Enhancing CX Using Generative AI
When I first wrote this in 2023, ChatGPT was a few months old and the question was "should we even look at this?"
That question is settled. The majority of contact centres now use AI in some form, and executives are pushing hard to do more of it.
So the useful question in 2026 isn't whether to use generative AI to enhance CX. It's where it genuinely helps, and where it quietly makes things worse — because the gap between those two has turned out to be much wider than the hype promised.
21%
of frontline agents are excited about AI. 55% are neutral to negative. The tools aren't the hard part.
1 in 5
consumers who used AI for customer service got no benefit at all — about four times the failure rate of AI generally.
86%
still say human interaction is important to their experience of a brand. AI augments the human; it doesn't retire them.
ℹ️ About this update
This is my original 2023 piece, brought up to date. The practical uses still hold — but the industry now has three years of real deployment data, and it tells a more useful story than the 2023 optimism did.
The current figures throughout are drawn from our latest Australian Contact Centre Best Practice Report and the primary research collected on our CX Statistics page.
Start With the Reality, Not the Hype
Before the uses, the context that makes them work.
Deploying generative AI into a contact centre without it is how organisations end up in the 40% of Agentic AI projects Gartner predicts will be cancelled by 2027.
Sources: our CX Statistics and latest industry insights.
The single most important number is a sentiment number, not a capability one
The latest Best Practice Report found that only 21% of frontline agents are excited about AI, and 55% are neutral to negative — even though the report also finds the technology is largely making their jobs easier.
Read that twice. The tools work.
What isn't keeping up is the communication and change management around them. Every use below is delivered through a human being who has to actually adopt it — so the technology is the easy half. The adoption is the half that fails, and it's an engagement problem before it's a technology one.
None of this is an argument against using generative AI. It's an argument for using it deliberately: on the right jobs, with the humans brought along, and with the guardrails in place from day one.
Why Generative AI Is a Bit Different
Quick grounding, because the distinction matters for where it helps.
Traditional AI mostly processes and analyses existing data — it classifies, predicts, routes.
Generative AI creates new content: text, images, summaries, replies, whole conversations, using large language models. ChatGPT is the example everyone has already played with.
For CX, that shift is the whole story. It's the difference between a system that can route a customer to the right agent and one that can draft the reply, summarise the case, or hold the conversation.
That's a much bigger surface area — and a much bigger opportunity to either help or embarrass yourself. Related concepts worth understanding first: conversational AI, Knowledge Management Systems, and first contact resolution all sit underneath what follows.
Practical Ways to Enhance CX With Generative AI
These are the uses that have earned their place through real deployment — not the demo-ware. Each one works. Each one has a way it goes wrong, so I've been explicit about both.
Conversational AI for genuine self-service
Generative AI lets you build chatbots and virtual assistants that hold natural, contextual conversations rather than the rigid decision-tree bots customers learned to hate.
Done well, they resolve the simple, high-volume contacts instantly, at any hour — and 74% of consumers now expect 24/7 availability precisely because AI has made it possible.
✅ Where it works: the repetitive, well-defined enquiries that have a clear answer.
⚠️ Where it backfires: the moment you use it to wall customers off from a human. Removing the option to reach a person on demand is a reliable recipe for churn — 85% of customers say they'll drop a brand that can't resolve their issue.
Deflection that just delays the contact isn't deflection; it's a worse, angrier contact later, and it shows up in the metrics that matter as repeat contacts and falling first contact resolution. Build the escape hatch to a human first, then automate around it — the same principle that makes omnichannel work.
Personalisation at a scale humans can't match
Generative AI is genuinely good at reading a customer's history and context and tailoring the response — product suggestions, next-best-action, surfacing the right knowledge to an agent mid-call so the customer never has to repeat themselves.
✅ Where it works: when the personalisation is obviously in the customer's interest.
⚠️ Where it backfires: the privacy line. Only 39% of consumers trust companies to use their data responsibly, and misuse of personal data is now their top concern about AI-automated interactions.
Personalise with data the customer knows you have and would expect you to use. The moment it feels like surveillance, the trust you were building evaporates — and that's where a genuinely seamless digital experience is won or lost.
