The Largest Online Community for Contact Centre Professionals
The Largest Online Community for Contact Centre Professionals
Close
Published On: 12th Oct 2026 08:00
Customer experience automation (CXA) can help contact centre leaders deliver faster, more consistent, and more personalized customer journeys. But automation can just as easily create new problems if it is introduced without considering the end-to-end customer journey.
So, what should you avoid when automating customer experience? We asked our panel of technology experts for the costliest mistakes out there right now.
Some enquiry types are suitable for automation, whereas other types of interaction are less so. At the base of the enquiry pyramid are high-volume, routine, and straightforward enquiries, which are the most suitable for automation (and are probably already automated).
As you go up the pyramid, you’ll find lower-frequency, higher-complexity use cases, or those requiring greater tact or sensitivity, so their suitability for automation decreases.
For many types of enquiries, such as those that are complex, urgent, or emotional (CUE), the person making the contact is doing so to speak to a human.
Fundamentally, if the automation is being seen as preventative, that will create frustration. Equally, the same enquiry can be suitable to automate for some customers but not others, such as older or more vulnerable people.
Contributed by: Martin Taylor, Co-Founder and Deputy CEO, Content Guru
A common CXA mistake is automating conversations without automating the work.
A bot that can’t authenticate, update a record, or process a refund becomes a gatekeeper, forcing customers to repeat themselves to an agent.
Real automation connects to the systems where work gets done, so journeys are completed end to end.
Too often, businesses measure deflection instead of resolution. You can contain 100% of interactions in self-service and still have unhappy customers.
Measure outcomes instead! Did the customer get what they wanted? CSAT is the true test of whether your strategy is working.
Too many businesses also trap customers. Escalation to a human should be easy, and the handoff should carry full context so no one starts over.
Contributed by: Matthew Clare, VP, Product Marketing, UJET
Poorly designed escalation from automated to human agents is one of the fastest ways to destroy the trust that CXA is supposed to build.
When a customer has already explained their issue to a virtual assistant and is then forced to repeat everything to a human agent, the automation has made the experience worse, not better.
The handoff must be seamless: full conversational context, intent data, and any information already collected should transfer to the agent automatically, so the interaction continues rather than restarts.
Contact centres that invest as much in designing the escalation experience as they do in the automation itself consistently see higher satisfaction scores. The moment automation reaches its limit is the moment that defines whether a customer stays or leaves.
No matter how sophisticated automation becomes, there will always be customers who need to speak to a human, whether due to complexity, vulnerability, or simply personal preference.
Making that option difficult to find, or removing it altogether, is one of the most damaging mistakes a contact centre can make.
Research consistently shows that customers are willing to engage with automation for routine tasks, but they expect a clear and easy path to a human agent when the situation demands it.
The goal of CXA should never be to eliminate human contact; it should be to ensure human agents are available for the interactions where they add the most value. Automation and human support are complementary, not competing.
Contributed by: Jonathan Kershaw, Head of AI and Contact Centre Experiences, Vonage
Contact centres should avoid assuming that every customer interaction is a good candidate for automation. Simple, repetitive, and predictable requests are often well suited for automated experiences because customers typically value speed and convenience in those situations.
However, complaints, cancellations, complex problems, and emotionally sensitive conversations often require empathy, judgement, and human problem-solving.
Over-automating these interactions can leave customers feeling trapped in a process when they need someone who understands the full context of their situation.
The goal should be to identify where automation adds genuine value and where human involvement creates a better outcome.
Effective CX automation should make it easier for customers to access the right level of support while allowing employees to focus their time on interactions where their expertise matters most.
Contact centres should avoid treating CX automation as a project that can be implemented once and left unchanged. Customer behaviour evolves, new issues emerge, and even well-designed automated processes can create unintended friction over time.
Organizations need a continuous feedback loop to understand where customers experience frustration, where automated journeys are breaking down, and where agents repeatedly step in to resolve problems.
Conversation intelligence can be especially valuable because it allows businesses to analyse customer interactions at scale and identify recurring patterns that may not appear in surveys or isolated feedback.
By continuously monitoring these signals, contact centres can refine their automation strategy over time and ensure technology is removing friction while preserving human interaction where empathy, judgement, and problem-solving remain essential.
Contributed by: Jonathan Kenu Escobedo, Customer Success Manager, MiaRec
The most successful CXA implementations we’ve worked on didn’t start with “what can we automate?” but with a more important question: “what should stay human?”
CXA done well is invisible. It removes friction before customers feel it. It surfaces context so agents don’t have to ask the same question twice. It handles the routine, so people can focus on the personal.
