Host Rhys Fisher, Associate Editor at CX Today, sits down with Tom Azernour, AI Product Manager at Diabolocom, to unpack why the virtual agent category is long overdue for a rethink. 

As contact centers move beyond legacy voice bots into the agentic era, Tom explains what "orchestration-first" AI actually means in practice and why the architecture decisions made today will define CX performance for years to come. 

If your virtual agent is still just deflecting calls, this is the conversation you need to watch. 

Most virtual agents deflect calls. Diabolocom's orchestrates them. Tom Azernour breaks down what separates the new generation of AI-powered virtual agents from legacy bots, and why the difference matters more than most leaders realize. 

From deflection to orchestration: Why the decision tree model is dead, and how Diabolocom's agentic approach gives virtual agents a goal and the autonomy to achieve it, connecting to CRMs, ERPs, and ticketing tools without overhauling your existing stack. 

The structured handoff: Why passing a raw transcript isn't enough, and how Diabolocom extracts intent, summaries, flags, and custom data fields to give the next agent, human or virtual, everything they need from the first second. 

Built for operators, not engineers: How Diabolocom designed a platform that non-technical teams can configure in minutes, without sacrificing the depth that technical users need. 

The CCaaS + telco advantage: 20+ years of end-to-end infrastructure ownership means Diabolocom optimizes its AI models and telephony stack in harmony, delivering latency that competitors wrapping third-party solutions simply can't match.

Transcript

SPEAKER_01 0:08

Hello and welcome to CX Today. I'm Reese Fisher, Associate Editor, and today I'm delighted to be joined by Tom Azenau, the AI product manager at Diabolacon. Tom, thanks for joining me. How are you doing today? Yeah, me too. No, I'm looking forward to the chat today. You know, we're gonna be discussing virtual agents, which are obviously a big topic in the CX space right now, and kind of really trying to unpack some of the big questions surrounding that topic. So I guess just to kick things off here, we know that contact centers, you know, have had virtual agents for quite some time now. But what do you think has actually changed more recently? And how is that linked perhaps to what you guys at Diabolacoma calling the orchestration layer?

SPEAKER_00 0:56

Sure. Um well indeed and I think there is quite a mix-up in what's what, uh, which is not surprising considering everything that is happening right now in the field. Um so to do some kind of a reminder, um, there is the traditional, let's say the legacy voice boots, which were the first generation of virtual agents, to be fair, uh, and that were about deflection of uh redundant requests. So it will be saying something predefined and it will follow a path, a predefined path. And truth is, um, it's great in uh specific situations, but it's uh recognized from a mile away, and sometimes it has bad understanding. Um but more importantly, you have to respect the flow or it may fail. And at this time, this is uh what we started with and what uh the market started with, because this is what technology enabled us. And and as time went, we went, let's say, uh a step deeper and we started with actual IVA technologies to answer uh all the immediate needs of contact centers. In this case, I'm talking about the ability to qualify a call, right? So there is a call or actually some other kind of interaction, obviously. Uh, but I will be talking about calls since it's the most demanding uh channel we address. So you you understand why the person is calling, right? So we can route it toward the proper human agent. Well, it was a huge milestone for a more efficient contact center, even though the virtual agents were not really specialized per se. But now I'm sure everybody's watching this video has heard about the agentic era, which is more um oriented for specific market needs. Um and this is what we are building, we've been building over the past few years now. Um as in agentic, uh, from agency to be able to act or to do, not exactly an agent, and because right now there is a lot of uh things heard and uh people discussing about what an agent is. The baseline here is that we want uh a fleet of agents of uh IVAs to be able to act toward a specific goal. Right? So to do that, we don't rely anymore around a decision tree with all the possible outputs uh planned out. Uh this is a way to cumbersome, and you will always be missing some outputs, right? So we rather give it a starting point, a goal, and the IVA will fill in the gaps. Um and in itself, it will be the judge to decide whether uh it's still missing some bits of information and it needs to ask some questions to the caller or to the to the client. Um but maybe it has enough context, maybe it understood what's requested, and it will use and rely on its connectors, its access to other applications to actually do something the same way a human would do. Um because the end game is to have an IVA that is as capable as a human for all these redundant tasks, even where when they're more complex. Um so uh to sum up, this is what we call the orchestration layer. This is the blueprint that we have been building and we are still improving on, so that any company can leverage their existing tools without changing their uh stack. Whether it's a CRM, a ticketing tool, um an accounting app, an ERP, you name it. You build the virtual agents around your needs the way uh it complies with your processes without all the hassle of change management. We comply to you, not the other way around.

