“The simple truth is that companies can achieve the largest boosts in performance when humans and machines work together as allies… in order to take advantage of each other’s complementary strengths.”
— Paul R. Daugherty, Human + Machine
Think of AI Agents as Rebalancing, Not Replacing
It’s 2025, and the conversation around customer service automation has matured. The hype around replacing support teams with chatbots has met the friction of reality. Now, leading companies are asking smarter questions: not “How can I automate more?” but rather “What’s the optimal blend of AI agents and human agents?”
At Loris, we believe the future of support is not binary. It’s collaborative. And that belief is rooted in hard data, operational lessons from the front lines of CX, and a deep understanding of where Natural Language Processing (NLP) excels—and where it still struggles.
To that end, we believe the best way to use AI Agents and human agents is to understand what each is best at. Below you’ll find the categories where AI accels, where it falls short, and a methodology to categorize your own customer service issues so you can build a program that serves customers more effectively, no matter who (or what) they are interacting with. Let’s explore how to build that right mix.
The Case for AI Agents: Speed, Scale, and Simplicity
There’s no doubt AI agents bring powerful benefits. Well-designed AI systems can handle repetitive inquiries around the clock, reduce time-to-resolution for common requests like order tracking, and surface relevant knowledge at scale. These benefits offer an obvious return on investment—especially for high-volume, low-complexity channels such as ecommerce or travel.
However, the ability to deflect simple inquiries shouldn’t be confused with the ability to deliver true customer satisfaction. According to Tidio, nearly half of users are satisfied with chatbot support—when their problems are simple. But that statistic flips the moment complexity or emotion enters the picture.
Where AI Agents Fall Short: Nuance, Empathy, and Judgment
Despite the rise of generative models, language modeling is not the same as understanding. As outlined in “It’s the Golden Age of NLP, So Why Can’t Chatbots Solve More Problems?”, AI agents can predict likely responses, but they often fail to grasp meaning, context, or emotional nuance. That leads to friction. Chatbots can misread tone, provide misleading answers, or fail to escalate fast enough—ultimately frustrating customers.
The consequences aren’t theoretical. Consider FedEx’s infamous AI phone bot that reportedly refused to connect customers with a live agent, even hanging up on them. That kind of automation backlash can cause long-term brand damage.
But diagnosing issues with your AI Agent doesn’t need to get to this critical level. There is data within customer interactions that can give you early indicators of issues. If your AI Agent is having a high number of exchanges with customers, especially for issues you consider straightforward, could mean there’s some disconnect or lack of understanding, either by the bot or the customer. Likewise, conversation intelligence tools can highlight changes in customer sentiment. A decline in customer sentiment is an immediate sign that something is going wrong with the AI Agent-customer conversation, and may even indicate that these issues are best handled by a human agent. Lastly, more advanced interaction analysis can provide markers like confusion, frustration, and other factors that show customers are not getting what they need from AI Agents.
Having this data gives you an immediate look at the quality of your AI Agents and help you understand whether there are gaps in training, or whether the AI Agent just isn’t ready for some issue types. But don’t think of the inability of AI Agents to handle every issue as a failure. This is why you have human agents, and where their unique abilities can make all the difference.
Human Agents: The Last Mile (And the Most Important One)
When customers are angry, confused, or in distress, nothing replaces human empathy. But more than that, complex cases—like multi-step resolutions, cancellations, or sensitive billing issues—demand judgment and creativity that no AI model can yet replicate.
However, the problem isn’t just about ensuring human agents are available. It’s about ensuring they are prepared. Too often, AI systems dump already aggravated customers into the laps of support agents without providing context or insight. This sets both the customer and the agent up to fail.
This is where customer intelligence can improve the customer and agent experience. Using a Customer Insight platform doesn’t just analyze conversations—it extracts meaning. This helps companies understand how customers felt, what they needed, how hard agents worked, and where coaching is required. That means your teams aren’t flying blind. They’re guided by insight, not assumption.
Avoiding the Automation Trap: Short-Term Wins, Long-Term Losses
As detailed in “Is Your AI Strategy a Short-Term Success, But a Long-Term Loser?”, many brands embraced automation as a blunt cost-cutting tool. But cost-cutting has a floor, while value creation has no ceiling. Reducing support headcount might help the quarterly P&L, but it doesn’t drive loyalty or insight.
