Thought Leaders
Best Supporting Agent: Why Agentic AI in the Contact Centre Still Needs a Human Lead

Key PointsAI assistance helped agents resolve about 15% more customer issues per hour in a study…
View more
- AI assistance helped agents resolve about 15% more customer issues per hour in a study of over 5,000 workers.
- Contact centres must define which tasks AI can handle alone and require human approval for high-value exceptions.
- Leaders should track repeat contacts and correction rates rather than celebrating lower handling times alone.
Agentic AI is moving fast from answering customer questions to taking action on their behalf. It can retrieve account records and move a request forward without waiting for an agent’s input. For contact centres under pressure to handle more conversations with limited resources, that is an appealing prospect.
But giving software permission to act also raises the question of what happens when a customer needs an exception, or when an automated decision is technically correct but plainly unfair. These are everyday realities of customer service, and also moments when trust is won or lost.
The opportunity is substantial, provided contact centres are deliberate about which decisions AI can make and when human lead is required. A strong option for contact centres is to use automation for routine tasks while giving agents the authority to resolve complex customer issues.
How Agentic AI Is Changing Customer Service Operations
Traditional contact-centre automation could only do so much. Customers were often stuck navigating phone menus or choosing from a limited set of options, and anything more complicated usually meant waiting to speak with an agent. Agentic systems can go further, and given the right permissions, they can follow a customer’s request across multiple systems and carry out approved steps. A request to change an appointment, for example, might involve checking eligibility and updating the booking.
That changes the role of the human agent. Instead of spending the first few minutes hunting for information, an agent can enter the conversation with a clear picture of what has already happened. A well-designed assistant can surface and summarize previous interactions before the agent has to ask another question.
AI can make a real difference in customer service. A study on customer support agents involving more than 5,000 workers found that agents using AI assistance resolved about 15% more customer issues per hour. The biggest improvements came from less experienced agents, who benefited from having helpful suggestions available while handling customer requests.
These findings are encouraging, but there’s an important detail to be considered. The AI was helping agents do their jobs, with people still making the decisions. That leaves an important question for contact centres exploring agentic AI: how much responsibility should these systems have when working directly with customers?
Speed is certainly part of good service, but customers also expect accurate answers and fair treatment. An AI agent might resolve a billing inquiry in seconds, but if it misunderstands the customer’s situation or applies the wrong policy, that quick resolution could turn into another complaint. Contact centres need to look beyond how quickly an issue is closed and consider whether the customer actually received the help they needed.
Where Does Human Judgment Still Matter?
Even with access to customer records, AI doesn’t always have the full picture. Take a billing dispute, for example. The AI checks the account and confirms that a charge matches the contract. However, the customer was previously promised a credit during a service outage, and that information was never added to the billing system. The AI may give an answer based on the available records, but an experienced agent can recognize that something is missing and investigate further.
In 2026, Alibaba’s customer service operations showed that AI shortened conversations but received lower customer ratings in AI-eligible interactions. The research also showed that the effectiveness of human intervention depended on when agents stepped in and what had gone wrong. These findings suggest that knowing when to involve a person is just as important as knowing what AI can handle.
For contact centres, the handoff from an AI agent to a human agent deserves as much attention as the technology itself. Customers shouldn’t have to repeat their concerns every time they’re transferred. Agents need access to previous conversations and the actions AI has already taken so they can pick up where the system left off. A smooth transition can make all the difference, especially when someone is already frustrated and simply wants their problem resolved.
Setting Boundaries for AI in Customer Service
Contact-centre leaders need to think carefully about how much control they give AI. Preparing a refund recommendation, for example, carries less risk than issuing a large credit without approval. Build these limits into the system from the start, with clear rules for when an agent needs to step in.
Contact centres can establish these boundaries by focusing on a few practical measures:
- Define approved tasks: Identify which requests AI can handle independently, such as checking order statuses or updating appointments.
- Set approval limits: Require agent authorization for high-value refunds, account changes or exceptions to company policies.
- Establish escalation rules: Automatically involve an agent when customer information conflicts, a request falls outside approved procedures, or the system cannot provide a reliable answer.
- Control data access: Give AI access only to the customer information necessary to complete an approved task.
- Review performance regularly: Monitor incorrect responses and customer complaints to identify where additional safeguards are needed.
As AI takes on more responsibilities, these boundaries should be reviewed and adjusted based on how the technology performs in real customer interactions. This allows contact centres to expand automation while maintaining accountability for the decisions being made.
What Contact Centres Should Actually Measure
A contact centre can celebrate a lower average handling time while customers lose confidence. If someone calls back the next morning because an AI agent promised a change that never happened, the first conversation was not a success. The reporting should make that visible.
Alongside speed and containment, teams should examine repeat contacts, reopened cases, correction rates, customer complaints and how often a human must reverse an automated action. They should review escalations for missing context and track whether the AI’s recommendations were useful to the people receiving them. These measures are important when automated actions affect a customer’s money or access to a service.
Leaders should also ask agents what the system gets wrong. Frontline agents hear the awkward pauses and recognize when a customer has lost patience. Their feedback is a source of operational intelligence. Give them a quick way to flag errors, and show them what changed as a result. If every correction disappears into a generic feedback form, the team will stop bothering.
However, using AI to understand agent experiences also raises questions about how closely workers should be monitored. Some tools claim they can detect emotions through a person’s voice. For example, the EU AI Act generally prohibits using AI to identify agents’ emotions in the workplace, with limited exceptions. The European Commission has even used call centres as an example. Businesses should check the laws in their region before introducing these tools.
Keeping People Accountable as AI Takes On More Work
There is a tempting vision of a contact centre where almost every conversation ends without a person. It makes for a neat presentation slide. Actual customers tend to be less tidy. They change their minds, misunderstand instructions, ask for exceptions and bring personal circumstances that a system may not be able to judge fairly.
As AI takes on more responsibilities, contact-centre leaders need to make sure their teams can recognize mistakes and correct them quickly. Customers shouldn’t have to repeat their problems or deal with the consequences of an automated decision that went wrong.
AI’s success in customer service depends on how well it supports agents in delivering the service customers expect.