Augmenting agents — not replacing them
This is where the strongest, least glamorous returns are: generative AI sitting beside the agent — drafting replies, summarising a long case in a sentence, suggesting the next step, writing the wrap-up notes.
It measurably lifts productivity — Deloitte's 2026 research has 64% of service leaders reporting gains — and it removes the dull, repetitive work agents are usually glad to lose, freeing them for the complex enquiries that actually need a person.
✅ Where it works: as a co-pilot the agent controls.
⚠️ Where it backfires: when "augmentation" is quietly a headcount-reduction plan with better PR. Only 20% of service leaders have actually cut agents because of AI, and Gartner now predicts half of those who did will rehire by 2027.
Automation removes the easy contacts first — which is exactly the work new starters cut their teeth on — leaving a denser, harder, more emotional queue. Plan for that, or you'll pay to replace the agents you just made redundant.
Generating and maintaining knowledge
The quiet powerhouse. Generative AI is excellent at drafting, structuring and keeping a knowledge base current — and knowledge management has gone from a nice-to-have to a core capability, rated important by 95% of senior leaders.
A good knowledge layer is what makes everything else work: the bot, the personalisation and the agent co-pilot are all only as good as the content behind them.
✅ Where it works: AI drafts, a human verifies, and the same source feeds the bot, the agent and the customer-facing help.
⚠️ Where it backfires: letting AI publish unchecked. A confidently wrong knowledge article scales its error across every channel at once.
Generative AI writes the first draft; a person still owns "true".
Analysing every interaction, not a sample
This is the one most businesses are sitting on without realising it. Traditional Quality Assurance and Voice of the Customer programs read a tiny sample — often 1–2% of contacts — because a human can only listen to so many calls.
Generative AI can read, summarise and theme 100% of your calls, chats and emails: the emerging complaint before it becomes a trend, the process that's generating repeat contacts, the coaching opportunity buried in a transcript.
✅ Where it works: surfacing themes and outliers for a human to act on — turning every interaction into insight instead of a QA lottery.
⚠️ Where it backfires: using it as a surveillance stick on agents rather than a lens on the operation. Point it at the process, not the person.
Done right, it's the cheapest way to hear what your customers are actually telling you — the raw material behind our own CX statistics and the metrics you already track.
Proactive, personalised outreach
The best customer service contact is the one the customer never had to make. Generative AI makes genuinely personalised proactive outreach affordable at scale — the delivery delay flagged before they chase it, the renewal explained before it lapses, the outage acknowledged before the phone rings.
✅ Where it works: when the message is timely, relevant, and saves the customer effort.
⚠️ Where it backfires: when "proactive" becomes a polite word for more marketing. If the outreach serves you more than it serves them, it's noise, and it trains customers to ignore you.
The test is simple: would the customer thank you for it? If not, don't send it.
A necessary honesty about jobs
Everything above is about customer service — using generative AI to help the people on the front line do their job better. Used that way it isn’t a headcount story, it’s a quality one.
But I won’t pretend that’s the whole picture. When you also point generative AI at the underlying processes, systems and customer journeys, you get smoother journeys that generate fewer calls and enquiries in the first place — and over time, fewer contacts genuinely does mean fewer people needed to field them.
I’d rather be straight about that. The honest version: the customer wins — the best contact is the one they never had to make — the frontline job gets better, and the total volume of work slowly shrinks. Plan for all three.
Looking for the technology behind these?
Browse suppliers by what you're actually trying to do. We don't rank or recommend anyone — it's a starting point for your own research.
Browse the full technology directory →The Guardrails That Make It Work
Every use above shares the same failure modes. Get these right and generative AI is a genuine CX advantage; get them wrong and it's an expensive way to annoy people.
🚪 Always leave the door to a human open
86% of customers still rate human interaction as important, and 50% actively prefer brands that don't push generative AI at them. Use AI to make the human interactions better and rarer — not to prevent them.
🔍 Be transparent that it's AI
95% of consumers expect an explanation when AI makes a decision that affects them. Tell people when they're talking to a machine and what it's doing with their data. Hidden AI is a trust problem waiting to surface.
