AI didn't suddenly show up in contact centres. So, what's new?
How much of today's innovation, from traditional automation to agentic AI, is genuinely new? And how much is the latest chapter in a story that has been unfolding for years?
Everything Is AI Now
Follow the CCaaS industry today, and you could easily think artificial intelligence arrived only a few years ago. Almost every announcement now comes with an AI label: generative, conversational, predictive, agentic or orchestration. The language seems to change almost as quickly as the technology.
I've worked through several generations of contact centre technology, and I keep coming back to the same question: When did AI really arrive, and how much of what we're seeing now is truly new? AI didn't suddenly appear with ChatGPT. The building blocks had been developing in contact centres for years. What has changed is their sophistication and, increasingly, their ability to act rather than simply analyze. Looking back helps put today's claims in perspective.
Before AI, There Was Automation
Long before anyone was talking about generative AI, contact centres were already automating customer interactions. In the 1970s, Automatic Call Distributors (ACDs) routed calls to available agents, while Interactive Voice Response (IVR) systems let customers work through menus, get information and complete simple tasks. Skills-based routing and computer telephony integration (CTI) later improved call distribution and connected phone systems with customer data and business applications. Most of this came before cloud delivery, but it laid the foundation for today's CCaaS platforms.
Was that AI? For the most part, no. These systems followed set rules. They could carry out instructions well, but they didn't learn or think for themselves. That distinction still matters because automation and AI are often treated as the same thing, especially in marketing.
AI Was Already Emerging
In the 1990s and 2000s, speech recognition and better statistical technology began to change how contact centres worked. Early systems picked out specific keywords; later versions supported more natural conversations. In 1992, AT&T began using speech recognition for collect and calling card calls, asking callers what kind of call they wanted to make. The system was expected to replace thousands of operator jobs. Around the same time, predictive analytics and machine learning created new ways to route calls, forecast staffing, analyze speech and automate quality management. Instead of simply following rules, these systems could find patterns in past data and use them to predict what might happen next.
Anyone who wrestled with an early speech-enabled IVR will remember its limits, particularly with accents, background noise or an unexpected question. Even so, it marked an important step. Contact centre technology was moving from following instructions to making predictions. The features didn't all arrive at once, and they weren't equally advanced. Some relied on basic statistics rather than machine learning, but together they prepared the ground for the AI now built into CCaaS.
Conversational AI Wasn't Born Yesterday
As natural language processing improved, chatbots and voicebots became better at understanding what customers wanted. Some learned from data, while others were still scripted systems behind a conversational interface.
In July 2018, Google announced Contact Center AI, which brought together virtual agents, real-time agent assistance and conversation analytics. Its virtual agent and agent assist capabilities became generally available in November 2019, three years before ChatGPT. Conversation analytics followed later as CCAI Insights.
Contact centre vendors were heading in the same direction. Amazon Connect launched in 2017 and was soon paired with Amazon Lex for conversational self-service. Contact Lens added AI-driven speech analytics in 2020. By early 2018, Genesys was promoting Kate as the face of its AI and automation work, and it had acquired Altocloud for machine learning that predicted customer behaviour across the journey. That technology became the foundation for Genesys Predictive Engagement.
This history matters because many features now associated with AI-powered CCaaS were already being developed and used. Cloud delivery later made them easier for organizations to adopt without running all the underlying technology themselves. The tools have come a long way, but the goal has barely changed: solve problems quickly, help agents do their jobs and give customers a good experience. Customers generally don't care whether they're dealing with AI. They care whether it solves the problem. If it doesn't, then the technology suddenly matters a great deal.
Then Generative AI Changed Expectations
When ChatGPT became publicly available on November 30, 2022, generative AI moved into the spotlight. For CCaaS, large language models created better ways to understand and generate language, summarize conversations and work with unstructured information. Agent assistance became more useful, with real-time suggestions, relevant information and practical guidance. Automated summaries reduced after-call work, while conversational self-service became more flexible when customers asked something unexpected.
Those are real improvements, although many address problems that existed long before generative AI. Take automated call summaries. Earlier technology could pull out keywords or fill in records, but generative AI can produce a clear summary that captures context and intent with far less setup. The business task isn't new. The improvement lies in how well and how flexibly the technology can do it.
One Customer Problem, Three Generations of Technology
Take a customer contacting a retailer about a delayed delivery. With a traditional IVR, the customer would work through several menus, enter an order number and hear a recorded update. If that didn't answer the question, the call would usually go to an agent.
