From IVR to AI agents: 40 years of promises that call centres would disappear

In 2025, it feels like every week brings a new headline about AI “replacing” support agents.

An AI agent that can handle 80 percent of chats.

A telco planning to “automate thousands of roles”.

Startups claiming that a single bot can do the work of an entire team. 

If you work in customer support, this storyline is not new. The industry has heard it before.

In the 1980s, it was Interactive Voice Response (IVR) systems.

In the 2000s, it was chatbots.

Now it is large language models and AI agents.

Each time, the technology genuinely improved parts of the experience. Each time, business cases predicted massive headcount reductions. And each time, the reality was more nuanced: contact centres did not vanish, they evolved.

This article walks through that history as a timeline, then looks at what it really tells us about AI in customer service today.

1960s–1980s: the birth of the call centre, and the first automation wave

Modern customer support operations started to take shape in the 1960s. Private Automated Business Exchange (PABX) systems made it possible to centralise incoming calls and route them to rows of agents, an early version of today’s contact centres. 

Through the 1970s and early 1980s, call centres scaled as telephone became the dominant service channel. Technology focused on:

• automatic call distribution

• call recording

• basic reporting

The objective was straightforward: handle rising call volumes more efficiently.

IVR arrives – and promises to shrink the workforce by half

In the late 1980s and early 1990s, Interactive Voice Response systems started to spread. IVR allowed callers to interact with recorded menus using touch-tone or spoken input instead of talking directly to an agent.

Business cases were aggressive. Some projected a 40 to 60 percent reduction in agent headcount thanks to self-service menus. 

Vendors framed IVR as a revolution that would:

• automate routine queries

• slash staffing costs

• keep customers happy with fast self-service

For a while, this sounded plausible. IVR did deliver clear benefits: it routed calls more intelligently, handled simple requests like balance checks or store hours, and helped large organisations absorb peak demand without hiring endlessly.

But something else also happened.

Studies and industry surveys over time showed that IVR quickly became one of the least liked service channels. For many customers, “IVR” turned into shorthand for long menus, dead ends, and the feeling of being trapped in a maze before finally reaching a human. 

Instead of eliminating agents, IVR:

• filtered out the most repetitive calls

• shifted agents toward more complex, emotionally loaded, or exception-heavy cases

• raised expectations of what “good” phone support should feel like

The call centre did not disappear. It specialised.

1990s–2000s: email, offshoring, and the first chatbots

By the 1990s and early 2000s, customer support was changing on several fronts at once:

• Email became a standard service channel.

• Offshoring to large BPO hubs expanded global contact-centre headcount.

• Early web self-service and FAQs started to appear. 

Then came the next promise: chatbots would finally replace human agents.

The first chatbot boom

Rule-based chatbots and simple scripted virtual assistants gained traction in the mid-2000s and early 2010s. They lived on websites, in banking apps, and later inside messaging platforms.

The story was familiar:

bots would deflect a huge percentage of contacts, work 24/7, and make large teams of agents unnecessary.

In practice, early chatbots:

• worked reasonably well for very narrow intents (order status, password reset, basic FAQs)

• struggled with anything outside their script

• handed off a significant share of conversations to human agents when customers got stuck or frustrated 

Global call-centre employment remained large throughout the 2000s and 2010s, even as automation spending increased. Automation changed the content of the job far more than the existence of the job itself.

Meanwhile, customer behaviour sent a clear message:

for complex, emotionally charged, or high-stakes issues, people still wanted to talk to another person.

2010s: from call centres to contact centres

The 2010s were less about a single “job-killing” technology and more about channel expansion.

Phone support was joined by:

• live chat

• in-app messaging

• social media

• WhatsApp and other OTT channels

The call centre morphed into the contact centre, reflecting the fact that customer issues now arrived from everywhere, not just the phone line. 

Behind the scenes, AI and automation did make a serious impact, but mainly in supporting roles:

• smarter routing

• speech analytics

• quality monitoring

• agent-assist recommendations

The narrative of “automation as a co-pilot” started to emerge. But at the same time, marketing language around chatbots still promised significant headcount reduction.