SPEAKER_01 5:11

Yeah, thanks for that, Tom. It's really interesting there listening to how you brought that orchestration layer in and you touched on something there, kind of running through how virtual agents perhaps have been viewed in the past, this idea that you know quite often they're viewed as just a system that can hand off to a human with a transcript, and that's kind of where the role stops. What does maybe a structured handoff look like? And what difference does it make to the agent that it does end up picking up that call when they get the structured handoff rather than the traditional?

SPEAKER_00 5:42

Sure. Okay, this is an important bit, even more when we know that the handoff could occur between a human and a machine, a virtual agent, the other way around, or between two virtual agents actually. And um just to understand where I'm going, uh we just need to acknowledge that one of the I would say the most important currencies in quantity centers is the time. If the client is spending more time that they should have, uh we don't get the benefit of the IVA because they've been saying the same thing over and over. They've been repeating themselves, they've been talking to uh 10 different virtual agents. Um so on this side of things, this is not great. On the other way around, we want the human agent to also benefit from the IVA. Um if they have to ask the same question over again, this is also uh time that is lost for them. And this was one of the complaints and uh notes we had from our uh partners is that some of the competitors we have, some of the solutions out of the box of the market, do provide limited context. The approach is pretty standard, uh, but it's been done by design from day one. We provide all the connectors so that you can retrieve anything that happened during the conversation. So obviously there are the basic blocks such as the transcript and the summary of the calls, always just uh get it directly after the call. But we also bring to the table and all the specific data that was uh extracted. So you can specify that you want to fetch the name, the age, uh, I don't know, the contact, the contract number, uh, the favorite number, the favorite color. Uh you decide what you want to fetch during the conversation, along with alert flags, detecting language, and much more. And you can use that in the context of upcoming agents, whether they're um human agents or virtual agents. Um we provide uh the framework, you leverage it the way you want. Um obviously you have access to all of that. Uh you can pick some of them to, as I said, to provide it to the next agent, or maybe you want to refine it. Uh you want to work on it, to send it to a CRM or something else. So this is how we make sure that you get what you need. Um but I think the other the other way around you should also be able to feed data to the virtual agent so it will handle the call with any kind of context. So maybe it wants to know what happened during the previous call. Maybe you want a welcome message that's personalized with the first name of the caller. Um all of that end-to-end is uh possible and baked in the solution. So we'll just focus on uh making wave flows. You can provide input context and output context.

SPEAKER_01 8:56

I see. It's really interesting there hearing you talk because it's obvious that you're kind of straddling that making things as user-friendly as possible, but at the same time, it is a piece of technology that has, you know, fairly complex capabilities. And I know part of what you guys are trying to instill was making it, you know, usable for those non-technical operational teams. How did that design philosophy, I guess, shape this product?

SPEAKER_00 9:20

Sure, you said it well. Um, our clients are professional in contact centers. Not everybody is as proficient in AI. Not everybody is um an expert or a tech leader. Um they're operational teams. They know what the agent, whether they're they're human or a virtual uh agent, should work, it should say, should behave. Um and truth be told, there is a base knowledge needed to produce efficient prompts and flows and to have uh a very uh proactive virtual agent. So it's why the platform is built around all the standard concept and components to, on the one hand, build virtual agents, on the other hand, monitor fleets of virtual agents, and um this is done out of the box, right? So you can get started from a template, and within minutes you can get started and have a virtual agent that is able to answer its first uh goals. Um if you're more on the tech savage side of things, because we have some as well, um, you can fine-tune your agent by doing microadjustment. For this, I just want to lay out that you have the full freedom and full access to these options, which are exactly the same one that we use to build uh complex workflows within the able virtual agents. Um and so um we made sure that uh we did answer the needs of let's say 99% of our users. We still have a remaining 1% for it which we have uh internal experts that are here to assist for any kind of needs.

