Driving customer support efficiency with AI

Don't let your AI transition become a customer exit

Rapid developments in Artificial Intelligence (AI) are transforming every industry, and many would argue that customer support has been at the head of the line. There is a version of the transition to AI-enabled customer support that looks like it’s succeeding across many dashboards, and yet in the background, customer relationships are being slowly and steadily destroyed. Somewhere behind the numbers, there’s often a growing category of customers who needed something the AI chatbot couldn't give them, and when they didn't get it, they didn't come back.

The problematic assumption underlying this situation is that AI will handle more routine and predictable customer interactions on its own and automatically, and that the benefits will follow like water off a hill.

Falling into this assumption is possibly the most expensive mistake operations leaders can make during the transition to AI adoption that’s happening at scale in every sector and every enterprise.

The reality is that AI can handle standard and predictable queries — the kind that could be described as “frequently asked questions” — but only when you design, build, and operate it to do so. It can add extraordinary value, but only when you can fully integrate it into live operations, and manage it with the same rigour that’s applied to your existing teams. 

AI can create a more capable team, with experienced people empowered to focus entirely on the complex, high-stakes interactions that need them most, and where the commercial consequences of getting it wrong are greatest.

What happens to the work AI doesn't handle

Gartner research published in 2025 predicted that agentic AI will autonomously resolve 80% of common customer service issues without human intervention by 2029. 

Lurking in the details, though, are two stark facts: first, that this is a projection, and second, that the projection relates only to “common” customer service queries.

Behind the attention-grabbing headline, therefore, is a much more complex reality. To start with, the 20% — at least — of common issues that AI still won’t be able to solve years from now, plus the 100% of more complex interactions, all requiring sound judgment, empathy, authority and autonomy.

These are not being automated away anytime soon. Instead,they’re becoming more concentrated, and more consequential. Indeed, a strong argument can be made that these interactions will have the biggest impact of any factor on brand loyalty and business success over the next 10 years.

When AI successfully handles what’s standard and predictable— order tracking, for example, or standard account queries and first-line diagnostics — the residual items are, by definition, more complex issues, often with several moving parts, and often leading to emotionally charged situations that a chatbot loop will likely make worse.

Another confounding factor is that the increased availability of AI-powered chatbots has in many cases led to two strange situations at once: a significant growth in the number of questions asked, and many questions that are not satisfactorily answered, causing potential damage along the way. After all, when there’s a chat interface a click away to answer anything you ask it, why take the time to peruse an FAQ section in the first place? On top of this, those new chat interactions can lead to escalations that might never have happened previously, and each of those interactions has implications for brand perception and trust.

All these factors together create a filtered queue of complex, high-stakes interactions that AI cannot deal with on its own. What’s more, this is not just a temporary problem to be dealt with on the way to full automation.

In fact, as issues become both more numerous and more thorny, dealing with the complex will be an imperative to consider alongside any automation initiatives.

The case for building your own team — and why it usually breaks

The first instinct for many operations leaders who recognise this challenge is to try to solve it in-house. That will involve building a specialist team, hiring for emotional intelligence and commercial judgment, where they believe their quality standards and company culture can be safely influenced and controlled.

This logic is understandable, but the execution is orders of magnitude harder than it looks, and often breaks entirely, for three primary reasons, each of which can compound on top of the other in a doom loop before the project is finally canned.

The talent problem

The people who handle the complexity of the filtered queue well will be those with sound judgment, genuine empathy, the ability to de-escalate under real pressure, and the commercial awareness to know when a resolution that costs something is worth it. 

Manpower Group's most recent global talent survey in 2025 found that more than 80% of employers in Ireland report difficulty finding the skilled talent they need, a figure that has exploded from just 34% in 2019 and represents an all-time high.

What’s worse, building a team with this profile is a sustained recruitment burden that pulls capacity away from everything else. The market for this talent is competitive in every geography, and the cost of getting it wrong comes in the form of high churn, inconsistent quality, and continuous retraining, elements that are rarely captured in the original business case.

The structural rigidity problem

Internal teams are built for predictability: stable headcount plans, agreed shift patterns, management layers designed around relatively known volume. By definition, however, the filtered queue is never truly stable.

As AI adoption accelerates, the day-to-day operational reality shifts with it: interactions that are complex today could be largely automated tomorrow, while variations of now simple issues could become more complex as products evolve and customer expectations rise. All of which requires a degree of operational flexibility that internal teams will struggle to deliver.‍

The management overhead problem

Every hour your VP of Operations spends managing the headcount needed to handle customer support requirements — performance issues, workforce management, scheduling, team attrition, training cycles, and technology implementation — is an hour of opportunity cost that could better be spent on improving core product and growth. 

This overhead scales with the team, creating a cost structure that grows in tandem with volume of customer interactions rather than benefiting from the efficiency of a specialist partner whose entire operational infrastructure is already designed, built, and running successfully for companies just like yours.

Build vs. Buy: Where to draw the line

The build-versus-buy decision in CX is often not a simple one. Cost is always a factor, of course, but the right framework is not primarily cost-related. It’s about figuring out where your organisation's competitive advantage actually lies, and being able to invest in maximising that most effectively.

Some things should stay in-house.

Core product strategy and brand development — particularly in the early stages of a business, when founders and senior leaders need to be close to customer conversations to understand what they're building — belong to the core team. 

But the complex, emotionally demanding, judgment-intensive work that arrives in the customer support queue after AI has handled everything it can (and half-handled the rest) is a different matter. This facet of the business requires specialist recruitment, often in languages and time zones that can become a serious administrative headache for the core business, plus operational infrastructure and deep expertise in AI deployment that is just not central to most companies’ true focus. 

Finding a partner whose entire operation is designed precisely around this kind of work offers a structural advantage that an internal build will always struggle to match at comparable cost.

The Otonomee view

The companies we work with understand something that is still underappreciated in the industry: AI does not handle routine customer interactions on its own. It can do it, once it’s designed, built, implemented, and monitored to do so. Automation can add extraordinary value, but it doesn’t happen out of the box. It happens with intentional design, expert model training, and the operational and technical savvy to integrate them into live operations. 

Otonomee designs, builds, owns, and drives AI automation for our clients. We ensure models handle Level 1 interactions end-to-end, without human involvement, but this is just the beginning of our engagement. When AI is working properly on what it should handle, our experienced teams kick into gear. 

We are proud to select and train our teams in “power skills” such as empathy and sound commercial judgment, and then empower them with the authority to make real decisions. 

This leads to better outcomes and fewer escalations from the complex interactions that need their unique skills and awareness, all creating the conditions that can lead to a customer-for-life.

That is the type of operation Otonomee has been built from the ground up to deliver. If your AI transition is creating a queue of more complex and more demanding customer support interactions your team is not fully equipped to handle, talk to us.