Reality stayed stubborn: voice remained one of the most widely-used service channels, and human agents were still central to the customer experience. 

2020s: generative AI and the third big wave of disruption claims

Today’s hype cycle is powered by large language models and generative AI.

Vendors now talk about AI agents that:

• handle 70 to 90 percent of customer contacts

• work across voice, chat, email, and messaging

• make traditional contact centres “obsolete”

There are real innovations here. New AI systems are much better at understanding natural language, keeping context, and generating fluent responses than the bots of the 2000s.

Companies experimenting with generative AI are already seeing:

• higher containment on simple queries

• shorter handling times

• richer analytics for training and quality

At the same time, there are echoes of the past.

A recent Reuters report on India’s IT and contact-centre sector highlighted startups claiming to have automated thousands of roles, with ambitions to cover 80 percent or more of routine contacts. Yet the article also notes that a majority of surveyed customers still prefer human agents for many interactions, and that the transition is creating real anxiety for workers. 

Meanwhile, newer research on IVR and automation shows that frustration spikes when customers are forced through automated flows for issues that are complex or emotional, and that IVR remains one of the least-loved channels overall. 

In other words, the technology has changed, but the human response has not.

The pattern that keeps repeating

Looking across IVR in the 1980s, chatbots in the 2000s, and AI agents today, a clear pattern emerges:

1. Overconfident forecasts

• IVR business cases in the 80s projected 40–60 percent staff reductions. 

• Early chatbot advocates talked about bots “replacing” large portions of frontline teams. 

• Today, some AI vendors again talk openly about reducing headcount by 70–80 percent in certain markets. 

In each cycle, the industry narrative overshoots what technology can reliably handle in real-world conditions.

2. Automation shifts work, it does not erase it

In practice, automation has:

• removed low-complexity tasks from human queues

• surfaced more edge cases, emotionally charged situations, and multi-step problems to human agents

• raised the skill bar for those remaining roles

Agents are now expected to do less repetitive work, but the work they do is harder, more nuanced, and more consequential for the customer relationship.

3. Customer preference acts as a brake

Across decades, customer satisfaction consistently falls when:

• automation is pushed as a rigid front door

• there is no clear, fast path to a human

• bots are allowed to “fight” escalation instead of enabling it 

Customers do not object to automation by default. They object when it gets in the way of resolving important issues with a person they can trust.

Why “this time is different” still has limits

Support leaders are right to explore AI. The gains are real. But history helps to explain why the idea of fully autonomous support deserves skepticism.

Four structural factors keep humans in the loop:

1. Complexity

   Real customer journeys cross systems, products, policies, and edge cases that no single model can perfectly encode.

2. Emotion and trust

   Refunds, medical questions, fraud alerts, travel disruptions, insurance claims – these are not just “tickets”. They are stressful life moments.

3. Accountability

   Someone has to own the judgment call when a case falls into a grey area. That responsibility does not disappear because a model suggested an answer.

4. Regulation and risk

   In finance, healthcare, insurance, and public services, the risk of a wrong automated decision is high. Automation here tends to be tightly constrained, not fully free-running.

For all these reasons, the most realistic future for customer support is not a contact centre without humans, but humans supported by increasingly capable automation.

What actually changes for support teams

Where AI does profoundly change things is in the profile of the job and the skills agents need.

If AI handles most of the straightforward volume, human agents are left with:

• the complex, multi-step problems

• the unhappy or anxious customers

• the situations where the bot guessed wrong

• the cases that touch on legal, financial, or safety concerns

That means the real pressure point moves from “how many people can we hire” to:

• how fast can we onboard people to this higher bar

• how well can we train them to work with automation instead of against it

• how consistently can we coach them on judgment, tone, and escalation

Which is exactly where most organisations are still under-invested.

Training in an AI era: the quiet gap behind the hype

Look back at IVR and early chatbots and another common thread appears:

most of the focus went into the tech itself, far less into training people around it.