SPEAKER_01 11:11

Yeah, I think that's that's really important, like you said, although you know it's it's obviously important for it to be usable for the non-technical support uh leaders. There there are going to be those those really technical people who want to get into the nuts and bolts of it and have a real play about and push it to its limits. So it's good to kind of cover all bases, I guess.

SPEAKER_00 11:31

Absolutely, yes. And um This is also something that we keep on working on. Obviously, this is um this is a very organic product, and the technology is going so fast. We just make sure that we have everything that is built in. So uh you're getting more and more assisted when you configure a virtual agent so you can actually uh build it around a specific need without struggling with the technical part. It's all planned out and we keep on improving on in this regard.

SPEAKER_01 12:04

Yeah, yeah, that's a good point about kind of the fluid nature of this technology. What do you guys do specifically maybe to make sure that you are kind of keeping up to date with any any of the changes in the in the space?

SPEAKER_00 12:16

Well, uh many things actually. Uh there is the obvious uh answer, which is participating to venues and exhibitions where the market leader is show the trends and talk about uh the new concepts, the new frameworks. But also I think keeping uh a foot in the field as a product manager, as a user, as someone from RD, because we have uh these people working within the company as well. Um, this is how to smell the smoke before the fire, you're starting to get insights from the market. Maybe there is uh a new framework that is coming up, you tried it and it's efficient. So even before the main players started to implement them, you can already start to tweak it, to play with it, maybe to implement it. Um and since before being a provider of AI solutions, as individuals, we are users of AI solutions. So we we know when things get hard, but maybe we're biased, like everybody else. Um it's why we regularly do workshops with our clients. Uh we do some testing to make sure about the usability of the product, and uh we are big on getting feedback and insights to continue making it better and better.

SPEAKER_01 13:41

Yeah, yeah, thanks for that, Tom. Like I like I kind of mentioned, it's it's we hear a lot about how these companies need to stay up to date and on trend with AI, but we don't have to hear perhaps how they are doing that. So that's uh yeah, some really, really valuable insights there. I guess I wanted to also pick your brain about another thing I find interesting about Diabolicom, and it's your position in the market as a CCAS platform, but also a telecom operator. How does that maybe make a difference when it comes to deploying AI voice solutions?

SPEAKER_00 14:13

Sure. All right, so um I think it's a good time to give a bit more of history regarding diabolical com uh because we've been doing that for uh more than 20 years now. And uh since then we completely cover the six environment. So we do provide the telephony system and infrastructure, but also the contact center solution. Now AI enabled, but even before that, for years, we already provided uh the CKA solution. Uh and I think it brings tremendous benefits because since we have end-to-end hosting and uh management of whether uh the infrastructure, since we are uh effectively a telephone telecom provider, but also all the AI models, fine-tuning, refining, hosting, management, we're able to uh optimize all these modules to work in harmony. And um so I one thing is uh to circle back to our question, we have so much data to work with, whether it's conversational data, so audio recordings, uh written channels with mails and chat boxes, SMS, you name it. Um we're able to get a continuous feedback loop to understand where it's not accurate enough or what needs to be tuned a bit. Um but also for the technical side of things. Since uh we have access to everything that builds the system, this is how we are able to uh tweak all the things that made that make the virtual agent more efficient faster. And I think this is one of the best takeaways is that uh we have a best-in-class latency. Since we have the low-level integration, we're able to uh um tune the things so they're they they feel and they are actually blazing fast. Um and this is why we're providing the out-of-the-box experience uh instead of wrapping many AI solutions for which we would like um efficiency to optimize. Um but I'm thinking about something else. I think this is one of the why most partners choose us. So we know the game, and we've been here for a while now, and uh all knowledge is internal. We tick the boxes in terms of functionality, of uh security and compliance, which is a big topic in Europe, but also in the world ofgether, uh sovereignty, and lastly, we're not dependent on other providers whose pricing is currently surging. We manage everything, so we make everything predictable and there's no surprises.

