AI has taken over the customer experience. According to CallMiner’s 2026 CX Landscape report, 99% of CX teams have automation built into some aspect of their customer interactions. Yet just 24% of them consider the experience to be “very positive.”

It creates an interesting challenge for CX leaders. Teams need to ask where AI can best support their CX interactions and where humans remain essential to delivering the experience customers expect. Leaders also need a clear way to measure whether those choices are working: Is AI resolving issues, reducing customer effort, and improving satisfaction? Or is it creating friction that leaves customers frustrated and employees handling more complex problems?

Whitepaper

Empathy & Automation: Three Ways to Make Automated Customer Experiences More Human

Discover how customer intelligence and conversation intelligence can help organizations build more human customer experiences with automation. Explore three practical ways to identify the right interactions to automate, design around real customer behavior and measure automation based on customer outcomes, not efficiency alone.

The solution isn’t one or the other but a combination of the two. It’s a deliberate balance, shaped by what customers actually say and adjusted as both the technology and customer trust evolve. Here’s how to find it.

What customers actually think about AI support

When consumers interact with AI-powered support features, there’s often still friction surrounding it. According to the 2026 Delight AI Index, which surveyed 1,001 U.S.-based consumers, 71% of respondents had interacted with some form of AI customer service. Only 57% of them had positive interactions.

Other research has also suggested consumers’ reluctance to embrace AI. In a study by Gartner that surveyed 5,728 customers, 64% said they would like companies to avoid using AI for customer service altogether, while 53% said they’d take their business to a competitor if a company continued to utilize artificial intelligence. The primary concern among respondents was AI limiting their ability to speak to a human representative.

Brands have much riding on improving these perceptions. When AI gets something wrong, 83% of customers blame the brand, not the technology, according to the Delight AI Index. Trust in AI also takes time: just 16% say they would allow AI to make decisions without their approval. Nearly half (47%) say they would like to confirm each of AI’s actions. Yet despite the challenges with trust, more than half of customers (54%) expect their trust in AI to increase over the next 12 months.

CX leaders are aware of this conflict, too. In CallMiner’s research, 43% of organizations say most customers prefer to talk with a human, and 40% say their AI has difficulty handling complex, emotional or risky situations. Both have declined year-over-year, suggesting that trust in automation is built one positive interaction at a time.

Where AI earns its place

AI shines where speed, scale, and consistency are valued over nuance. That’s typically when:

• Requests are repetitive and predictable. Order status updates, password resets, store hours, and balance inquiries account for the majority of contacts at most organizations. Automating these tasks enables your customers to get instant replies while freeing up agents to do the work that requires a human.
• It’s after-hours or you’re experiencing high volume. Your AI solution can work overtime, every day. Whether you’re dealing with a holiday influx or a spike caused by an outage, there’s no overtime pay.
• The transactions are structured. Routine doesn’t necessarily mean simple. A customer return may involve multiple steps, but as long as there’s a defined process, AI can typically handle it. However, if a customer is irritated about a charge and wants to dispute it, that’s where you’ll need humans.
• It’s triage and context gathering. AI can determine a customer’s intent, confirm identity and gather relevant information so that when an agent does need to step in, they’re already armed with context.
• You want to reach out first. Automated alerts about outages, renewal messages and delivery notifications can prevent customer inquiries before they happen.

Beyond where it’s applied, AI’s capabilities are growing. Gartner forecasts that by 2029, agentic AI will resolve 80% of common customer service inquiries without human intervention and reduce business operational costs by 30%. As AI takes on more of the low-hanging fruit, accurate resolution of non-routine issues will become increasingly critical.

Where humans remain essential

There are still times when humans are best suited to respond. Humans should be in the driver’s seat when:

• The situation is highly emotional. Fraud reporting, bereavement, or financial distress aren’t times to feel like a number.
• Issues are complex or involve numerous obstacles. Several roadblocks, multi-step troubleshooting, or scenarios without a known solution call for human agents who can improvise, not recite.
• The customer’s desired outcome is unclear. If it’s not clear what a customer is asking for, a human can ask the right probing questions.
• When there’s a lot riding on the outcome. Issues involving finances, health, or legal/regulatory risk are costly to get wrong. The Delight AI Index reports consumers embrace AI for day-to-day requests but expect a human when these specific scenarios arise.
• When sales are on the line. Account retention, win-backs, upsells, and high-value accounts all benefit from human judgment.

CX leaders understand this intuitively. According to CallMiner’s 2026 CX Landscape Report, 95% of companies believe human agents are more valuable than AI for at least one type of interaction. In particular, this is true for interactions that include complex problem-solving, high business impact, customer vulnerability or empathy.

The broader market is also correcting course. By 2027, Gartner predicts that 50% of organizations that intended to reduce their customer service staff size significantly because of AI will abandon those plans. It also expects that no Fortune 500 company will have eliminated human customer service entirely by 2028.

AI or human agent? 5 questions to decide

Who should own an interaction? Start by answering five questions:

1. Is it predictable?
2. Is there emotional baggage?
3. Are there high stakes?
4. Is it relationship-driven?
5. Is it reversible?

