Can AI Agents Really Handle 50–60% of Customer Service?

The claim that AI agents can now handle 50–60% of customer service interactions is no longer a far-off projection. This figure has started to appear repeatedly in industry research throughout 2025–2026. The question is no longer “is it possible”, but “which kinds of interactions” can truly be taken over by AI, and where the limit is.

Data That Supports This Claim

Several independent studies reinforce this figure from different angles:

• Cisco research on nearly 8,000 decision makers in 30 countries projects that more than 56% of customer interactions will be managed by AI in 2026, rising to 68% by 2028.
• For the voice channel specifically, AI voice agents can currently resolve 40–60% of incoming calls for structured request types, such as balance checks, appointment scheduling, and order tracking.
• AI-based chatbots that can take action (not just answer) resolve 60–80% of customer questions without human involvement, far above older rule-based chatbots, which only reach around 20–35%.
• Salesforce reported that its Agentforce recorded resolution of up to 84% of more than 380,000 support interactions, while several research aggregators put the industry average at around 69% resolution without human help.

The 50–60% range is a realistic and conservative figure compared to more aggressive claims. It suits the average company that has already implemented AI, not just front-runners like Salesforce or Klarna that have reached 70–80%.

Types of Interactions Easiest for AI to Take Over

AI agents excel in categories of questions that are high in volume but low in complexity:

The pattern is clear: the more routine and structured the question, the higher AI’s ability to resolve it on its own. Once it enters emotional territory, disputes, or cases that need long context, AI’s success rate drops sharply.

Why There Are Still Limits

Even though AI resolution rates keep rising, research also shows the other side:

• About 60% of customers still prefer to talk to a human for complex or emotionally charged problems.
• 77% of customers say they are frustrated when they cannot reach a human agent at all.
• Public trust in AI to handle billing disputes without human oversight is still low, at only around 29%.

This confirms that the most effective model is not “AI fully replacing humans”, but rather AI filtering and resolving routine cases, with humans focusing on high-value cases. The combination of the two has even been shown to boost customer satisfaction (CSAT) scores by 15–20% compared with AI alone.

Implications for Businesses in Indonesia

For companies that run call centers or omnichannel customer service, this trend has several practical consequences:

1. Routine work can be automated first. Order status checks, rescheduling, FAQs, and data verification are the most mature candidates for AI agents.
2. Escalation must be smooth. A good system makes sure AI knows when to hand the conversation over to a human agent, rather than forcing customers to repeat their story from the beginning.
3. Integration with back-end systems determines quality. Companies that connect AI with CRM, order systems, and billing record resolution rates 2–3 times higher than AI that relies only on a static knowledge base.
4. Cost per interaction drops significantly, from around USD 3–6 equivalent for human handling to around USD 0.25–0.50 equivalent for AI-resolved interactions.

Conclusion

The 50–60% figure is not empty hype. It reflects the average state of the industry when AI agents are mature enough to handle the majority of routine interactions but still need humans for complex and sensitive cases. The most realistic strategy for businesses is not to choose between AI or humans, but to design workflows in which the two complement each other, with AI handling volume and humans protecting the quality of the customer relationship at the most decisive points.