🧑🤝🧑 Bring your own people along
The 21% agent-excitement figure is the whole ballgame. Involve agents in the rollout, show them the tool removes drudgery rather than their job, and the adoption problem — which is where AI programs actually die — largely solves itself.
✅ Keep a human on "true"
Generative AI is fluent, not reliable. It drafts; a person verifies anything customer-facing. The failure rate — 1 in 5 AI service interactions delivering no benefit — is mostly unowned output meeting a real problem.
None of this is exotic. It's the same principle underneath every use and all three years of deployment data: generative AI is a tool for people, deployed by people, adopted by people.
The organisations getting a return treat it that way. The ones that end up cancelling treat it as a headcount line item.
Worth reading alongside why human interaction is becoming the premium skill and how to rethink automation so it actually saves you time.
Frequently Asked Questions
What is generative AI in the context of CX?
Generative AI is the branch of artificial intelligence that creates new content — text, replies, summaries, whole conversations — using large language models, rather than just classifying or routing existing data like earlier AI. In customer experience that means it can draft an agent's reply, summarise a case, power a conversational chatbot, analyse every interaction, or generate and maintain knowledge content. ChatGPT is the best-known example. The practical difference is surface area: it can act on far more of the interaction than traditional AI, which is both the opportunity and the risk.
Will generative AI replace contact centre agents?
The evidence so far says no — it changes the job rather than removing it. Only about 20% of service leaders have reduced headcount because of AI, and Gartner predicts half of those will rehire by 2027. What AI removes first is the simple, repetitive work, which leaves a denser, more complex, more emotionally demanding queue for the humans — so skill requirements and the case for good coaching actually go up. Treat AI as agent augmentation, not agent replacement, and both the economics and the customer outcomes are better.
What's the biggest mistake businesses make with AI in CX?
Using it to wall customers off from a human. The technology isn't perfect, 86% of people still value human interaction, and 85% will abandon a brand that can't resolve their issue — so a chatbot with no escape hatch to a person doesn't deflect the contact, it delays it and makes it angrier. The second biggest mistake is deploying to hit an executive AI target (91% of leaders report that pressure) without bringing agents along, which is why so many programs stall on adoption rather than technology.
How do I use generative AI without losing customer trust?
Three things. Be transparent that AI is involved and what it's doing with data — 95% of consumers expect an explanation when AI makes a decision about them. Only personalise with data the customer knows you hold and would expect you to use, because just 39% trust companies to use their data responsibly. And keep a human verifying anything customer-facing, because generative AI is fluent but not reliable. Trust is the currency here, and it's far easier to spend than to earn back.
Where should a contact centre start with generative AI?
With agent augmentation, not a customer-facing chatbot. Putting AI beside your agents — drafting replies, summarising cases, writing wrap-up notes — delivers measurable productivity (64% of leaders report gains) with far less risk than a public-facing bot, and it builds internal confidence and knowledge before you point AI at customers. Pair it with a solid knowledge base, since everything else depends on the content underneath it. Then, once your people trust it, extend to self-service.
Where to Next
Related Articles
Summary
In 2023 the question was whether to touch generative AI. In 2026 it's where to point it.
A handful of uses have earned their place: conversational self-service, personalisation at scale, agent augmentation, knowledge generation, analysing every interaction, and proactive outreach.
Every one of them helps when it's deployed for the customer and adopted by your people — and backfires when it's deployed to hit an AI target or to wall customers off from a human. The data is unambiguous on that: the technology is largely ready, and the failures are almost all human — trust, transparency, and the 55% of agents who aren't yet sold.
Generative AI will play a big role in the CX future. It just won't do it on its own — and the businesses that remember that are the ones it will actually reward.
- About the Author
- Latest Articles
After more than 30 years in contact centres and CX, I’ve built, run, outsourced and transformed just about every kind of customer operation there is — and I’m still as fascinated by getting it right as I was on day one. That’s what I love about this profession: there’s always more to learn.
As founder and CEO of ACXPA, I’m on a mission to raise the standard of customer experience as a profession — Australian-founded, and increasingly on the global stage. Through ACXPA and our specialist training arm, CX Skills, we help businesses and their frontline teams deliver customer experiences that actually work — on the phone, digital and in person — with the skills, standards and insight to back it up.
