The term "AI" for artificial intelligence is one of the hottest buzzwords in technology. Like most buzzwords, it's being overused, misused, and used by some people just to sound like they're in-the-know. But also like most buzzwords, there's a kernel of important truth and value behind it.
ECG focuses on this landscape from the perspective of unified communications (UC) or voice service providers, and what they can do to compete and to add new features and functions to their network. This review of some AI technology basics and analysis of some recent products is intended to help enlighten and inform UC and voice service providers in their growth and development.
What Is AI in Unified Communications & Enterprise Communications?
AI is commonly defined as the capability of a system or algorithm to imitate intelligent human behavior. While the term is often used loosely, it's important to distinguish between simple automation (like macros or rule-based logic) and more advanced machine learning (ML) applications that adapt based on data patterns.
In enterprise communications, AI is being used to:
-
Solve routine customer issues without human escalation.
-
Assist contact center agents in real-time.
-
Enable live translation in global teams.
-
Fully automate end-to-end customer interactions.
However, it all starts with data quality. As the phrase goes, “garbage in, garbage out.” Whether using transcripts, call metadata, or logs, training data must be relevant and vetted for ML models to produce accurate and valuable outcomes.
There are three major types of machine learning at play:
-
Supervised Learning: The system learns from labeled data (e.g., good vs. bad call transcripts).
-
Unsupervised Learning: The system finds patterns in unlabeled data (e.g., fraud detection).
-
Reinforcement Learning: The system learns via feedback and aims to maximize rewards (e.g., adapting to customer behavior over time).
AI in Action: Tools and Vendors Shaping the Future
The real value of AI emerges when it solves specific operational problems. Let’s take a closer look at how leading telecom vendors are building AI into their products – and the strategic questions they raise.
1. NICE InContact – Augmenting Human Agents with AI
NICE offers AI tools to assist agents in real time – sometimes without needing to record the actual audio. Their “Enlighten Copilot” attempts to monitor calls live, analyzes sentiment (the mood of the customer), suggests de-escalation strategies, and surfaces knowledge base articles. Notably, NICE reports that their platform operates without retaining audio files, potentially avoiding legal complications in jurisdictions like California where call recording laws are strict. This system aims to better empower customer service agents while preserving compliance and reducing customer friction.
Product Strategy Question: Do you have the data available with which to equip your customer service agents? I.e., do you have knowledge base articles and good call content you can use to train the database?
2. ASAPP – Fully Autonomous Customer Service with Human Oversight
ASAPP takes a bolder approach: they aim to handle up to 90% of calls entirely with AI. Their “Generative Agent” system manages the full customer interaction, only alerting a human supervisor when necessary. Think of it as a contact center AI operating with a human-in-the-loop model – supervisors oversee multiple calls and step in when needed. ASAPP’s system necessarily integrates directly with enterprise backends, improving via active learning and analyzing escalated cases to refine future responses.
Product Strategy Question: When are your users/customers going to be better served by a voice-based interface to enterprise backends, and when would they be happier using an app-based or web-based interface to the same backend platforms? Most humans are calling into contact centers to talk to humans.
3. DeepL – Real-Time Voice Translation for Multilingual Teams
DeepL, known for its deep learning-based translation engine, now integrates with Microsoft Teams to provide real-time voice translation (e.g., German to English). Its standout features include:
-
Dictionary lookups
-
Alternative phrasing suggestions
-
Contextual clarification tools
While currently integrated only with Microsoft Teams, this kind of real-time translation is poised to enter broader VoIP systems. We would love to see DeepL publish a SIP interface allowing interoperation with conventional calls.
4. Alianza – Fraud Prevention in Live Calls via AI Whispers
Alianza is pioneering AI usage in fraud prevention during live calls. By duplicating the call audio stream and analyzing it in real-time, their system can inject an alert (a "whisper") to the called party if suspicious behavior is detected. For example, if the calling party appears to be impersonating a bank to solicit personal information, the system can warn the receiver mid-call.
This approach leverages scalable cloud infrastructure to do what’s infeasible for smaller voice providers – processing live audio at scale with AI and offering new layers of protection. Alianza positions this as part of a concept they call "Telco 4.0," where voice telecom takes advantage of AI-based technology to assist and support the user.
The Importance of Data Quality
You should always ask where the data is sourced when looking at any AI tool. Nearly all of them are based on Machine Learning. AI’s effectiveness hinges not just on clever algorithms, but on the quality and origin of the data it’s trained on:
-
Externally sourced data (e.g., websites, textbooks, or publicly scraped text) can be useful, but may not match specific enterprise needs. Companies like NICE InContact can actually provide "Small Language Model" or fine-tuned Large Language Model (LLM) systems.
-
Internally sourced data (e.g., transcripts, logs, customer interactions) is more relevant but must be properly labeled and vetted. For products using your data in this way, you need to be able to provide the data in a useful format to train the system. You can also be prepared for the costs of training and fine-tuning systems to adjust them as the data changes.
Vendors differ in whether they require access to recordings or work with transcriptions, whether training is manual or automatic, and whether AI operates pre-call, mid-call, or post-call.
Understanding where the data comes from, who owns it, and how it’s used is crucial for any organization evaluating AI tools.
AI Strategy Starts With the Right Questions
At ECG, we’ve spent years helping voice service providers evaluate, implement, and optimize new technologies – and AI is no exception. The vendors and tools covered in this post represent just a few of the many directions AI is taking in the telecom world, from real-time agent assist to fraud prevention and translation.
What matters most is applying the right kind of AI to solve the right kind of problem, and knowing the limitations and requirements of the models behind it. Whether you’re considering tools that analyze call sentiment or deploying systems that interact directly with your users, the decisions around data sourcing, compliance, and system integration are critical.
If you're exploring AI for your voice network and need a trusted partner to help you navigate the options, get in touch. We’ll help you make informed decisions, avoid the hype, and put your resources toward solutions that actually move the needle.
