AI agents are entering customer service

AI agents are becoming more than simple chatbots.

They can respond to customer questions, recommend products, process refunds, support bookings, send messages and help customers complete tasks.

The CMA has published guidance for businesses using AI agents to engage with customers. It says AI agents can support customer queries, refunds, product recommendations and marketing campaigns, but they must be used responsibly and in line with consumer law.

This matters because customer-facing AI is moving quickly from experiment to everyday business tool.

For SMEs, the opportunity is clear.

AI agents can reduce repeated enquiries, speed up responses and help staff focus on more complex customer issues.

But the risk is also clear.

If the AI agent gives the wrong answer, the customer may suffer and the business may be responsible.

The risk is poor customer outcomes

The CMA says businesses are responsible for what an AI agent does in the same way they are responsible for what an employee does, even where a third party provides the AI agent.

That is an important point.

A business cannot simply say, “The AI made the mistake.”

If the AI agent gives misleading information, handles refunds wrongly or makes it harder for a customer to exercise their rights, the business may still be accountable.

Common risks include:

• Wrong information about prices
• Incorrect refund decisions
• Misleading product recommendations
• Failure to explain important limitations
• Unclear labelling when customers are dealing with AI
• Poor escalation to a human adviser
• AI outputs that sound confident but are inaccurate
• No monitoring of customer complaints or feedback
• No quick correction when problems appear

The CMA also says businesses should consider telling customers when they are dealing with an AI agent if that fact could affect their decision, and should not overstate what AI can or cannot do.

That is why customer-facing AI needs governance.

What SMEs should do before deploying AI agents

SMEs should not wait until a customer complains before reviewing AI agent controls.

Before using AI in customer service, refund handling, marketing or recommendations, businesses should ask:

• What is the AI agent allowed to do?
• What is it not allowed to do?
• What customer rights could be affected?
• What information must be accurate?
• When should the customer be told AI is being used?
• When must the AI escalate to a person?
• Who reviews AI conversations?
• How are complaints and mistakes tracked?
• How quickly can prompts or workflows be fixed?
• Who owns legal and operational responsibility?

The CMA says testing is an important part of training AI agents, and businesses should monitor whether agents are delivering the right results, behaving as intended and complying with consumer law.

CAIT Group Ltd helps organisations create this structure.

CAIT supports customer-facing AI readiness, workflow automation reviews, AI governance, human escalation design, staff training and AI risk readiness.

The goal is not to stop AI agents being used.

The goal is to make sure customer-facing AI is useful, fair, monitored and controlled.

Practical impact by organisation type

Individuals: Customers should receive clear, accurate information and have access to human support where needed.

Small businesses: Simple controls can prevent AI agents from giving wrong refund, pricing or product information.

Medium businesses: Monitoring and escalation rules help customer service teams use AI consistently across departments.

Large businesses: Governance supports auditability, legal compliance, customer trust and operational control.

Multinationals: Customer-facing AI needs consistent rules across markets, languages, product lines and consumer protection requirements.

Public sector organisations: AI agents used in public-facing services must be transparent, reviewed and supported by human escalation where users may be affected.

Thinking about using AI agents for customer support, refunds or enquiries?

We can help you define safe use cases, test AI workflows, create escalation routes and put practical monitoring around customer-facing AI.