The mistake isn’t automating too little or too much. It’s automating without understanding the customer journey, and whether improving efficiency in one part of that journey could come at the expense of trust elsewhere.
Organizations getting this right measure success through loyalty, not just deflection. They design automation around the moments that matter, rather than simply the moments with the highest volume.
We’ve learned through years of CX transformation that technology alone doesn’t create great customer experiences. It amplifies the experience you’ve already designed.
Automate a broken experience, and you just break it faster!
Contributed by: Agam Kohli, Senior Technical Solution Consultant, Route 101
You score your agents on how they handle the customer. Your bots, instead, get a containment rate, which only tells you that the conversation stopped before it reached a real person.
So, a bot that actually sorted the customer’s problem and a bot that just wore the customer down until they gave up look pretty much the same, if we’re going by containment rate.
You’ll probably see the difference later, though. That “contained” interaction will reappear as a repeat contact, or a complaint grounded in context that your agents have no awareness of.
Automated conversations need their own scorecard, because the evaluation questions are different. You need to know if the bot understood what it was being asked, and if it followed handover protocol or just exhausted the customer.
And while the scorecard might be different, it’s better to keep it all on the same platform. It’s all part of the same customer experience in the end.
Once you’ve scored both elements, bot and human agent, you can see which conversations automation is genuinely resolving and which ones it’s just kicking down the road. That second group is your next CX priority.
Contributed by: Derek Corcoran, CEO, ScorebuddyCX
Customers may interact through WhatsApp, webchat, email, voice, or other channels. What matters is whether the organization maintains the context of the interaction.
If customers are forced to repeat information when moving between channels, much of the benefit of automation is lost.
A well-designed CXA strategy should preserve relevant context throughout the journey, including when an automated interaction transitions to a human agent.
AI is only as reliable as the knowledge available to it. Customer-service information is often spread across FAQs, product documentation, policies, procedures, and legacy systems – and may be incomplete, outdated, duplicated, or contradictory.
If these inconsistencies are not addressed, AI can generate responses that sound convincing but are not necessarily correct.
A more powerful AI model cannot compensate for poor or contradictory knowledge. Knowledge must be accurate, current, structured, consistent, and continuously maintained as products, policies, and processes evolve.
A CXA strategy should match the AI architecture to interaction needs. LLMs (large language models) suit complex, open-ended, or less predictable conversations where broad language capabilities and reasoning add value.
Many customer-service use cases involve defined domains, predictable intents, and controlled knowledge. Here, small language models (SLMs) can provide a focused approach that complements the broader capabilities of LLMs.
Combined with high-quality knowledge, retrieval-augmented generation (RAG), business rules, and guardrails, specialized SLMs can deliver predictable, relevant responses.
Depending on the use case and deployment, they may also offer practical advantages around real-time voice latency, generated content and response control, infrastructure cost predictability, and data and security flexibility.
SLMs can also be deployed in private or on-premises environments when data governance, security, or operational requirements call for greater control over where sensitive customer data is processed.
Rather than asking “SLM or LLM?”, organizations can select (or combine) the architectures that best support each customer-service use case.
Contributed by: Walid Karam, Presales Director, Enghouse
The biggest mistake I see is automating the queue rather than the question. Teams bolt a chatbot onto every inbound channel, declare the project complete, and wonder why satisfaction dips.
Automation works when it handles the things customers don’t want to wait for — a balance check, a delivery update, a password reset. It falls apart the moment it tries to contain a complaint or a nuanced billing dispute.
A second trap is building the bot in isolation. If your virtual agent can’t hand the full conversation to a human with context intact, the customer repeats themselves and trust evaporates.
When Buckinghamshire Council rolled out Zoom Virtual Agent, it kept advisor transfers low precisely because the handover carried context forward.
Automate the repetitive. Protect the relationship. Test from the customer’s chair, not the project plan.
Contributed by: Ben Neo, CX EMEA Lead, Zoom
For more great insights and advice from our panel of experts, read these articles next:
Megan Jones, Editor, Call Centre Helper
Megan interviews inspiring people across the contact centre industry to capture these amazing voices and turn them into best practice articles – covering all elements of what it takes to deliver truly outstanding employee and customer experiences.
Connect with Megan on LinkedIn
Read more by Megan JonesReviewed by: Jo Robinson
The Latest Customer Contact Research, Reports and White Papers straight to your inbox!
Get the latest insights from Call Centre Helper by signing up for our email content.
Close
13th October 2026
14th October 2026