• Agents often discovered new IVR flows live on calls.

• Bot escalation logic was unclear, so customers arrived angry and confused.

• Supervisors had little visibility into how automation shaped conversations.

The industry cannot afford to repeat that pattern with generative AI.

In an AI-heavy support environment, agents need regular practice in:

• taking over from an AI mid-conversation without losing context

• correcting wrong or outdated suggestions tactfully

• explaining to customers what the AI did and what will happen next

• navigating exceptions that automation cannot safely handle

You cannot build these skills with a static slide deck or an annual training.

This is where simulation-based training platforms like Smart Role come in: they let teams rehearse realistic, AI-era scenarios before they play out with real customers, and give structured feedback on the skills that matter most.

What leaders should take from 40 years of “the end of the call centre”

If there is one lesson from IVR in the 80s and chatbots in the 2000s, it is this:

Every major automation wave changed the shape of customer support,
but none of them erased the need for skilled humans.

Generative AI will be no different.

The organisations that come out ahead will not be the ones that gamble on a fully autonomous contact centre. They will be the ones that:

• use AI aggressively where it fits

• stay honest about its limitations

• and invest just as heavily in the humans who handle everything that falls outside the happy path

History is not an argument against AI.

It is a reminder that hype fades quickly, but customer expectations do not.

About the author

Thibaut Martin is the co-founder and COO of Smart Role, an AI-powered platform helping customer support teams build real skills through realistic scenario-based training. Before launching Smart Role, Thibaut spent nearly a decade in customer experience roles at Google and Otrium, leading teams, scaling operations, and navigating multiple waves of automation hype from the inside. He now works with global brands and BPOs to help their agents ramp up faster, improve performance, and deliver more confident, human customer support in an AI-driven world.

FAQ

What is Smart Role?

Smart Role is an AI-driven training platform for customer support teams. It lets companies create realistic simulations, train agents on complex scenarios, and evaluate skills with automated feedback.

How does Smart Role improve training?

Instead of static onboarding or long PDFs, Smart Role lets agents practice real customer situations through dynamic role-play. The platform coaches them on tone, accuracy, empathy, and decision-making.

Who uses Smart Role?

Smart Role is used by BPOs and in-house support teams in industries like travel, fintech, retail, and marketplace platforms. Clients include Bumble, Etraveli Group, and Pontica Solutions.

Can Smart Role work alongside AI agents and chatbots?

Yes. Smart Role is designed for hybrid environments where automation handles simple queries and humans manage complex cases. The platform helps agents build the skills needed to step in when AI reaches its limits.

Is Smart Role secure?

Smart Role is SOC 2 Type 2 and ISO certified. All data is processed securely and hosted on enterprise-grade infrastructure.

Sources

• CMSWire, “AI in Contact Centers: Leveraging Lessons From the Past”

• Parloa, “IVR in Contact Centers: Why It No Longer Works”

• RABYIT, “Evolution of Call Centres”

• Jet Interactive, “How AI-Powered IVR Systems Are Transforming Call Centres”

• Reuters, “Meet the AI chatbots replacing India’s call-center workers”

• Tafaseel, “The Evolution of Contact Centers”

• Crafter.ai, “Exploring Technology in Call Center Automation”

• Wikipedia, “Interactive Voice Response”

• Telegraph, “Sky to replace thousands of call centre workers with chatbots”

• TechTarget, “History and Evolution of Contact Centers”

• CallCenterStudio, “IVR Optimization for Personalized CX”

• CallMiner, “A Comprehensive History of AI in the Call Center”

• NSUWorks, “The Evolution of Technology in Call Centers”

• SelectCall, “The Evolution of Call Centres Through History”

• Infomineo, “AI Chatbots in Customer Service: Can They Truly Replace Humans?”

• VCC Live, “From Call-Based to Multichannel”

• Teneo.ai, “The State of Interactive Voice Response Technology”

• Botsplash, “Chatbots: A Brief History Part II”

• Teledirect, “History of Call Center Technology”