With conversational AI, the customer can explain the problem in their own words. A virtual assistant might find the order, check the delivery information and explain the delay. A more complicated issue may still need an agent or a predefined process.
Agentic AI promises something more. With the right permissions and system connections, an AI agent could investigate the delay, consider the available options, arrange a replacement or refund within set limits, update the CRM and tell the customer what it did.
The difference is more than a smoother conversation. AI can move from providing information to taking the steps needed to resolve a problem. Earlier automated systems could also complete transactions through set integrations, but agentic AI is meant to be more flexible about which actions it chooses and the order in which it takes them. That would be meaningful progress, provided it works reliably outside a controlled demonstration.
When Does Innovation Become AI Washing?
AI washing isn't a new concern. It happens when established technology is repackaged with a new AI label, or when future capabilities are described as if they're already available. CCaaS is particularly vulnerable because predictive routing, speech analytics, automated quality checks, workforce optimization and conversational self-service have been around for years. AI has improved many of them, sometimes significantly.
Still, does every improvement deserve to be called revolutionary? When a product is described as agentic, can it genuinely plan and complete a task, or is it following a set workflow through a more polished chat interface?
Improving established technology is valuable in its own right. Small changes can produce real business benefits, and a sensible mix of fixed workflows and AI may be the best answer. The problem begins when buyers can no longer tell the difference between an established capability, a genuine advance and a future promise. AI washing isn't simply a question of whether a product contains AI. It's about whether the claims match what the product can do today. For organizations comparing CCaaS platforms, that distinction matters.
Is Agentic AI Running Ahead of Reality?
A polished demo of an AI agent completing a customer transaction can be impressive. Handling thousands of unpredictable, real-world interactions is a much harder test.
The analyst forecasts reflect both the promise and the risk. In March 2025, Gartner predicted that agentic AI would autonomously resolve 80% of common customer service issues by 2029. Three months later, it predicted that more than 40% of agentic AI projects would be cancelled by the end of 2027 because of rising costs, unclear business value or inadequate risk controls.
Real contact centres are messy. Customer information may be missing, a business application may be unavailable, or a transaction may need an extra approval. An AI agent has to recognize when it lacks the information or authority to act and when it should bring in a person. It also has to respect security, privacy and legal requirements.
These details determine whether an organization can trust autonomous AI with customer transactions. The question isn't simply whether an AI agent can complete a task. It is whether the agent can do so reliably and stay within the rules the business has set.
The Bigger Change May Be Architectural
As AI becomes more independent, choosing a CCaaS platform is no longer only about routing, channels and workforce management. Organizations also have to decide where customer journey orchestration belongs, how AI connects with CRM and other business applications, and which systems can carry out transactions. These decisions reach well beyond the contact centre. The platform still matters, but understanding how AI will work across the business may matter just as much.
Are We Measuring the Right Outcomes?
Contact centres have long tracked measures such as average handling time, service levels, abandonment rates and agent use. AI may improve those numbers, but speed isn't always the most important outcome.
If AI handles the straightforward questions, human agents may spend more time on complex or sensitive issues. Their average handling time could rise even as the customer experience improves. An AI interaction also isn't a success if the customer has to come back because the problem wasn't resolved the first time.
Klarna learned this in public. In early 2024, the company said its AI assistant was handling most customer service chats and doing the equivalent work of 700 agents. By May 2025, its CEO acknowledged that the emphasis on cost had lowered quality, and the company began recruiting human agents again so customers could always reach a person. That experience suggests we may need to pay more attention to first-contact resolution, customer effort, successful outcomes and the total cost of solving a problem. The real test of AI is what improves for the customer and the business, not how widely the technology has been deployed.
So, What's Really Changed?
Looking back at the history of contact centre technology, I see steady progress rather than a single revolution. Rules-based automation led to predictive systems, then conversational AI and generative AI. Now we're exploring systems that can make decisions, work across business applications and act on a customer's behalf.
Each stage builds on the last, but agentic AI could have a greater impact than earlier shifts. The technology is moving beyond helping agents or supplying information. It may be able to resolve more complicated customer issues on its own. That creates opportunities to improve the experience, along with harder questions about oversight and responsibility.
The CCaaS industry has plenty to be excited about. At the same time, when AI appears in almost every announcement, buyers need to separate improvements to existing features from genuinely new capabilities and promises that haven't yet been proven. Perhaps the biggest change isn't only what AI can do. It's what organizations are now being encouraged to expect from it.