Some frustrations in life are fairly universal.
Whether it’s someone stealing the parking spot right outside your house, accidentally stubbing your toe, or having to explain the same issue twice on a customer service call.
We all know the drill. You’ve called a contact center with a billing dispute; you’ve navigated a virtual agent, repeating your account number twice and providing a detailed explanation; and you finally get through to a human agent who immediately asks: “So, what seems to be the problem today?”
The frustration is real…and somewhat warranted.
In instances like this, the virtual agent has handled the interaction; it just hasn’t done anything useful with it.
That gap, between what virtual agents process and what they actually pass forward, is one of the more persistent frustrations in the contact center industry.
According to Tom Azernour, AI Product Manager at Diabolocom, it reflects how the category has been defined from the start:
“The traditional virtual agents, the legacy voice bots, were about deflection of redundant requests. They follow a predefined path. They’re great in specific situations, but if the caller doesn’t perfectly follow the flow, they may fail.”
Thankfully, the industry has evolved since then. Intelligent virtual agents have brought intent detection and smarter routing.
However, the underlying logic of handling what they can and handing off the rest has largely stayed the same. The handoff is still the problem.
From Deflection to Orchestration
For Diabolocom, fixing that problem means rethinking what a virtual agent is actually for.
The decision tree model, where every possible outcome is mapped in advance, is obsolete; instead, contact centers should be striving for goal-driven autonomy.
“We give the virtual agent a starting point, a goal, and the virtual agent fills in the gaps,” Azernour explains.
“Virtual agents judge whether they need more information or whether they have enough context to act. Then, they use connectors and access to other applications to actually do something – the same way a human would.”
The practical effect is that the virtual agent stops being a front door and starts being a functioning part of the contact center’s workflow. Diabolocom calls this the orchestration layer, and the distinction matters most when the interaction ends.
“One of the most important currencies in contact centers is time,” Azernour says.
“If the client has been saying the same thing over and over, we don’t get the benefit of the virtual agent. And if the human agent has to ask the same questions again, that is also time lost.”
The answer is structured output. Where most virtual agents hand off a transcript, Diabolocom’s approach extracts intent, summaries, flags indicating dissatisfaction or potential fraud, and adds any custom data fields the business specifies, as Azernour explains:
“You decide what you want to fetch during the conversation, whether it is a contract numbers, sentiment, or previous call context. You can cherry-pick what you pass to the next agent or feed it directly into a CRM.”
The handoff becomes a continuation rather than a reset.
Why Voice Training Matters
Most AI models are general-purpose, trained on broad datasets, and retrofitted for contact center use. Diabolocom’s virtual agent is trained on real contact center voice conversations, with ongoing fine-tuning from client-approved interaction data.
“The devil lies in the details,” Azernour says. “We work with partners who agree to share data because a system designed with their actual needs performs better.
“The retraining and fine-tuning on real contact center conversations is where the difference shows.”
As both a CCaaS provider and a telecom operator, Diabolocom also owns the infrastructure on which its AI runs, a combination that shows up most visibly in latency.
“We have best-in-class latency because of the low-level integration between the AI modules and the telephone system,” he says.
“And we are not dependent on other providers whose pricing is currently surging. We manage everything, so everything is predictable.”
Built for the People Running It
Advanced capabilities mean little if the people configuring the system need specialist support to use them.
Azernour is direct about who Diabolocom’s users are:
“They are not PhDs in AI. They are operational teams who know what the virtual agent should say and how it should behave.”
The platform reflects that, with tiered access that lets most users get started from a template within minutes, while giving technical users the same configuration options Diabolocom uses internally.
“They have the keys to the castle,” he says. “They just decide whether to go with the simple solution or get into the fine details.”
Full agentic orchestration, in which specialized virtual agents communicate autonomously and hand off between themselves without breaking flow, is Azernour’s stated priority for the year, alongside expanded language support and broader omnichannel coverage:
“We want to bring it to the table before anyone else, in an efficient and lightweight way, without an over-engineered system where you need a PhD just to build a virtual agent.”
For contact center leaders still running systems that deflect calls and hand off blank slates, that gap is closing faster than many might expect.
You can find out more about Diabolocom’s unique approach to AI by watching this interview with Rémi Guinier, Head of AI Product at Diabolocom.
You can also learn more about Diabolocom’s Virtual Agent for Contact Center by visiting the website.