SPEAKER_01 16:59

Yeah, yeah, I know we've uh yeah, we've spoken in the past about kind of uh Diabolic Commons approach, that kind of non-generic AI, unlike your outlander. It's uh I think it's definitely a strong differentiator in the market. I guess my final question here, Tom, I want to pick up a little bit on what we discussed earlier about obviously you guys how you work to stay up to date with your AI, but I was also interested in what you think maybe the future of Diabolic Commons virtual agent is and kind of anything maybe you have in the works.

SPEAKER_00 17:28

Sure thing. Um well uh that's very interesting. Um first of all, there is uh one thing we had it coming, uh, but it's uh getting a reality now. The full identification orchestration to make a long story short, it's the ability to chain multiple boats, virtual agents in autonomy to switch seamlessly between uh all of them depending on what's the client requesting. Um, because an IVA system is not a simple virtual agent answering anything, it's more of a uh collection of specialized IVAs. Um so we've been going this direction for uh a month now, and it's getting a reality this year. Um but I'm also thinking about um extended outbound capabilities, right? Because we've been talking about uh um getting and addressing calls and conversation that were coming your way, for which you will answer a certain way. But when you're doing outbound, you shuffle the carrots again. It's not working the same way. Um you could recover a stalling shopping carrot, or you would like to collect some information from a client before a technician comes to their house. So there are quite a bit of things that you need to tweak so that the conversation feels natural. Uh so we've been working on that for a while now, and it's uh it's coming this year as well to improve even more on our capabilities. But I can also think about a few other things. Um we are also big on language support. Uh we are uh actually adding five more languages this month, and we have way more coming up uh this year. So we cover more and more the needs and the ability to um answer immediately to any languages, even though it switches during the conversation. And if I could, yeah, if I could give one uh if I could give you the last one, I would think about Omni channel orchestration, uh, regardless of the channels. Um of course, there are countless ways of uh getting in touch with uh with an agent. I told before SMS, mails, voice, but there are actually way more ways to ask for help or to contact the company. So we are building on uh bringing that natively to the solution, uh step by step this year, and we would be glad to have some people uh actually jumping on board and uh testing our new things coming up. But of course, uh this is all going very fast. Uh I cannot disclose everything that we are working on, but we are very ecstatic about what's coming up.

SPEAKER_01 20:26

Yeah, yeah, that's uh obviously some big things on the horizon. Of the ones that you could talk about and you've discussed so far, is there one that you're kind of especially excited about or you think will particularly resonate with uh with your customers?

SPEAKER_00 20:40

Um I think the biggest thing coming up is um what I call the full agentic orchestration. Um we start we are starting to have a good overview of uh where the market is heading, but also where the market stands right now. And I'm not here to do some finger pointing, um but I'm also a user of uh agentic system and the IGAs altogether. And uh it seems that uh we still we are still in the midwater, is being between the old systems and the new way. Uh AI is still trying to find its way, and right now it's kind of a mixed bag. Um sometimes you have to still keep the let's say the legacy way of orchestrating agents. You have an independent agent that will uh route context and information to the next virtual agent, but they're not communicating altogether. Um this is actually uh why the agentic orchestration is something that we are prioritizing right away. Since we have a very good uh velocity in terms of what we are shipping, we think and we we are very, very hopeful that uh we can bring that to the table before anyone else in an efficient and lightweight way, uh, and not having an over-engineered system where you have to to get a PhD to uh to engineer the virtual agent.

SPEAKER_01 22:15

Yeah, well, watch this space. Very uh yeah, sounds like really exciting stuff, Tom. Um yeah, thank you so much uh for for the conversation. It's been really interesting. I think like you outlander, the junk really, it's it's virtual agents kind of it covers such a wide range and different facets now, but that there's this tendency to kind of lump it all in together. And I think uh this has been a really useful conversation to kind of outline the differences and what different vendors are doing differently. So yeah, I think the audience is gonna get a great deal from this. So thank you very much for your time.

SPEAKER_00 22:47

Thank you very much for having me today. It was a pleasure as always to be discussing uh customer experience with you.

SPEAKER_01 22:53

Absolutely. I did just want to quickly also thank our audience uh for tuning in. If you enjoyed this, and I'm sure you did, please remember to like and subscribe to the channel and head on over to cxtoday.com for more stories like these. Until next time, thanks for watching.