Most importantly, make sure your decisions are data driven. Don’t fall into assumptions. For example, one CallMiner customer assumed calls about bills would be ideal to automate. Upon analyzing the conversations, they found that a lot of customers were actually calling in to see if they could get assistance enabling AutoPay. With that insight, they were able to automate AutoPay for customers who ask for it, eliminating the need for that call entirely. The problem is that just 22% of organizations are leveraging CX insights to inform what they should automate.

Designing the AI-to-human handoff

The handoff is the moment that determines whether a hybrid experience succeeds or fails. Pay attention to these details.

Show the human path. The inability to reach a human is the number one customer complaint about AI. Never conceal the off-ramp.

Define escalation triggers. Escalate when a customer circles back without making progress, sounds upset or pressured, requests a human, is part of a high-value segment, mentions something compliance-sensitive or hasn’t received resolution in the predetermined time. Advanced sentiment analysis and vulnerability detection help tighten these triggers.

Forward the context. The agent should be able to see the entire conversation along with the customer’s inquiry and reason for the escalation. The customer shouldn’t have to repeat themselves.

Be transparent. If customers are talking to AI, tell them. Passing off chatbot technology as a human being undermines the trust you’re trying to create with the handoff.

Escalation isn’t a failure. CallMiner’s research found that organizations reporting very positive automated experiences escalate to human agents more readily than those with mixed or negative results: 76% versus 68%. The best automation is designed to recognize its own limits.

AI that augments human agents

Balancing automation with your agents isn’t just about directing customers to the right responder. AI also empowers your agents to excel at the conversations that are the most challenging. CallMiner’s research found 85% of organizations are using AI technology for at least one human-agent use case. The most common use cases are real-time assistance, productivity, and training and coaching.

This can look like pulling up knowledge articles in the middle of a conversation, transcribing and summarizing customer interactions so agents don’t have to spend time on after-call activities, and highlighting when a customer’s sentiment is changing. If AI can take care of high-volume, low-complexity tasks and assist agents with complex interactions, agents can focus less on busywork and more on the conversations that require a human touch.

Continuously tune your AI-human balance

It’s not enough to set the right balance once. You need to tune it like an instrument, and that requires measuring the right data.

Begin by monitoring AI and human performance metrics separately. An integrated CSAT score can mask a frustrating automated journey. Consider containment rates and CSAT together as well. Low escalation with poor CSAT typically indicates contained customers, not satisfied ones.

Don't forget customer effort score, first contact resolution, and repeat contact rate to ensure your customers' issues are truly being solved. Lastly, track sentiment post-IVR to gauge satisfaction with automated interactions. A decrease in sentiment highlights friction points to improve.

This is where most organizations fall short. CallMiner found that 68% say they don’t use their CX data to its full advantage, and only 30% use insights to improve automated interactions. Nearly all (94%) struggle to align CX data and feedback across teams.

Getting it right pays off. Of organizations that report their automated experience is very positive, 49% report they continually optimize automation with CX insights, compared to only 8% of organizations that report their automated experience is mixed or negative. Measuring and gaining insight from 100% of your interactions (both AI-handled and human-handled) can tell you where automation is performing well, where it can be improved, and where a human delivers the better outcome.

Finding the ideal balance

“AI vs. human” isn’t the right mindset. Leading organizations use AI for tasks that AI excels at, humans where empathy and judgment are needed, and are customer intelligence driven to know where to draw that line. That line will continue to shift as AI technology advances and customers become more trusting of AI, so this is an ongoing process.

CallMiner Eureka allows you to find the perfect balance between human and AI with intelligence from every conversation. By analyzing 100% of your omnichannel interactions, both AI and human-handled, CallMiner demonstrates where to automate and where a human touch is required. CallMiner also provides agents with real-time guidance and coaching for those moments that require a human.

Frequently asked questions

What does a hybrid customer service model look like?

Humans and AI working side-by-side in the same service organization. AI manages structured, low-context conversations and collects context before routing. Human agents own the most complex cases and anything that might elicit emotion or high stakes for the customer. Conversations should seamlessly hand off between AI and human agents, so the customer never has to start over.

Will AI replace human customer service agents?

Not entirely. Research firm Gartner predicts that by 2028, no large enterprise (Fortune 500) will have totally eliminated human customer service. Instead, agents will spend more time managing high-complexity, high-value conversations while AI absorbs high-volume, structured tasks. Agents will also be supported by AI throughout the conversation, helping them handle more traffic.

Do customers prefer AI or human support?

It depends on the task. Consumers are generally comfortable using AI for quick, everyday requests like order tracking. They prefer a person when money, health, urgency, or strong emotions are involved. Trust in AI is growing, but it builds gradually.

Which customer service tasks should be automated first?

Every support organization should look at routing high-volume, structured, low-risk questions to AI first (think password resets, order status, and FAQs). Look at your conversation data to see exactly why customers are contacting you. You may be surprised by what actually drives your customers to reach out.

When should AI hand a conversation off to a human?

Humans should take over when customers have to repeat themselves without progress, sound angry or upset, explicitly ask for a manager or agent, or mention sensitive financial, health, or legal concerns. If the account is high-value or the conversation goes past a time threshold without resolution, escalate as well. Make sure any automated handoff includes the entire conversation for context.