Klarna is one brand that experienced this firsthand. After betting heavily on automation, they’ve rebalanced toward models that put humans back in the loop—an acknowledgment that real-time problem-solving still requires a real person.
Learning from Klarna: Exposing the Myths & Facts of AI Agents
Join industry experts as they analyze Klarna’s strategic shift and what this means for broader AI agent adoption.
A Better Model: Insight-Driven AI, Human-Led Support
The ideal customer support model doesn’t start with automation. It starts with understanding. That’s why Loris recommends brands start with Customer Insight Analytics and Quality Intelligence before they implement an AI agent.
This approach helps identifies which types of interactions consistently lead to poor outcomes and why. That allows CX leaders to intervene proactively, coach precisely, and track whether quality is improving—not just whether tickets are closing. It also helps you determine which issues are more likely to be handled well by AI agents and which are still too complex.
In addition, having an AI approach to QA can surface friction patterns, recurring issues, and agent behaviors that traditional QA methods miss entirely. And because AI-based QA is faster, you get strategic insight without the lag.
Real-World Ratios: What’s the Right Mix?
The optimal balance between AI agents and human agents isn’t fixed—it depends on your industry, channel mix, and customer expectations. A fintech company may require more human oversight than a food delivery service. Email might allow for asynchronous escalation, while live chat demands immediate triage.
That said, many brands find success using AI for 40% to 60% of Tier 1 inquiries, while maintaining human control for high-complexity or high-emotion cases. The important part isn’t the ratio—it’s the readiness. Are your agents informed when they take over? Do you know where AI is working—and where it’s quietly failing?
In order to have confidence around which conversations you route to AI Agents and which you send to your human agents, you should start with what you know about the conversation. The grid above presents these elements within different categories on a task nature and complexity axes. Grouping your top contact issues into a grid like this can be helpful to understand where are the easiest places to start with AI Agents, where should you go next, and where do you need highly skilled agents for the near future.
- The “Simple Transactional” category is likely where you already have a chatbot, FAQ, or AI Agent working. These are the simplest tasks that most customers should be able to self serve. For those that do need assistance, an AI Agent should be able to guide them and even complete the transaction on their behalf.
- The “Simple Consultative” may be the next logical area for AI Agents. Generative AI has made this type of interaction possible, since the information is likely available across your website and other resources. The bigger question here is whether you trust the AI with this type of interaction or whether you see this as a potential revenue opportunity where a human agent would work better.
- The “Complex Transactional” category is a lower-risk but harder area for AI Agents. As they get better at solving more nuanced issues, this is an area that you would want to automate, since the value is relatively low. But, there are decisions and logic that would have to be built in other for the AI Agent to successfully complete the transaction. At a minimum, the AI Agent could complete some of the initial steps and pull in the human agent when a decision is needed.
- The “Complex Consultative” may be the area that AI will struggle the most – at least at handling independently. But these are also the areas where human empathy, problem solving, and creativity are assets that are your best options. AI may be useful in things like routing, intent tagging, and summarization to assist the human agent in solving these issues as well as highlighting the frequency and severity of these issues in top-level reporting.
Beyond Coverage: Measuring the Right Outcomes
Automation alone should not be your North Star. It’s a means to an end. And automation should be applied to cases where human agent and customer interactions are either low value or where it’s the fastest way for the customer to achieve their goal. But you need to analyze all your conversations to understand where to apply automation and if automation is positively impacting your business and your customers.
To do that, measure how AI agents are performing just as you measure your human customer service agents. Are escalations smoother? Is customer churn decreasing? How are conversations between customers and AI agents impacting the customer’s experience?
These are the outcomes that comprehensive AI analytics helps you track.
The Takeaway: Automate with Intention, Operate with Insight
The future of support isn’t about choosing between humans and machines. It’s about creating intelligent systems where each does what it does best. That begins with insight, not automation. And it ends with agents who are empowered—not overwhelmed.
At Loris, we help you strike that balance. Schedule time with our team and discover how conversation intelligence drives customer and AI Agent success.
